Part 3 ended with a question. Margin calls, leveraged ETFs, and options dealers can each independently cause selling — for entirely different reasons, on entirely different schedules. What happens when those schedules line up?

That’s where this piece picks up.

Three systems. Three clocks.

Three different things can force selling in a market, and they don’t run on the same clock. Worth laying them out side by side before asking what happens when they go off at the same time.

Table 1.

These aren’t the same kind of leverage, and they don’t play by the same rules. They also aren’t equally visible. That last point turns out to matter a lot, and it’s worth pausing on before we put all three together.

The clock that doesn’t wind down on its own

Here’s a natural assumption about the options-dealer piece. As a stock keeps falling, the puts a dealer hedged against a while ago go “deep in the money” — meaning they’re now almost certain to be exercised. An option like that barely reacts to further price moves anymore, so the dealer doesn’t need to keep adjusting their hedge for it. Shouldn’t that mean the selling pressure from this source naturally fades as the crash goes on?

Not necessarily. New puts keep getting bought. When people are panicking, they don’t only buy insurance far below where the stock has already fallen to — a lot of that buying happens close to wherever the price is right now. So as the price keeps dropping, a fresh wave of “this option really matters to hedge” activity can form right behind the old one, even as the old batch fades out.

In other words: just because the old hedges have stopped mattering doesn’t mean new ones aren’t taking their place. Nothing about how options work guarantees this pressure fades just because the price has already fallen a lot.

That said, this isn’t automatic either. “People buy insurance” doesn’t mean they always buy it at-the-money — plenty of real insurance-buying happens well below the current price, exactly because that’s the kind of protection people actually want. And sometimes the flow runs backward: people cash in insurance they already own instead of buying more, which reduces this pressure rather than adding to it.

There’s also a real condition hiding in this whole idea: it depends on new buying actually showing up. If markets get so chaotic that spreads blow out and buyers simply freeze, that fresh wave of insurance-buying can stop arriving — and the price can blow straight through a big cluster of old bets without anything new forming behind it. Once that happens, that old cluster isn’t just failing to get replaced. Its own pull on the market drops to essentially zero, because those options are now so deep in the money they barely move differently from the stock itself.

So: this pressure doesn’t necessarily last forever. It lasts for as long as panic keeps generating fresh demand to replace what just stopped mattering.

The fastest clock may be the hardest one to see

There’s an important catch here.

A margin account getting a call, or a leveraged ETF’s rebalancing need, can both be reasonably estimated by an outsider — the rules are public.

Options dealer positioning is different. Nobody outside the options dealer’s own desk actually knows, in real time, how exposed they are.

Analysts can make an educated guess. They take yesterday’s known numbers, add today’s option prices, volatility, and how much the stock has moved, and build a running estimate of what dealers are probably holding and how they’re probably hedging it. But it’s still a guess, continuously updated — not a direct view into the dealer’s actual book.

By the time price action makes it look like dealers are adding to a selloff, that hedging may already be part of what caused the move you’re using to detect it. So a model of dealer positioning isn’t a security camera pointed at the dealer’s book. It’s a best guess, checked constantly against how the market is behaving.

When all three clocks go off at once

Now: what happens if all three of these clocks go off at once?

Nobody needs to coordinate anything for that to happen. The market itself can hand each system the same signal at the same time — and here’s the uncomfortable part: nobody watching from outside necessarily has to see all three happening together for it to be real.

Imagine a specific afternoon:

1:00 p.m. A cluster of semiconductor stocks breaks lower.

2:00 p.m. Investors who borrowed heavily on margin, start getting close to their limits and begin selling to reduce risk.

2:30 p.m. Volatility rises. Some brokers raise the equity cushion they require, right in the middle of the stress. The cliff moves.

3:00 p.m. Dealers appear to be positioned in a way that makes their hedging add to the selling — the same desks could easily have been positioned the other way, adding buying instead. Today, they aren’t.

3:30 p.m. As the close approaches, leveraged ETFs’ rebalancing needs become locked in. They need to sell to bring their exposure back in line with their stated multiplier — a “3x” fund has to stay at roughly three times the index’s daily move, no more, no less, and a down day means trimming exposure to hit that target.

3:45 p.m. Everyday buyers, sensing more forced selling might still be coming, stop stepping in aggressively at today’s price.

Four completely separate systems have now landed on the same answer — sell — for four completely unrelated reasons. None of them had to talk to each other. The price did the talking for them.

Here’s the loop: forced selling pushes the price down. That changes volatility. That changes what dealers need to hedge. Their hedging pushes the price down further. That lower price increases what leveraged ETFs need to rebalance, and pushes other borrowers closer to their own limits. And the cycle continues.

Nobody had to plan this. And nobody watching from outside necessarily saw all three pieces click into place until afterward, if at all.

That’s the synchronization problem.

One honest caveat: that afternoon is a made-up scenario, built to show how three unrelated triggers could land at the same time — not a record of a real day, and not a claim that this exact sequence has ever been documented happening. The mechanism behind it is real. The clock-by-clock story above is an illustration of how it could work, not a chart of something that did.

The second lever

The equity cushion isn’t the only thing a broker can move. The cost of the loan itself can move too, and for a related but separate reason.

A broker really has two separate dials. One is the maintenance requirement — how much of your own money you need to keep in the account. The other is the interest rate on the money you’ve borrowed. A broker can squeeze a leveraged account just by raising what it charges, without ever issuing a margin call — or squeeze it by raising the equity requirement instead. In a real stress event, both can move at once.

These two dials aren’t protected the same way, and that’s worth knowing. The equity requirement can be raised instantly, with no warning at all — that’s explicitly allowed.

The interest rate works a little differently: brokers are required by law (SEC Rule 10b-16) to tell you upfront how your rate can change. Most margin agreements use this to say your rate floats with a benchmark, like the Fed funds rate — meaning it can move without a fresh warning each time that benchmark moves, because you already agreed it would. Changing the actual terms of your agreement for some other reason is different — that requires at least 30 days’ written notice.

So a floating rate moving with its benchmark isn’t a “change” in that legal sense. It’s the agreement doing exactly what it said it would do.

There’s a smaller wrinkle too. Some brokers charge a lower rate on larger balances. So paying down your loan enough to drop into a smaller bracket can actually make what’s left of your loan cost more per dollar — a strange side effect where deleveraging makes your remaining debt more expensive to carry.

This is a slower-moving lever than the equity requirement. A cushion change can create an emergency instantly; a rate change just quietly eats into your cash over days and weeks of interest. But the shape is the same: one dial can move with zero warning, and the other can move without a fresh warning whenever it’s just doing what it already told you it might do. Either way, the account gets squeezed without the customer getting a new heads-up.

Then came the denominator

Up to this point, the focus has been on one side of the equation: how much forced selling could happen. But selling pressure only matters relative to how much buying is actually there to absorb it.

Call that the “effective liquidity” — not the total amount that trades in a normal day, but specifically how much can be sold right now, near today’s price before the price has to move to find a buyer.

Here’s a simple version of why that distinction matters. Suppose forced sellers need to get rid of $20 billion worth of stock. At today’s price, buyers are only willing to take $8 billion of it. That leaves $12 billion unsold. So the price drops. At the new, lower price, another $3 billion of buying shows up. Still not enough. The price drops again.

Eventually, someone says: fine, I’ll buy at this price.

The buyers didn’t vanish. They just backed away and waited for a better price.

That distinction matters a lot. A market can show enormous trading volume while the amount of liquidity sitting right near the current price has become very thin. And falling prices don’t just clear out the sellers who needed to sell — those new, lower prices become the new benchmark everyone else’s collateral gets measured against. Which can trigger the next round of forced selling.

Worth being upfront: the $20 billion and $8 billion figures above are made up, just to make the mechanics easy to follow — not real measurements. Nobody publishes real-time U.S. numbers for what an event like this would actually cost. The closest real example is from Korea, where forced ETF rebalancing genuinely ran to about 1.6% of Samsung’s and 2.1% of SK Hynix’s normal daily trading value — and at one point, trading in Samsung’s leveraged ETFs actually outpaced trading in Samsung’s actual stock, 3.7 trillion won to 1.25 trillion won (see Part 2). That’s a different market and a narrower slice of the picture than the full three-system U.S. scenario above — but it’s real and measured, and it’s a useful sense of scale to keep in mind rather than treating the made-up numbers above as anything more than a teaching example.

The numerator can attack the denominator

There’s a further complication: the amount of liquidity available isn’t fixed either. It can shrink exactly when you need it most.

As one-way selling gets urgent, the people who’d normally buy can pull back — show less size, widen their prices, or simply stop competing for the trade.

Even the U.S. government bond market — normally the deepest, most liquid market on Earth — saw this happen badly in March 2020. That episode was mostly a “everyone needs cash at once, and the banks in the middle can’t keep up” story, with several contributing causes that economists still argue about the relative size of. It’s mentioned here for one narrow reason: to show that even the deepest market in the world can have its available liquidity collapse under enough stress. It isn’t offered as proof that this piece’s three-clock story specifically was what happened in bonds that week — that’s a different, related situation with its own plumbing.

The Federal Reserve eventually stepped in on a massive scale, committing to buy at least $500 billion of Treasuries and $200 billion of mortgage-backed securities on March 15, then removing any fixed cap on March 23 and buying whatever was needed to keep the market functioning.

That same month also triggered the market’s automatic circuit breaker four separate times — March 9, 12, 16, and 18.

So the amount of forced selling and the amount of available liquidity aren’t independent of each other. More selling can shrink the liquidity that’s supposed to absorb it, while volatility rises at the same time. That’s the dangerous feedback loop: the pressure itself can make the market’s ability to handle that pressure get worse.

Different kinds of selling aren’t automatically equivalent

There’s one more mistake worth correcting. An earlier, cruder version of this idea just added up the dollar amounts from each source — margin selling, plus ETF rebalancing, plus dealer hedging — as if a dollar from each one pushed the price by the same amount.

It doesn’t.

Margin selling is often plain stock sales. ETF rebalancing might happen through futures or swaps instead. Dealer hedging might happen through index futures rather than individual stocks. These flows can move between markets through arbitrage — but a dollar of futures selling doesn’t necessarily move the price the same amount as a dollar of actual stock sold.

So a more honest version of “total selling pressure” isn’t a simple sum. It has to account for which market each flow actually lands in, and how each one translates into real price pressure once it gets there. That’s harder to pin down, but it’s the more accurate way to think about it — even though it means the real number is something nobody outside the dealers, ETF managers, and swap desks involved could actually see in real time. That’s a real cost, not a free improvement. It’s still the right trade to make. It’s just worth naming honestly.

A quick note on the math, for anyone who wants it (entirely skippable): economists who study this have found that for a single large trade, price impact tends to grow roughly with the square root of the trade’s size relative to normal daily trading volume — bigger trades move the price more, but less than proportionally. The commonly cited form looks like this:

Where I is the price impact, Q is the size of the trade, σ_D is that day’s volatility, V_D is that day’s normal trading volume, and c is a constant researchers fit from real data. This comes from a well-established line of research (Farmer, Gerig, Lillo and Waelbroeck; more recently tested again on Apple stock by Vasaikar, though that newer paper hasn’t been peer-reviewed yet).

Here’s the catch, and it’s an important one: that formula is tested against normal daily trading volume — the observable, published number. It has never been tested against the much harder question this piece actually cares about: how much liquidity is available right near today’s price, during an actual panic, which can be far smaller than a normal day’s volume. So while the underlying math is solid for calm, single-trade situations, it can’t just be plugged into a synchronized-crash scenario and trusted to give a real answer.

Which, again, is really the whole point of this piece: the one number that would let you actually calculate the danger is the one number nobody can see while it’s happening.

And timing matters just as much as the total. If $10 billion of forced selling happens at 2:00 and $10 billion of buying from a differently-positioned dealer happens at 3:00, the net effect might look like zero on paper — but the market lived through both moves, in that order, and that’s not nothing.

Worth naming plainly what this piece leaves out, rather than pretending it’s the whole picture: short sellers buying back their positions, companies buying back their own stock, and routine portfolio rebalancing can all supply real buying pressure during a decline. Dealers also adjust their hedges in response to changing volatility and the simple passage of time, not just price moves — those effects can shift hedging flows independently of anything discussed here. This piece is deliberately built around the amplifying side of the story, because that side was the one going unaccounted for. It isn’t a claim that nothing ever pushes back.

A warning line, not a magic number

Here’s a way to organize all of this feedback into one number. Call it R — the size of one round of price disturbance, divided by the size of the round before it.

You may recognize this shape from an unfortunately familiar context: it’s the same logic behind the “R number” used during COVID. An R above 1 means each round is producing more than the last, so it grows. Below 1, it shrinks and fizzles out on its own. This piece borrows that same logic for price disturbances instead of infections.

But this is not an official, measurable statistic — it’s a way of thinking about the problem, not something anyone is currently calculating in real time. And R going above 1 once doesn’t prove the market has crossed some fixed, guaranteed threshold into a crash.

Maybe the most leveraged accounts are already gone by the next round. Maybe new buyers show up. Maybe dealers’ positioning flips. Maybe the next round comes in much smaller and the whole thing dies out. It’s also possible for the broad market’s R to look fine while one specific corner of it — a semiconductor basket, say — is still amplifying on its own.

So R crossing 1 is a warning line, not proof that a phase transition has occurred.

(For anyone who wants the more technical version: researchers who study financial networks have a more rigorous cousin of this idea, based on representing the whole system as a matrix of who owes what to whom, and checking whether its largest eigenvalue exceeds one — a real, established way to test for this kind of instability in a simplified, straight-line version of the system. The real system isn’t straight-line, though: liquidity and constraints keep changing as a crisis unfolds, and there’s genuine research showing a system can pass that simplified test while still supporting a much messier kind of contagion once those complications are put back in. Either way, nobody has the account-by-account data needed to actually build that matrix for the real market. Which is exactly why this piece treats R as an intuition, not a calibrated instrument.)

A warning line above 1 tells you a cascade is feeding itself, in that moment. It doesn’t tell you what actually stops one — circuit breakers, trading pauses, ordinary buyers eventually finding a price they like, or, where regulators are actually allowed to do it, official intervention when private buyers won’t show up. Worth being careful with that last one here, since this piece has been about stocks specifically: the central bank’s intervention in March 2020 was in government bonds, mortgage bonds, and short-term lending markets — not direct purchases of stocks. What actually happens when private buyers disappear in the stock market itself, rather than the bond market underneath it, is a genuinely different and more interesting question.

That’s Part 5.

Regards,
Luke Kandia
Chasing Clarity

Luke Kandia writes Chasing Clarity, a Substack covering healthcare systems, financial markets, public policy, and science communication — usually by starting with something small, like an old coin, a data point, or a footnote, and following it wherever it actually leads.

Nothing here is investment, financial, or legal advice — just a curious fellow who is currently wondering where the margin debt went in July 2026. Do your own homework before you make decisions with your money.

Sources and references

1. Federal Reserve — Monetary Policy Report, June 2020. Retrospective discussion of extraordinary financial-market stresses and Federal Reserve intervention during March 2020. Federal Reserve — Monetary Policy Report, June 2020

2. Federal Reserve — Financial Stability Report, May 2020. Documents severe deterioration in liquidity and market functioning during the March 2020 shock, including stresses in normally deep markets. Federal Reserve — Financial Stability Report, May 2020

3. Federal Reserve — Implementation Note, March 15, 2020. Directed purchases of at least $500 billion of additional Treasury securities and $200 billion of agency mortgage-backed securities. Federal Reserve — March 15, 2020 Implementation Note

4. Federal Reserve — “Federal Reserve announces extensive new measures to support the economy,” March 23, 2020. Shifted Treasury and agency-MBS purchases to the amounts needed to support smooth market functioning and effective monetary-policy transmission. Federal Reserve — March 23, 2020 Measures

5. NYSE — “Assessing NYSE Model Performance During 2020 Volatility.” Confirms activation of the Level 1 market-wide circuit breaker on March 9, March 12, March 16 and March 18, 2020. NYSE — Assessing NYSE Model Performance During 2020 Volatility

6. Farmer, J. Doyne; Gerig, Austin; Lillo, Fabrizio; Waelbroeck, Henri — “How Efficiency Shapes Market Impact.” Foundational work on concave market impact and the approximate square-root relationship between directional order size and price impact. NBER Conference Paper — How Efficiency Shapes Market Impact

7. Vasaikar, Aniket — “Empirical Confirmation of the Square-Root Law of Market Impact in a U.S. Large-Cap Equity,” June 2026. Independent, single-author preprint using reconstructed Apple metaorders from Nasdaq order-book data, finding support for a square-root impact specification calibrated against V_D (daily traded volume). Not yet peer reviewed; cited here as supporting color alongside the established Farmer et al. literature, not as an equally weighted result. arXiv — Empirical Confirmation of the Square-Root Law

8. KCMI (Korea Capital Market Institute) — “Market Trends and Implications Around the Launch of Single-Stock Leveraged and Inverse ETFs,” July 2026. Source for the 1.6%/2.1% average daily rebalancing figures for Samsung Electronics and SK Hynix cited as a real-world scale anchor in this piece (previously cited in Part 2). KCMI — Market Trends and Implications

9. The Korea Herald — “Samsung leveraged ETFs outtrade underlying shares; Tesla’s never have.” Source for the ₩3.7 trillion vs. ₩1.25 trillion Samsung leveraged-ETF-turnover figure cited in this piece (previously cited in Part 2). The Korea Herald — Samsung leveraged ETFs outtrade underlying shares

10. SEC Rule 10b-16 (17 CFR § 240.10b-16). Paragraph (a)(1)(iv) requires brokers to disclose at account opening the specific conditions under which interest rates can change without prior notice — the basis for a pre-disclosed, benchmark-linked base rate repricing without a fresh notice. Paragraph (b) separately requires not less than thirty days’ written notice for changes to the disclosed terms and conditions themselves, except where a change lowers the customer’s rate. Cornell Legal Information Institute — 17 CFR § 240.10b-16

11. Federal Reserve — “Market Liquidity in Treasury Futures Market During March 2020.” Source for the finding that cash-futures basis traders were not found to be an important driver of the March 2020 disruption specifically in the Treasury futures market, complicating a simple basis-trade-unwind narrative. Federal Reserve — Market Liquidity in Treasury Futures Market During March 2020

12. Bardoscia, Marco; Battiston, Stefano; Caccioli, Fabio; Caldarelli, Guido — “Pathways towards instability in financial networks,” Nature Communications, 2017. Source for the dominant-eigenvalue/spectral-radius condition as a local-stability criterion in linearized financial-network models, referenced in this piece’s technical aside on the warning-line section. Nature Communications — Pathways towards instability in financial networks

This story began with what seemed like a harmless question.

How do you get a prospector’s licence in Manitoba?

The answer surprised me. An individual lifetime prospecting licence costs all of $15.

That sent me wandering through Manitoba’s iMaQs mining maps, looking at old claims, abandoned mines and mineral belts. Before long I was staring at names that were much more than dots on a map to me: Wallace Lake, Bissett, Bird River, Cat Lake, Aikens Lake, the Gammon and the Bloodvein.

I need to explain why those names stopped me.

I had been travelling through this country for almost fifty years without ever really understanding what lay beneath it.

I knew the country long before I knew the rocks.

And that, as it turns out, is part of the story.

I knew the country before I knew the rocks

In 1977, when I was 15 and in Grade 10 at Winnipeg’s storied St. John’s High School, I joined 31 others on a canoe expedition through eastern Manitoba.

The trip was supposed to cover roughly 160 miles in 14 days.

It took us 16.

The man leading us was our geography teacher, Gerry Nicholson.

Gerry always seemed to me like a man slightly more out of place standing in front of a classroom than he was with a paddle in his hands. Put him somewhere in the Canadian Shield, dressed like a lumberjack from another century, and he looked entirely at home.

Only much later did I discover just how deep that connection went.

When Manitoba celebrated its centennial in 1970, one of the major undertakings was Brigade ’70, a huge canoe event involving dozens of boats converging on Winnipeg. The official Manitoba Centennial Corporation records list G. Nicholson as chairman of the Brigade ’70 committee.

Another member of Gerry’s Brigade ’70 committee was Don Starkell.

The same Don Starkell from Winnipeg who, in 1980, would push off from the Red River in East Kildonan and paddle more than 12,000 miles to Belém, Brazil, near the mouth of the Amazon.

By canoe.

That Don Starkell.

The journey would become the subject of Paddle to the Amazon and establish Starkell as one of Canada’s legendary long-distance canoeists.

And there he was, a decade before departing for the Amazon and seven years before our school expedition, serving alongside my geography teacher, Gerry Nicholson, on the committee organizing Brigade ’70.

The same Manitoba Centennial Corporation records also place Nicholson on its Aquatic Events committee. This wasn’t simply a geography teacher with an unusually ambitious outdoor hobby. Gerry was deeply involved in Manitoba canoe culture before I ever walked into his classroom.

So when Gerry decided in 1977 that taking a group of Grade 10, 11 and 12 students on a 160-mile wilderness canoe trip was a perfectly sensible educational exercise, he wasn’t exactly experimenting.

This was Gerry’s world.

We started at Wallace Lake, then worked our way through a series of lakes and some decidedly non-trivial portages toward Obukowin Lake and Carroll Lake. There was a fishing lodge at Carroll that I remember as newly built.

Eventually we reached Aikens Lake, where there was another fishing lodge, and where we stayed overnight largely because the fishing was too good to leave.

From there we travelled northwest on the Gammon River until reaching the Bloodvein River, which carried us west through some spectacular rapids and eventually onto Lake Winnipeg at what we then called the Bloodvein Reserve, today Bloodvein First Nation.

There were no cell phones, GPS units or satellite communicators quietly sending our coordinates back to Winnipeg. Gerry got us turned around a few times, although never seriously and usually for only a few hours.

Looking back, “lost” may even be too strong a word. Gerry generally knew approximately where we were.

It was the lakes that occasionally disagreed with him.

Mostly, though, we were late because the fishing kept interfering with the itinerary.

At fifteen, I considered that an entirely legitimate reason to be late.

Submarine Moose and the fish that stopped an expedition

One morning, on a lake I remember only as Moose Lake, we watched a moose enter the water.

Being teenagers, several of us naturally decided that paddling after it would be an excellent idea.

We discovered very quickly that a canoe full of enthusiastic teenagers was no match for a swimming moose. Someone christened our aquatic friend “Submarine Moose,” and the name fit beautifully. It moved through the water with astonishing efficiency while we thrashed away behind it.

The moose won.

The more enduring memory from that morning, though, involved fish.

A light drizzle had begun as we broke camp. By that point we’d experienced enough serious rain that this barely qualified as weather. We ate our usual breakfast of oatmeal and dried fruit and nuts, loaded the canoes and pushed off.

Every canoe carried at least one fishing rod, so lines went into the water almost immediately.

Within minutes, virtually every canoe had a three- or four-pound northern pike — a jack — on the line. It didn’t seem to matter what lure anyone was using.

The lake had apparently decided we weren’t leaving.

So we turned around, brought out the heavy cast-iron frying pans Gerry insisted were worth carrying over every portage, cleaned and battered the fish, and proceeded to have one of the finest fish meals I can remember.

It probably cost us half a day.

Almost fifty years later, I can still remember that fish fry.

The itinerary may have said otherwise, but Gerry was right about the frying pans.

Sleeping above the Bloodvein

Another night remains equally fixed in my memory.

We had reached a major set of rapids on the Bloodvein where the river narrowed through something resembling a small canyon. The water ran perhaps 30 or 40 feet below us, while above it was exposed Canadian Shield.

That was where we slept.

We simply rolled our sleeping bags out on the rock.

I had seen truly dark skies before while camping with Plast, the Ukrainian scouting organization, but this was the first time I remember actually falling asleep beneath one.

There was no moon. Thousands upon thousands of tiny pinpricks filled the sky. We watched satellites move slowly overhead and waited for shooting stars.

Sleeping uncovered in northern Manitoba normally introduces one rather obvious complication: mosquitoes — the scourge of the Manitoba north.

But the rapids below us created their own little weather system. Huge volumes of water crashed over and around the rocks, moving enough air across the cliff to create a persistent breeze.

It was just enough to keep the mosquitoes away.

For one night, a group of teenagers lay directly on the Canadian Shield and slept beneath the stars.

I have never forgotten it.

Apparently, we were missing

There is another memory from that trip that became considerably funnier with the passage of time.

Remember, we were supposed to be gone for 14 days.

We were now on Day 16.

From our perspective, nothing was particularly wrong. We were paddling. We were fishing. Everybody was alive.

From the perspective of parents sitting hundreds of kilometres away in Winnipeg, however, 32 people had simply disappeared into the Manitoba wilderness.

There was no text message saying:

Fishing fantastic. Two days late. Don’t worry.

As I remember it, enough parents had contacted the RCMP that on the second-last day of the trip, somewhere before reaching Bloodvein, we suddenly heard something enormous approaching.

A Canadian Armed Forces helicopter tasked with search-and-rescue appeared overhead.

Someone had apparently decided it might be prudent to come looking for us.

I don’t remember precisely how Gerry communicated with the pilot. What I do remember is Gerry deliberately waving his paddle overhead in a very particular manner.

He looked like a man who had done this before.

Whether he actually had, I never thought to ask.

The helicopter eventually departed. We continued paddling and emerged onto Lake Winnipeg two days late, everyone intact, carrying considerably better memories than we might have had if we’d kept rigidly to the itinerary.

If only we’d known

Nearly half a century later, I opened Manitoba’s modern mining maps and one particular name jumped off the screen.

Wallace Lake.

There was Bissett beside it, due west off Hwy 304. Sitting on the north shore of a lake.

Rice Lake.

There was the Rice Lake gold country.

The map was dotted with old mines, shafts, prospects and claims spread throughout country we had travelled without ever giving much thought to what lay beneath our feet.

For more than two weeks we had scrambled over outcrops, carried canoes across exposed Canadian Shield and travelled through some extraordinary mineral country.

We brought fishing rods.

We did not bring geological hammers.

If only we’d known.

Although perhaps that was fortunate. Thirty-two teenagers deciding that every interesting chunk of quartz needed to accompany them for the remaining 150 miles might have sunk the canoes.

That 1977 trip began a relationship with eastern and northern Manitoba that kept resurfacing throughout my life.

Years later, friends and I returned to the same general country by vehicle, canoe and on foot. We travelled into the Cat Lake area and visited the abandoned Irgon lithium mine site, back when lithium was sufficiently uninteresting that a developed mine site could sit there largely forgotten.

Decades later, lithium stopped being quite so uninteresting.

We travelled through Bissett without really appreciating that we were standing in the middle of Manitoba’s historic gold country.

Then, in 2016, almost forty years after Gerry took us out of Wallace Lake, my work took me much farther north.

To Garden Hill on Island Lake.

Different age. Different reason.

Same Shield.

This time I wasn’t carrying a paddle.

I was carrying a laptop.

I had been there

In October 2016, while I still owned an IT services company, I flew north to audit the state of the network infrastructure at the Garden Hill Band Office.

My report from that trip begins with me landing at Island Lake airport and trying to figure out how I was supposed to get across the water to Garden Hill. Eventually someone told me to walk down toward the lake and find a “taxi.”

It turned out that an Island Lake taxi could be a small aluminum fishing boat.

Eventually I found the dock, raised my arm at a passing boater and watched him turn around and come back.

“Can I get a ride across?”

“Hop in.”

So hop in I did.

The boat was about 15 feet long. I couldn’t see a PFD for him or for me. I had a backpack full of computer equipment, a feather parka and a cold, wet aluminum bench for a seat.

Then I spotted a square of cardboard.

It had probably never looked so inviting.

I put it on the bench, sat down and my captain opened the throttle. About a minute later I was standing on the other side.

Ten dollars.

Not bad for a one-minute taxi ride.

Garden Hill gave me an immediate appreciation for what remote actually means.

There is no permanent road connecting the community to southern Manitoba. Boats and aircraft are fundamental during open-water season; in winter, ice roads temporarily change the transportation equation.

In the report I wrote afterward, I described the mud, the roads and the enormous difficulty of doing things there that most of us barely think about. I walked past a vehicle abandoned with a flat tire and reflected on how something trivial in Winnipeg could become a serious problem in a place where ordinary services were sparse.

I was there looking at routers, cabling, Internet connections and computer infrastructure.

What I did not know while crossing Island Lake was that somewhere in that enormous maze of water and islands lay a tractor, two sledges and the remains of a gold shipment that had disappeared through the ice more than eighty years earlier.

And one man had remembered exactly where it was.

The train that never arrived

Gold had been discovered in the Island Lake area in the early 1930s, and Island Lake Mines Limited developed a genuine producing operation.

Manitoba’s official Mineral Inventory records 5,986.6 ounces of total gold production during 1934 and 1935.

On March 18, 1934, a tractor train set out carrying a shipment of gold concentrate.

Behind the tractor were two sledges

Contemporary accounts later described approximately 500 bags of concentrate, each weighing around 100 pounds. Taken literally, that works out to roughly 25 short tons.

Manitoba’s official Mineral Inventory gives the shipment as up to 27.2 tonnes, or about 30 short tons.

The surviving accounts therefore put the original load somewhere in the range of roughly 25 to 30 tons of concentrate.

The tractor train never arrived.

Somewhere on Island Lake, it broke through the ice, taking the sledges and their cargo down with it.

Contemporary newspaper accounts also reported that five sled dogs chained to the second sledge went down with the train.

Mounting a recovery operation in such a remote northern setting would have been extraordinarily difficult in 1934.

The ice closed over. The mine eventually closed too, and years became decades.

But a young Island Lake trapper named Abraham McPherson knew where the tractor had gone down.

Abraham McPherson remembered

According to Winnipeg Free Press reporter Bob Lowery, McPherson had been out on his trapline when the tractor train set out.

On his way home, he crossed part of the route the machine should have travelled and noticed something strange.

There were no caterpillar tracks.

When he reached the community, he learned that the tractor train had never arrived.

Spring eventually came. The snow disappeared from the lake, and McPherson went looking for the place where the train had broken through.

He found the hole.

Before breakup erased the evidence, he carefully lined its position up with landmarks along the shoreline.

Then he remembered them.

For 46 years.

No GPS. No satellite imagery. No waypoint stored in a phone. Just a young trapper standing on a frozen northern Manitoba lake, looking toward shore and committing a position to memory.

Over the years, others apparently became interested in finding the wreck. Lowery reported that McPherson refused to reveal its location unless he received a written agreement giving him an equal share of whatever was recovered.

Knowing where the gold is happens to be an excellent negotiating position.

Three men who didn’t know where the gold was

By 1980, gold had become extremely interesting again.

Prices had exploded during the late 1970s, briefly reaching US$850 an ounce in January 1980. Old mines, abandoned workings and forgotten deposits suddenly looked very different.

Three men now entered the story who knew about the gold but didn’t know where to find it.

There was Don McIvor, a prominent Manitoba Métis Federation leader and former mayor of Wabowden who would later serve as president of the MMF.

There was Bruce Kondratuk, a Manitoba Hydro engineer and experienced diver.

And there was Wayne Wamboldt, of Halifax.

They had heard about the lost shipment.

Lowery reported that they had initially been looking around Gods Lake, roughly 120 kilometres away, before the search eventually led them to Abraham McPherson.

On June 18, 1980, the men approached him.

According to the Free Press, it took only about 15 minutes to make the deal.

McPherson would tell them where the tractor had gone down. In return, he became a full one-quarter partner.

The other three would finance the recovery operation until the concentrate reached the surface and was delivered for refining.

Four equal partners.

After nearly half a century, Abraham McPherson took them back onto Island Lake.

Off by 20 feet

Bob Lowery wasn’t reporting this story entirely from a desk in Winnipeg.

He travelled to Island Lake, photographed the men at work and documented the salvage operation while it was happening.

According to Lowery’s July 11, 1980 report, McPherson guided the men to the position he had carried in his head since 1934.

A diver went down.

The wreck was there.

Lowery reported that McPherson’s remembered position was off by only about 20 feet.

After 46 years.

They hadn’t merely found gold-bearing sediment somewhere in the vicinity. The Linn tractor itself was still sitting on the bottom.

So were the sledges.

The crew attached a rope to the axle of the submerged tractor, giving divers a direct guide through roughly 70 feet of water to the wreck.

The treasure hunt was over.

Recovering the treasure was another matter entirely.

Building a gold vacuum

The recovery system they devised was crude, ingenious and remarkably practical.

The men built two barges, each roughly 18 feet square, using steel drums and lumber. According to the Free Press account, a sludge pump mounted on the recovery platform drew material through a flexible hose from the lake bottom.

At the diver’s end was essentially a vacuum nozzle.

The diver worked around the wreck, sucking concentrate into the hose. At the surface, the gold-bearing slurry emptied into a 45-gallon steel drum where the heavier material settled. Water and finer material overflowed through additional settling drums before the water was eventually returned to the lake.

The Free Press reported that divers worked in roughly 30-minute shifts, using the rope attached to the tractor axle as their guide down to the wreck.

Four men had effectively built an improvised underwater mineral-recovery plant in one of the more inaccessible parts of Manitoba.

And it worked.

Just not nearly as well as they initially hoped.

Ten ounces per ton

By August 1980, the Free Press reported that the recovered concentrate had assayed at approximately 10 ounces of gold per ton.

That was the grade of the concentrate, not the original rock coming out of the mine.

The material had already been processed into concentrate, removing much of the waste rock and concentrating the gold-bearing fraction. Ten ounces per ton works out to roughly 340 grams of gold per tonne of concentrate — a rich product, certainly, but entirely different from claiming that the original mine ore itself ran 340 grams per tonne.

That figure finally gives the lost shipment some scale.

Using the surviving estimates for the size of the load, roughly 25 to 30 short tons, the shipment may have contained somewhere around 250 to 300 ounces of gold.

Some members of the crew estimated in 1980 that the entire load might gross somewhere around $150,000 to $200,000.

McIvor dreamed considerably bigger.

He speculated that the concentrate might contain additional “chemically-bound gold” that wasn’t being reflected in the assay and imagined a multimillion-dollar payday.

His partners were considerably more restrained.

Nothing I have found suggests subsequent refining ever vindicated McIvor’s theory. But there is something wonderfully human about the episode. Four men finally find a treasure that has been underwater for 46 years, somebody hands them an assay result, and almost immediately one of them begins wondering whether there might secretly be even more gold.

Gold fever apparently survives immersion.

As I finish this piece at the end of August 2026, gold is trading at roughly C$6,195 per troy ounce. At that price, 250 to 300 ounces would represent approximately C$1.55 million to C$1.86 million in gross metal value.

Not bad for something that went through the ice during the Depression.

Six weeks becomes four months

The men began the recovery with considerable optimism.

The Free Press reported that Kondratuk thought they might recover as much as 80 percent of the concentrate and that the job could perhaps be completed in six weeks.

Instead, the operation stretched into roughly four months.

The reasons aren’t difficult to imagine. Diving at 60 or 70 feet in a remote northern lake is slow work. Visibility changes. Pumps clog. Equipment breaks. Divers can remain down for only so long.

Every hose, drum, replacement part and gallon of fuel has to get there somehow.

Then everything recovered has to get back out.

By the end of the operation, Manitoba’s Mineral Inventory records that approximately nine tonnes of concentrate had been recovered from a depth of 21 metres during four months of dredging.

In late October 1980, 140 bags were flown to Thompson.

Then something rather odd happened.

The recovered concentrate sat there for more than a year.

It remained under lock and key until November 1981, when it was finally sent to Hudson Bay Mining & Smelting in Flin Flon for processing. Manitoba’s mining record independently confirms that sequence.

The treasure hunters meet arithmetic

On January 4, 1982, Lowery caught up with McIvor again.

McIvor still didn’t know exactly what the final refined shipment would produce. He told the Free Press that, based on 1980 gold prices, the recovered concentrate might gross somewhere between $40,000 and $70,000.

Recovery costs were estimated at around $25,000.

That is the less romantic side of treasure hunting.

Finding lost gold and making money recovering lost gold are two entirely different propositions. The men had spent months constructing barges, operating pumps, diving, moving equipment, flying concentrate out of Island Lake and eventually transporting it to a smelter.

Even with gold near historic highs, nobody was guaranteed a fortune.

But one sentence in that January 4, 1982 story completely changes the ending.

McIvor said there was a good chance they would return to recover a:

“large amount of the concentrate which was left on the sleds.”

That wording matters.

He wasn’t talking about microscopic traces of gold scattered somewhere through acres of lake-bottom mud. One of the men who had spent months working directly over the wreck believed a large amount of concentrate remained on the sledges themselves.

McIvor said another recovery attempt might be made the following summer.

Summer 1982.

And then the story disappears

So I went looking for the next chapter.

I searched 1982, then 1983, then everything after that.

I searched the Winnipeg Free Press archives for Don McIvor, Abraham McPherson, Bruce Kondratuk and Wayne Wamboldt. I searched Island Lake, gold concentrate and salvage, along with combinations of those names and terms.

As far as I can determine, the January 4, 1982 article is the last Winnipeg Free Press story connecting those four men with the Island Lake gold all the way through 2026.

Forty-four years of silence.

That doesn’t prove they never returned.

Perhaps they did quietly. Perhaps the final refinery settlement demonstrated that another recovery attempt made no economic sense. Perhaps ownership became complicated. Perhaps the remaining concentrate was buried too deeply. Perhaps another attempt failed without attracting newspaper attention.

Or perhaps four men who had already devoted months of their lives to an extraordinary adventure simply moved on with their lives.

What makes the Island Lake story so fascinating is that almost none of its basic structure depends upon legend.

The mine existed. The shipment existed. The Manitoba government records the March 18, 1934 loss of 500 bags containing up to 27.2 tonnes of gold concentrate. It records McIvor, Kondratuk, Wamboldt and McPherson by name, the nine tonnes they recovered from 21 metres of water, the 140 bags flown to Thompson, and their eventual shipment to Flin Flon for processing.

The contemporary Free Press reporting supplies the remarkable human details: McPherson’s shoreline bearings, the tractor still sitting on the bottom, the improvised gold vacuum, the reported 10-ounce-per-ton assay and McIvor’s statement that substantial concentrate remained on the sledges.

The mystery isn’t whether the treasure existed.

It’s how the story ended.

I crossed that water

Reading Lowery’s stories eventually sent me back to the report I had written after my own Garden Hill trip ten years earlier.

Near the end of that day in 2016, I needed another boat taxi to get back to Island Lake airport.

This time the lake was choppy.

The driver opened the throttle and the aluminum hull began pounding the waves hard enough that I remember hanging onto the gunwales with what I described at the time as a “death-like grip,” laptop and all.

At one point I became convinced the driver was heading the wrong way.

“I think the airport is that way,” I helpfully offered.

He ignored my contribution to northern navigation.

A moment later I understood why.

He was steering toward the protection of the shoreline, where the waves almost disappeared. Once we reached calmer water, the remainder of the trip was remarkably smooth.

I tucked my back-seat boating expertise away.

When we reached the airport, I handed him twenty dollars and told him to keep the change.

He had earned it.

The strange thing, looking back ten years later, is that I had travelled directly through the setting of this story without knowing the story existed.

Somewhere out there was the place where a tractor train carrying gold had vanished beneath the ice in 1934. Somewhere out there, 46 years later, Abraham McPherson had looked toward shore and reconstructed a set of bearings he had carried in his head since he was a young man.

The connection only became visible decades later.

In 1977, Gerry Nicholson had taken a fifteen-year-old kid through Manitoba’s Shield country with a canoe paddle and a fishing rod.

Years later, I returned to Cat Lake and Bissett without understanding much about the mineral wealth beneath them.

In 2016, I crossed Island Lake carrying a laptop.

Only in 2026, staring at mining maps on a computer screen, did all those pieces of geography begin connecting to one another.

I had crossed Island Lake with a laptop.

Four men before me had crossed it looking for gold.

And they found it.

Finding Wayne

That eventually brought me to Wayne Wamboldt.

Following the four men forward through old newspaper archives, I found a 1989 marriage announcement for a Wayne Wamboldt of Halifax. Then I found a 1999 Manitoba newspaper story involving a Wayne Wamboldt working in the water-treatment business.

Eventually I found a LinkedIn profile for a retired Director of Waste Management in Truro, Nova Scotia, who had attended the University of Manitoba.

The pieces fit unusually well.

I still don’t know with certainty whether he is the same Wayne Wamboldt who stood on a homemade barge on Island Lake in 1980, helping haul gold-bearing sludge out of 70 feet of water.

So I sent him a message.

I explained that I was researching the Island Lake story and had found the old Winnipeg Free Press articles naming a Wayne Wamboldt of Halifax as one of the four men involved. I told him that the newspaper trail seemed to end in January 1982, with a large amount of concentrate reportedly still left on the sledges.

Then I asked whether I had found the right Wayne.

As of this writing, I haven’t heard back.

Perhaps I have the wrong man. Perhaps the message disappeared into the black hole where unsolicited LinkedIn messages go to die. Or perhaps, after 46 years, Wayne simply has no interest in revisiting a wild chapter of his younger life.

I hope he does.

Because if I found the right Wayne, there is still one question I desperately want to ask him:

Did you ever go back?

Regards,
Luke Kandia
Chasing Clarity

Luke Kandia writes Chasing Clarity, a Substack covering healthcare systems, financial markets, public policy, history and science communication — usually by starting with something small (an old coin, a data point, a footnote or an old photograph) and following it wherever it actually leads.

Nothing here is investment, financial, or legal advice — just a retired IT guy who now can’t stop wondering if the boys ever went back. Do your own homework before you make decisions with your money.

Sources and notes

The principal contemporary sources I have personally examined for the salvage story are Bob Lowery’s reports in the Winnipeg Free Press:

July 11, 1980 — “Four find lost Island Lake gold”

August 22, 1980 — “Group hopes to net lake’s cache of gold”

January 4, 1982 — “Gold from lake bottom on way to bank”

Manitoba’s Mineral Inventory also cites a June 22, 1981 Winnipeg Free Press article in connection with the nine-tonne recovery after four months of dredging. I have not yet examined that clipping directly, so details derived from it here come from the provincial Mineral Inventory rather than from my own reading of the newspaper.

The Province of Manitoba’s Island Lake Mine Mineral Inventory independently documents the March 18, 1934 loss of the 500 bags, the names of the four members of the later salvage operation, approximately nine tonnes recovered from 21 metres of water, the 140 bags flown to Thompson and their eventual shipment to Hudson Bay Mining & Smelting in Flin Flon.

The Manitoba Historical Society’s records for the Manitoba Centennial Corporation list G. Nicholson as chair of the Brigade ’70 committee and Donald G. Starkell among its members. The same records also list G. Nicholson on the Aquatic Events committee.

The Manitoba Historical Society’s account of Paddle to the Amazon identifies Don Starkell and his sons as East Kildonan residents and records their 1980 departure from the nearby Red River on the journey that would eventually carry Starkell more than 12,000 miles to Brazil.

The descriptions of my 2016 arrival at Island Lake, Garden Hill and the boat crossings come from the infrastructure/network assessment I wrote during that trip.

And somewhere beneath Island Lake, a Linn tractor may still be sitting where it disappeared through the ice more than ninety years ago.

Perhaps with some of the gold still beside it.

Two years ago, I was getting thoroughly fed up with our original Starlink system at the trailer.

It worked beautifully — for about 40 - 50 seconds at a time.

Then, seemingly every minute, the connection would disappear for 10 or 20 seconds before coming back.

For browsing the web, it was annoying.

For anything requiring a continuous Internet connection, it was maddening. Thank god for buffering technology.

The problem was trees.

We had one of the original first-generation Starlink dishes, the big round one with the motorized positioning system. Our trailer sits in a heavily treed area north of Gimli, Manitoba, and the dish sitting on top of the trailer simply didn’t have a clean enough view of the sky.

So I came up with the obvious IT-guy solution.

No not buying a chainsaw, though honestly that thought did cross my mind. More than once.

Build a tower.

If you can’t see the horizon, go higher

Not a little pole.

A proper satellite tower high enough to get the Starlink dish above the treetops, giving it a much larger, unobstructed view of the sky.

We priced it out. The contractor that finally came out all the way to Gimli to scope out the installation said that it wouldn’t be a problem. It would cost though.

About $5,000. Including a concrete pad, and a tower about 50 ft tall.

We approached the trailer park board and asked permission to install it.

They said no. Nice try. But no.

I wasn’t particularly happy about that at the time.

This summer, we were planning to try again — with renewed vigor. I was ready to make the case that reliable Internet had become about as essential at the trailer as electricity and running water.

Then Manitoba summer intervened.

We left the trailer during a stretch of extremely hot, humid weather. There was no air conditioning running while we were away.

When we came back, the Starlink that I had powered up at the beginning of the season, was dead.

I pulled the cable from the old power brick and stared at the built in connectors.

Corrosion.

Moisture.

It probably didn't require extraordinary humidity. Sustained summer humidity, dust and repeated temperature swings would have created ideal conditions for corrosion. If those exposed contacts ever cooled below the dew point, even briefly, a microscopic film of condensation could form on them. Add electrical power and several summers of accumulated contamination, and corrosion had everything it needed.

Wonderful.

The old Starlink system was toast, rather than go hunting for another Gen V1 PoE power brick, we replaced it with the current generation. A square dish. No motor. Actually smaller it seemed than the old one.

As I installed it, I noticed that the ends of the cables not only came out of the satellite (my old gen had a permanently mounted 100 ft cable attached at the satellite end), but that both ends had tight fitting rubber grommets where the cable met the hardware - leaving little opportunity for air or moisture to reach the electrical contacts and corrode them.

I plugged it in and used the app to align it with the Starlink satellites in the sky.

Which led me to another rather astonishing fact.

SpaceX has now launched more than 11,000 Starlink satellites, with roughly 10,700 considered active in orbit.

Think about that number for a moment.

As of August 2026, there are only about 16,000 active satellites of all kinds orbiting Earth. In other words, roughly two out of every three working satellites up there are now part of Starlink.

And this happened incredibly quickly.

At the end of 2022 there were 6,718 active satellites orbiting Earth. Of those, 3,394 were Starlinks. Everything else humanity had in orbit amounted to 3,324 active satellites.

Starlink had passed them by 70.

Starlink had essentially reached the point where its satellites outnumbered every other active satellite humanity had in orbit — combined. In late 2022. Having launced its first sattelite only 5 years prior to that momentous day.

Think about the speed of that.

Sputnik 1 became humanity’s first artificial satellite in October 1957. It then took the world 65 years to build an active satellite population of roughly 3,300 non-Starlink spacecraft.

Starlink’s first two experimental satellites didn’t reach orbit until February 2018.

Yet by the end of 2022 — less than five years later — Starlink had 3,394 active satellites in orbit, slightly more than everybody else combined.

Sixty-five years of humanity putting satellites into orbit.

Starlink matched it in less than five.

And that was only three and a half years ago.

Since then, SpaceX has launched thousands more.

Which probably helps explain what happened next.

I went onto the roof. Placed the unit down in the middle of the roof. Connected the cable to the satellite dish. Ran the cable down the side of the trailer thru a built in access hole, plugged by steel wool for those pesky rodents. Ran the app. Aligned the unit.

And the Internet just...

worked.

I started pinging continuously.

Ping.

Ping.

Ping.

Ping.

No 20-second gaps.

No disappearing Internet.

No waiting for the system to recover.

Nada.

The trees hadn’t moved.

The trailer hadn’t moved.

And I certainly hadn’t spent $5,000 putting a satellite dish 50 feet into the air.

Starlink itself had changed.

There were vastly more Starlink satellites overhead than when we first installed the system in 2021, newer ground hardware — my motorless Starlink Standard 4 terminal — and a far denser and more sophisticated network for handing my connection from one rapidly moving satellite to another.

Whatever combination of obstructions, satellite availability and handoffs had plagued our original installation was effectively gone.

Which meant something else was gone too.

The tower.

We didn’t need it anymore.

The approximately $500 failure sitting in front of me had just eliminated the need for a $5,000 solution.

Net savings:

$4,500.

And suddenly I had a considerably different opinion about the trailer park board refusing my tower two years earlier.

They hadn’t cost me two years of better Internet.

They had inadvertently prevented me from spending five grand solving a problem that technology was about to solve for me anyway.

Two more things

The experience also changed my thinking about buying the replacement hardware.

Starlink offered me two choices.

Buy the new hardware outright for about $499.

Or rent it for $10 a month.

I rented it.

At $10 a month, I’ll spend $360 over three years. The mathematical break-even point against a $499 purchase is a little over four years.

But that’s not really why I chose it.

My first Starlink lasted about five years. In those five years the hardware changed enormously. The dish changed. The connectors changed. The router changed. The satellite constellation exploded in size. And the service itself became dramatically better.

What will Starlink hardware look like five years from now?

I have absolutely no idea.

And that is precisely the point.

I’m no longer particularly interested in owning the hardware. Like I’m not interested in owning a home.

The rented equipment remains Starlink’s property. If three years from now an entirely different generation of terminal makes today’s dish look as antiquated as my old round Gen V1 looks today, I haven’t sunk another $499 into a piece of rapidly aging technology.

I’m paying for flexibility.

There was another improvement as well.

Seasonal service used to be clumsy. Years ago, when we shut down the trailer for winter, the practical solution was essentially to cancel Starlink and reactivate it the following spring.

Today Starlink offers Standby Mode on eligible plans. Instead of completely disconnecting, the terminal can remain on a low-speed standby service and be switched back to full service when needed.

For somebody who uses a trailer for roughly six months of the year, that’s a much more sensible arrangement.

There is a caveat: Starlink’s current rules exclude some rental kits from Standby Mode, so whether that option is available depends on the particular hardware and plan. And Standby Mode does not technically guarantee that your old Residential plan will still have capacity when you return.

No guarantees. But I haven’t heard of Starlink refusing anyone recently either.

Still, the direction is obvious.

The hardware is getting better.

The constellation is getting bigger.

And the service is becoming much easier to turn on, turn off and adapt to the way people actually use it.

Lessons learned

There’s probably a broader lesson in there somewhere.

Sometimes the machine that finally dies isn’t the problem.

Sometimes it’s the thing that keeps you from spending a fortune keeping an obsolete solution alive.

And sometimes being told no by a trailer park board turns out to be worth $4,500.

A little Manitoba heat.

A little humidity.

A little condensation.

And poof.

My old Gen V1 Starlink died.

And saved me $4,500.

Regards,
Luke Kandia
Chasing Clarity

Lucas Kandia writes Chasing Clarity — a retired IT guy who writes about governments, economics, public policy, medical issues and pretty well anything else that interests him.

BTW, this isn’t a plug for Starlink. Just the plain, honest truth. And as it turns out, truth is sometimes stranger than fiction — and occasionally a lot more relevant.

References

1. Union of Concerned Scientists. Stephen Young, “How Many Satellites Are in Space? The Spike in Numbers Continues,” July 11, 2023. At the end of 2022, UCS counted 6,718 active satellites, including 3,394 Starlink satellites.
   UCS satellite-count article

2. Jonathan McDowell, Jonathan’s Space Report. Satellite and Debris Population: Past Decade. As of August 13, 2026: 16,278 active satellites, including 10,742 active Starlinks.
   Jonathan’s Space Report statistics

3. Jonathan McDowell, Jonathan’s Space Report. Starlink Satellite Statistics. Updated August 27, 2026; detailed launch, orbital-status and operational records for the Starlink constellation.
   Starlink satellite statistics

4. Space.com. “Starlink Satellites: Facts, Tracking and Impact on Astronomy,” updated August 27, 2026. Reports 11,102 Starlink satellites launched as of that date.
   Space.com Starlink satellite count

5. Starlink. “Obstructions Explained.” Explains how terminals build real-time obstruction maps and proactively switch between satellites many times per minute to route around trees and other obstructions.
   Starlink Beam Switching

6. Starlink Support. “What is the rent Starlink option?” Canadian rental hardware is currently offered for $10/month in eligible markets. Starlink notes that rental kits remain subject to specific service-plan restrictions.
   Starlink Canada rental information

7. Starlink Terms of Service. Rental equipment remains the property of Starlink and must generally be returned following cancellation.
   Starlink Terms of Service

There is an old lodge sitting on the south shore of Lake Manitoba.

It is boarded up now.

The windows are covered. Most of the buildings that once surrounded it are gone. The students, researchers, naturalists and visitors who once filled the place have disappeared.

But the lodge is still there.

And I know it because I once slept there.

Long before I knew anything about its history—before I knew who built it, before I understood the extraordinary landscape surrounding it—I knew it simply as the University of Manitoba’s Delta Marsh Field Station.

I went there as part of a U of M outdoor education course.

Outdoor survival.

It was a three-day experience in learning how to function outside in a Manitoba winter. Among other things, we built quinzees: snow shelters made by piling up snow, allowing it to settle and then hollowing out the inside.

And then we slept in them.

As luck would have it, Manitoba decided to make the exercise considerably more authentic.

A blizzard rolled over Delta Marsh that weekend.

By the time it was finished, snow had accumulated almost to the windows of our cars. It took us hours to dig the vehicles out.

Meanwhile, being university students presented with an enormous quantity of fresh snow, we discovered another recreational opportunity.

We went tobogganing.

Off the roofs of one of the supply buildings.

It was one of those ridiculous, wonderful weekends that gets filed somewhere in your memory and stays there for the rest of your life.

I didn’t know then that I was staying in a building with an extraordinary history.

I certainly didn’t know that the man who built it had apparently once refused to let Clark Gable stay there.

But I’m getting ahead of myself.

The Squire of Delta Marsh

Mallard Lodge was built in 1932 by Donald Henderson “Dan” Bain.

Calling Bain a successful Manitoba sportsman doesn’t quite do him justice.

He captained and played centre for the Winnipeg Victorias when they won the Stanley Cup in 1896 and 1901. He was later inducted into the Hockey Hall of Fame, Canada’s Sports Hall of Fame and the Manitoba Sports Hall of Fame.

And hockey was only one of his sports.

At various times, Bain was a Manitoba champion in roller skating, gymnastics, cycling and trapshooting. He played lacrosse. And in 1930, at 56 years old, he won a Canadian pairs figure-skating championship.

Apparently Dan Bain didn’t believe much in specialization.

He was also a very successful businessman. As president of Donald H. Bain Limited, a grocery brokerage with offices in several Canadian cities, he accumulated a substantial fortune.

His 4,500-square-foot Winnipeg home at 69 East Gate even contained a spring-mounted dance floor.

Dan Bain was quite a character.

And he loved Delta Marsh.

Over time, Bain assembled a huge private estate there—more than 3,000 acres, including a long stretch of Lake Manitoba shoreline. He was an avid waterfowl hunter, outdoorsman and dog lover, and Delta became his sanctuary.

In 1932, he built Mallard Lodge.

This was no primitive hunting shack.

It had electric lights supplied by its own generator years before rural electrification reached the area. It had running water, insulation for year-round occupation, a concrete basement and a large living room dominated by a magnificent fieldstone fireplace.

There is a wonderful story about its construction.

Engineers apparently told Bain that the sandy ground wouldn’t support the building.

So Bain fired them.

Then, with help from his nephews, he built the foundation himself.

Eight decades later, historians examining the building noted the absence of cracks in the basement floor.

Apparently Dan knew what he was doing.

The lodge became Bain’s private retreat. He spent long periods there during summers and autumns and occasionally Christmas.

But getting invited was another matter.

Bain guarded his privacy fiercely.

Which brings us to Clark Gable.

Clark Gable Couldn’t Stay Here

In 1938, Clark Gable came to Delta Marsh.

Yes.

That Clark Gable.

One of the biggest movie stars on Earth.

Delta Marsh had already become famous among prominent waterfowl hunters, and Gable came to Manitoba to hunt there.

Over the years, a story developed that Clark Gable had stayed at Mallard Lodge.

Apparently he didn’t.

According to Bain family members, Dan Bain refused him.

The Manitoba Historical Society’s history of Bain records the family’s recollection and notes that it was entirely consistent with Bain’s personality. He was fiercely protective of his privacy, fencing his property and putting up warning signs to discourage uninvited visitors.

Think about that for a moment.

Clark Gable could come to Delta Marsh.

He just couldn’t stay with Dan.

Nearly ninety years later, Bain may inadvertently have written one of the greatest hotel slogans imaginable:

Clark Gable couldn’t stay here.

You can.

But the celebrities, wealthy businessmen and hunters weren’t really the most important thing about Delta.

The real star was outside the lodge.

One of the World’s Important Wetlands

Delta Marsh stretches along the southern shore of Lake Manitoba.

It is a vast freshwater wetland of shallow bays, channels, wet meadows and emergent vegetation. Much of it is separated from Lake Manitoba by a narrow, wooded sand ridge.

And that peculiar geography matters enormously.

Today the Delta Marsh Important Bird Area encompasses roughly 308 square kilometres. It is also internationally recognized under the Convention on Wetlands.

Why?

Birds.

Lots and lots of birds.

For generations, Delta was famous primarily for its waterfowl—mallards, canvasbacks, redheads, lesser scaup, pintails, wigeon and many others.

Aerial surveys have detected concentrations of roughly 100,000 waterfowl at Delta.

But scientists eventually came to appreciate that something else extraordinary was happening there.

That narrow wooded ridge between Lake Manitoba and the marsh forms a natural migration corridor.

Birds travelling north and south encounter an enormous body of water and follow the ribbon of habitat along its edge.

Waterfowl stage in the marsh.

Shorebirds use the beaches and wetlands.

Migrating songbirds follow the ridge.

In May, warblers pour through on their journey north.

In autumn, the traffic reverses.

Some birds will eventually travel thousands of kilometres farther south.

Which is why calling Delta a “Manitoba wetland” doesn’t really capture its importance.

Birds don’t recognize borders.

A wetland in Manitoba can be one critical piece of an ecological system extending across a continent.

Think of Delta Marsh as a gigantic service station on an aerial highway.

Food.

Shelter.

Rest.

Then onward.

Remove enough service stations from a migration route and eventually the highway stops working.

That is why wetlands such as Delta matter far beyond the communities beside them.

And humans eventually began asking a question about all those ducks passing through Delta.

How do we make sure they keep coming?

Before the University: Delta Changes the Science of Ducks

There is another part of the Delta Marsh story that I hadn’t appreciated.

Long before the University of Manitoba established its Field Station at Mallard Lodge, another scientific institution appeared elsewhere on the marsh.

And it eventually became an international conservation organization.

Its name is Delta Waterfowl.

Its institutional roots stretch back to an American conservation organization founded in 1911. But the part of the story that concerns Delta Marsh begins with a wealthy American businessman named James Ford Bell.

Bell founded General Mills.

He was also an avid waterfowl hunter and particularly loved canvasbacks, which migrated through his home state of Minnesota.

Bell became worried about declining duck populations.

His first solution was wonderfully straightforward.

If his hunting camp shot ducks, he would raise ducks to replace them.

Bell established a hatchery on his property at the north end of Delta Marsh.

But eventually he began asking a much more consequential question.

Instead of simply producing more ducks, why not understand why duck populations rise and fall in the first place?

For help, Bell turned to one of the most important figures in the history of modern conservation.

Aldo Leopold.

Leopold was helping establish the emerging science of wildlife management. He would later write A Sand County Almanac and become one of the foundational thinkers in modern conservation.

Bell enlisted Leopold’s help in developing a scientific waterfowl research program.

Leopold sent a young graduate student named Hans Albert Hochbaum to Delta.

In 1938, Hochbaum arrived.

And something remarkable began.

The hatchery evolved into the Delta Waterfowl Research Station.

Research at Delta helped scientists understand fundamental aspects of wild-duck biology: breeding ecology, territorial behaviour, nesting, re-nesting, homing, predation and habitat.

Delta Waterfowl describes the station as eventually becoming an epicentre of waterfowl and wetland research in North America.

Students trained there went on to universities, wildlife agencies and conservation organizations throughout Canada and the United States.

Think about what that means.

This marsh outside Portage la Prairie wasn’t merely a spectacular place to shoot ducks.

Scientists working here were helping develop the modern science of how to conserve them.

And Delta Waterfowl didn’t disappear.

It evolved.

Today it is a binational conservation organization operating in Canada and the United States, working on waterfowl production, habitat, research and hunting across North America.

An organization now known thousands of kilometres from Manitoba still carries the name of the marsh where so much of its scientific identity was formed:

Delta.

And then something even more unusual happened.

Another field station arrived.

Two Field Stations. One Marsh.

In the 1960s, the University of Manitoba became interested in establishing its own permanent research and teaching facility at Delta Marsh.

The University knew about the Delta Waterfowl Research Station.

Indeed, Jennifer Shay’s history of the U of M Field Station makes clear that the two institutions were expected to collaborate.

But they were not the same field station.

Delta Waterfowl concentrated particularly on waterfowl.

The University envisioned something broader.

And it had inherited one extraordinary piece of infrastructure in which to do it.

Dan Bain’s old lodge.

After Bain died in 1962, the disposition of his estate became complicated.

In March 1964, a huge auction was held at Mallard Lodge. People who had spent years wondering what lay behind Bain’s fiercely protected gates finally got inside.

The lodge was essentially stripped to its walls.

The property subsequently passed through a private company and was resold weeks later to the Province of Manitoba.

Then came the University.

In 1966, University of Manitoba ecologist Dr. Jennifer Shay established the Delta Marsh Field Station around Mallard Lodge.

What the University inherited wasn’t a pristine research campus.

Shay recalled a collection of aging structures: Mallard Lodge, cottages, the barn that had housed Bain’s automobile collection, a generator house, old railway cars, a game-hanging shed and other buildings.

With limited money and considerable ingenuity, they turned it into a functioning scientific station.

And suddenly Delta Marsh had something extraordinary.

Two field stations.

One marsh.

There are relatively few wetlands anywhere with even one permanent biological field station.

Delta had two.

One grew out of the concerns of hunters and conservationists about declining waterfowl.

The other grew out of a university’s desire to put students and researchers directly into an ecosystem and ask how it worked.

Different institutions.

Different histories.

But ultimately variations on the same question:

How does this extraordinary place work—and how do we keep it alive?

Forty-Four Years of Science

The University Field Station became something remarkable.

For the next 44 years, students and scientists came to Delta Marsh.

Undergraduates learned there.

Graduate students conducted research there.

Scientists studied birds.

Plants.

Wetlands.

Water.

Migration.

Ecology.

And they produced an extraordinary body of work.

More than 130 graduate theses.

370 scientific publications.

Think about those numbers.

Hundreds of pieces of scientific research.

Generations of students.

Decades of observations.

All radiating outward from a lodge built in 1932 by an eccentric Stanley Cup champion who wanted somewhere private to hunt ducks.

The Field Station became particularly important to ornithology.

Researchers associated with Delta studied migration, breeding biology, bird populations and environmental change.

The station’s influence extended far beyond Manitoba.

And somewhere during those decades, I arrived.

My Three Days at Delta

I wasn’t there to conduct ornithological research.

I wasn’t writing a thesis.

I was trying to learn how not to freeze outside.

Our U of M course brought us to the Field Station for three days of outdoor education and survival training.

We built quinzees.

Then we crawled inside and slept in them.

There is something wonderfully clarifying about lying inside a hollowed-out pile of snow, knowing that the temperature outside is very much a Manitoba temperature.

Then came the blizzard.

The snow piled up.

And piled up.

Our vehicles slowly disappeared until the snow was almost to their windows.

Eventually we had to dig them out.

It took hours.

But before that—or perhaps because we were young enough not to regard the situation with appropriate seriousness—we discovered that the accumulating snow had created excellent conditions for another activity.

Tobogganing.

We climbed onto one of the supply buildings.

And went tobogganing off the roof.

Repeatedly.

I remember it as an absolutely amazing weekend.

And then life moved on.

As it does.

I left university.

The years passed.

The Field Station continued.

Students came after me.

Researchers came after me.

Theses were written.

Papers were published.

Birds migrated north.

Birds migrated south.

And Mallard Lodge remained there beside the marsh.

Until 2011.

What Really Happened to the Field Station?

The simple version of this story goes like this:

Lake Manitoba flooded catastrophically in 2011.

The Delta Marsh Field Station was damaged.

The University closed it.

End of story.

Except that isn’t quite the story.

The flooding was unquestionably serious. Lake Manitoba experienced extraordinarily high water. The Field Station suffered damage. The University of Manitoba permanently closed the facility.

Most of the surrounding Field Station buildings were subsequently demolished or removed.

Mallard Lodge survived.

But scientists associated with the station strongly disputed any suggestion that Delta was simply an obsolete facility whose time had come.

In 2012, four scientists with deep connections to the station—Keith Hobson, D. Ryan Norris, Gordon Goldsborough and Spencer Sealy—published an editorial with a heartbreaking title:

Requiem for a Field Station: The Loss of a Canadian Ornithological Treasure

They looked at the numbers.

And the numbers told an important story.

Scientific output hadn’t collapsed.

Over the Field Station’s 44-year history, researchers had produced 370 publications—an average of about 8.4 per year.

Thesis production had generally increased.

And research and teaching use during the 1990s and 2000s had been among the highest periods in the station’s history.

There was no evidence of some decades-long decline into irrelevance.

The Canadian Field Stations organization subsequently made the point even more explicitly: Delta’s demise could not reasonably be attributed solely to flood damage or declining use.

They argued that something larger was happening—the chronic difficulty of securing institutional and governmental support for biological field stations.

That changes the story.

Because perhaps Mallard Lodge wasn’t simply lost to nature.

Perhaps we also chose not to save what had been built there.

A facility responsible for more than 130 graduate theses and 370 scientific publications disappeared.

Forty-four years of teaching and research ended.

The students stopped coming.

The researchers stopped coming.

The doors closed.

And the marsh continued without them.

Then, in 2020, I Found It Again

Almost a decade later, Lydia and I were exploring the area.

And we came across the old Field Station.

I recognized it.

Mallard Lodge.

Except this wasn’t the place I remembered.

It was abandoned.

Boarded up.

Silent.

The surrounding buildings where students and researchers once worked were mostly gone.

I stood looking at this place where I’d had that ridiculous, unforgettable weekend decades earlier.

Where we’d built quinzees.

Where we’d slept outside.

Where we’d survived our Manitoba blizzard.

Where we’d dug our vehicles out.

Where we’d gone tobogganing off a roof.

And now nobody was there.

That bothered me.

It still does.

Because an abandoned building isn’t always just an abandoned building.

Sometimes it represents accumulated human memory.

Someone designed this place.

Someone built it.

Dan Bain sat beside that fieldstone fireplace.

His dogs wandered through those rooms.

His friends came through those doors.

Clark Gable apparently didn’t.

Decades later, Jennifer Shay and others somehow turned an old hunting estate into a functioning university research station.

Scientists discussed their discoveries there.

Students arrived knowing almost nothing about wetlands and left knowing considerably more.

More than 130 graduate students produced theses connected to the place.

Hundreds of scientific papers came out of it.

Thousands of students, researchers and visitors must have passed through.

Somebody met their future spouse there.

Somebody decided to become a scientist there.

Somebody saw their first canvasback there.

Somebody stood outside on a cold autumn morning and watched thousands of birds lift off the marsh.

And at least one group of university students went tobogganing off the roof during a blizzard.

I know.

I was one of them.

So Who Owns It?

That question has been bothering me since I started looking into this again.

Who owns Mallard Lodge now?

And the answer turns out to be considerably more complicated than I expected.

For decades, the story seemed simple enough.

The Province of Manitoba acquired the old Bain property after his death, and the University of Manitoba operated the Delta Marsh Field Station there beginning in 1966.

But then I found a Manitoba government document from June 21, 2006.

Its title leaves very little room for ambiguity:

Delta Marsh Field Station Building Site Transferred to University of Manitoba

The Province announced that it was formally transferring ownership of the Delta Marsh Field Station to the University of Manitoba.

The announcement quotes Manitoba Conservation Minister Stan Struthers saying:

“We are also pleased to transfer ownership of the Delta Marsh Field Station to the University of Manitoba so that outstanding research work may continue.”

And Gordon Goldsborough, then director of the Field Station, thanked the Province for what he called its:

“generous gift of the land and buildings at our site.”

The same government announcement separately renewed the University’s lease on approximately 700 hectares of surrounding Crown land for another 21 years.

That distinction matters.

The surrounding Crown land was leased.

The Field Station site and its buildings were transferred.

So by 2006, the University wasn’t simply occupying Mallard Lodge under a provincial lease.

It owned the core property.

Then came 2011.

The Field Station closed.

Most of the surrounding buildings were later demolished or removed.

Mallard Lodge remained.

And then the story becomes murky again.

A 2016 report in the Manitoba Co-operator described Mallard Lodge and one other surviving Field Station building as being owned by the Province of Manitoba.

Which raises an obvious question.

If Manitoba gave the Field Station property to the University in 2006...

How did Manitoba apparently come to own it again?

Was the property transferred back after the Field Station closed?

Was there a condition attached to the original gift?

A reversion clause?

A negotiated conveyance?

An Order-in-Council?

A deed sitting somewhere in Manitoba’s Land Titles system that explains the whole thing?

I don’t know yet.

But now there is clearly a transaction—or legal mechanism—missing from the public story.

And that makes the question considerably more interesting than simply asking who owns an abandoned lodge.

There is a paper trail here.

I’m going to find it.

But while researching the ownership, another question occurred to me.

A much bigger one.

Why should this place remain abandoned?

Look at what it already has.

A lodge built in 1932.

A remarkable Manitoba character who built it.

Stanley Cups.

Clark Gable.

A magnificent fieldstone fireplace.

One of the most important wetlands on the Canadian Prairies outside its door.

A continental bird-migration corridor.

The birthplace of enormously influential waterfowl research nearby.

A conservation organization that still carries Delta’s name across North America.

Forty-four years of University of Manitoba history.

More than 130 graduate theses.

370 scientific papers.

And generations of former students with memories of the place.

You couldn’t manufacture that history if you spent $100 million trying.

It’s already there.

Imagine Mallard Lodge Alive Again

Not demolished.

Not replaced by condominiums.

Not converted into some generic luxury resort that could be anywhere.

Restored.

Keep the fieldstone fireplace.

Keep the log walls.

Find the old photographs.

Put the history back on the walls.

Tell people who Dan Bain was.

Tell them about James Ford Bell.

Aldo Leopold.

Hans Hochbaum.

Jennifer Shay.

Tell them what scientists discovered here.

Tell them about the migration happening outside their windows.

And perhaps Mallard Lodge doesn’t have to choose between being a lodge and being a place of learning.

Bring university field courses back occasionally.

Invite researchers.

Partner with conservation organizations.

Bring naturalists.

Birders.

Photographers.

Families.

School groups.

People who simply want to spend several days beside one of Canada’s great wetlands.

Imagine waking before sunrise, walking outside with a cup of tea and hearing tens of thousands of migrating birds out on the marsh.

Imagine sitting around Dan Bain’s fieldstone fireplace that evening.

Maybe somebody else builds a quinzee.

Maybe somebody else experiences a Manitoba blizzard.

Perhaps we discourage the tobogganing off the roof this time.

Probably an insurance issue.

Mallard Lodge has already had three lives.

Dan Bain’s private wilderness retreat.

The University of Manitoba’s scientific field station.

An abandoned building.

Maybe it’s time for a fourth.

A place where people can once again come to experience Delta Marsh.

A place that preserves its past rather than embalming it.

A place that introduces another generation to the marsh rather than leaving them to look at photographs of what used to be there.

There are old buildings whose useful lives have ended.

I’m becoming increasingly convinced this isn’t one of them.

In fact, the more I learn about Mallard Lodge, the more extraordinary it becomes.

And the more absurd it seems that it should simply sit there.

Boarded up.

Waiting.

This place deserves another chapter.

So I’m going to find out who owns it.

Then I’m going to find out what condition it’s really in.

And then I’m going to ask the slightly crazy question that started forming in my head when I began this investigation:

What would it actually take to bring Mallard Lodge back?

Next: Who Owns Mallard Lodge?

And what would it take to save it?

Regards,

Luke Kandia
Chasing Clarity

Luke Kandia writes Chasing Clarity, a Substack covering healthcare systems, financial markets, public policy, history and science communication — usually by starting with something small, like an old coin, a data point, a photograph or a footnote, and following it wherever it actually leads.

Nothing here is investment, financial, or legal advice — just a curious fellow who is currently wondering who owns Dan Bain’s old Mallard Lodge. Do your own homework before you make decisions with your money.

Sources & Further Reading

Manitoba Historical Society — Dan Bain: The Squire of Delta Marsh
The wonderful detailed history of Bain, Mallard Lodge, his Delta estate, the Clark Gable story and the disposition of his estate.

Manitoba Historical Society — Donald Henderson “Dan” Bain (1874–1962)
Biographical background on Bain’s remarkable sporting and business career.

Manitoba Historical Society — Mallard Lodge / Delta Marsh Field Station
History, photographs and location of the surviving lodge and former Field Station.

Jennifer M. Shay — Reflections on a Marsh: The Delta Marsh Field Station’s First 20 Years, 1966–1986
Shay’s first-person history of establishing the University of Manitoba Field Station and its relationship with the Delta Waterfowl Research Station.

Hobson, Norris, Goldsborough & Sealy — Requiem for a Field Station: The Loss of a Canadian Ornithological Treasure
Avian Conservation and Ecology, 2012. A contemporary examination of the closure and the Field Station’s scientific output.

Canadian Field Stations — Tribute to the Delta Marsh Field Station
History and assessment of the Field Station’s 44 years of research and teaching.

Delta Waterfowl — History
History of James Ford Bell, Aldo Leopold, Hans Hochbaum and the development of the Delta Waterfowl Research Station.

Manitoba Important Bird Areas Program — Delta Marsh
Current information about the marsh’s bird populations, habitat and importance as a migratory corridor.

CBC News — U of Manitoba Shuts Down Delta Marsh Field Station
Contemporary reporting on the University’s 2011 decision to close the station.

Government of Manitoba — June 21, 2006
Delta Marsh Field Station Building Site Transferred to University of Manitoba
This contemporaneous provincial announcement confirms that Manitoba transferred ownership of the Field Station site and buildings to the University of Manitoba while separately renewing the University’s lease on approximately 700 hectares of surrounding Crown land.

Parts 1 and 2 covered two kinds of mechanical selling that have nothing to do with anyone panicking: Part 1: Margin calls, and the daily rebalancing that Part 2: Leveraged ETFs are contractually required to do regardless of what anyone believes about the market.

There’s a third one. It works differently from the other two, and it’s the hardest to pin down.

A quick grounding in options

An option is a contract giving the buyer the right, not the obligation, to buy (a call) or sell (a put) a stock at an agreed strike price, by an agreed date. One standard equity option contract covers 100 shares. The price of the contract itself is the premium.

Buyers of options have capped, known risk — the most they can ever lose is the premium. Unhedged sellers can face a very different risk. A dealer who sells a call without owning the stock is on the hook to deliver shares at the strike no matter how high the stock climbs — there’s no ceiling. A dealer who sells a put is on the hook to buy shares at the strike even if the stock is nearly worthless — bounded at zero, but that can still mean losing almost everything relative to a small premium collected.

That asymmetry is why dealers hedge. They’re not in the business of betting on direction — they’re in the business of being the counterparty, all day, to whoever wants an option, and earning the bid-ask spread for providing that liquidity. Every option they end up holding, long or short, is directional risk they didn’t choose. So they offset it by trading the underlying stock.

The number that tells them how much stock to hold is delta — roughly, how much the option’s value moves for every $1 the stock moves. A delta of 0.40 behaves like 40 shares of exposure; sell that call, and the dealer buys about 40 shares to stay neutral. As the stock moves, delta changes — that rate of change is gamma — and the dealer has to keep rebalancing the hedge to keep up.

(For the full walkthrough — why people buy options in the first place, worked leverage examples, exactly how delta and gamma behave step by step, and why dealers can lose money on trades their own hedging is designed to manage — see the companion piece, Understanding Options: A Primer.)

Which direction the hedge runs

This is where it gets interesting, because the direction a dealer’s hedge runs depends entirely on which side of their own book they’re on.

Short gamma is fundamentally the dealer offsetting the P&L exposure created by a contract it actually sold. Having sold the option, the dealer loses money as the stock moves toward — and through — the strike, the same way an owner of the opposite position would. So the hedge chases that exposure in the direction that matches it: buying more stock as the price rises toward a call they’ve sold, or selling more stock as the price falls toward a put they’ve sold.

A dealer who sold a call and watches the stock rise must keep buying more stock into that rising market, to cover a growing upside liability as delta climbs. That’s not a coincidence. Overall, their aggregate gamma exposure — the combined effect of every option on their book, added up — is negative. A dealer with negative net gamma is short gamma, and short-gamma hedging tends to run with the market: stock falls, they sell; stock rises, they buy.

That’s procyclical — it can add fuel to whichever direction the market’s already moving. It’s also common whenever there’s heavy customer demand to buy options, like calls in a hot stock or protective puts ahead of an earnings report — someone has to sell those contracts to the customers who want them, and that someone usually ends up short gamma.

Worth being precise about what actually determines the sign here, because it isn’t simply counting how many contracts a dealer sold versus bought. Gamma isn’t evenly spread across an options book — it concentrates heavily near the money and can become enormous there as expiration approaches, so a handful of near-the-money contracts about to expire can carry more gamma than a much larger pile of far-out-of-the-money or long-dated ones.

What matters is the sum of gamma across every position, weighted by how much each one actually contributes — not the raw count of contracts on each side.

The opposite case is a dealer whose aggregate gamma exposure nets out positive — say, a desk that’s accumulated a large book of options through market-making over time, or one deliberately holding options as a position of its own. That dealer is long gamma, and long-gamma hedging tends to run against the market: stock falls, they buy; stock rises, they sell. That’s a stabilizing flow — it can lean against the move.

Put the two cases side by side and the difference is really one thing: whether the dealer’s aggregate gamma exposure nets out negative or positive. Negative means short gamma, hedging that tends to run with the market. Positive means long gamma, hedging that tends to run against it.

And that sign isn’t fixed for the day just because nobody’s traded another contract. Spot can move through — or sharply away from — a strike, expiration can arrive, and the gamma concentrated at one part of the book can disappear or reshuffle surprisingly quickly — the composition of what’s driving the book’s net exposure keeps shifting even when the dealer hasn’t touched their options position at all.

Which is exactly how a dealer can be an accelerant in the morning and a brake by the afternoon, without anyone on the desk deciding to change anything.

None of this is a person watching a screen

At a modern electronic market-making desk, this isn’t a trader glancing at delta once an hour and deciding to place a trade. Risk systems recalculate delta and gamma on every trade and on essentially every meaningful tick of the underlying — sub-second, not hourly — and an automated hedging layer can fire the rebalancing trade on its own, the moment net delta drifts past a threshold a human set in advance.

Not a fixed schedule. A trigger.

One large market maker’s own SEC filing describes exactly this — a system that assimilates market data and reevaluates quotes many times per second, automatically rebalancing positions throughout the trading day, with humans monitoring risk parameters rather than manually placing each trade. (Interactive Brokers Group — 2025 Form 10-K)

Humans set the rules — how tight to keep the band, how aggressively to chase the market versus wait for a better fill, when to override the system entirely in a genuine crisis. But the moment-to-moment reality is a machine acting on its own, continuously, the instant its threshold is crossed.

Which is worth sitting with for a second, because it’s the same shape as everything else in this series. The margin cliff in Part 1 wasn’t a person deciding to hurt anyone — it was a formula that fires the moment a threshold is crossed, with a human standing nearby who could technically intervene but usually doesn’t need to. This is that, again, in a completely different department, for a completely different reason.

A dealer’s hedging book is a cliff too — automated, threshold-triggered, running without waiting for permission. It just happens to be a cliff that can lean against the market or potentially add to it, depending on which side of gamma the dealer happens to be sitting on that day.

Is this actually dangerous?

Cboe — the exchange operator that created listed options trading in the U.S. and remains the exclusive home of SPX options — describes this distinction directly in its analysis of SPX 0DTE market-maker hedging. SPX options are options on the S&P 500 index itself, rather than on any single stock — one of the most heavily traded option products in the world. 0DTE stands for “zero days to expiration”: a contract expiring that very trading day, regardless of when it was originally opened — it might have been listed weeks earlier and simply arrived at its final day. Whatever gamma exposure it creates becomes an intraday risk-management problem rather than one that can play out over weeks — a dealer can choose to warehouse some of that risk rather than trade it away instantly, but the window to make that decision keeps shrinking as the clock runs down.

Worth pausing on one difference from everything covered so far. Every example in this piece has used a single-stock option — the kind where, if exercised, the dealer actually delivers or buys 100 real shares. SPX options don’t work that way. They’re cash-settled: instead of handing over shares of “the S&P 500,” which isn’t a single tradable thing anyway, the contract settles in cash based on the difference between the settlement price and the strike. There’s no delivery to scramble for. What a dealer hedges with instead is S&P 500 futures, or an ETF like SPY that tracks the index — not the index itself, which can’t be bought or sold directly. The delta and gamma math works the same way either way; what changes is only the instrument the dealer actually trades to stay neutral.

The other thing worth naming: near the strike, 0DTE gamma isn’t just “the same gamma, compressed into a shorter window.” It can be structurally much larger. Gamma for an option sitting near the strike price grows sharply as expiration approaches — the closer to the deadline, the more violently delta can swing on even a small stock move, because there’s almost no time left for that swing to reverse itself. A three-month option’s delta might drift gradually over weeks. A same-day option’s delta can lurch from 0.20 to 0.80 in an afternoon on a move that would barely register for a longer-dated contract. That’s the actual mechanical reason 0DTE hedging draws so much attention — not merely that it happens fast, but that the hedging itself gets more forceful the closer the clock runs to zero.

Because Cboe runs the exchange itself, their research draws on real trading and positioning data crossing their own systems, not outside estimates. (Cboe — Evaluating the Market Impact of SPX 0DTE Options)

Cboe’s work is also an important antidote to sensational claims about 0DTE options, which get a lot of headline attention for supposedly whipping markets around. Gross option notional — roughly, the underlying value represented by all those contracts, before adjusting for delta, direction, or offsetting positions — is not the same thing as net dealer hedging demand. Financial headlines love quoting trillions in 0DTE “notional volume,” but that raw figure doesn’t tell you the dealer’s net Greek exposure, particularly net delta and gamma, and therefore doesn’t tell you how much underlying hedging may actually follow. A dealer who’s long gamma on one strike and short gamma on another isn’t adding those two positions’ notional together; the exposures partly cancel. Cboe found customer flows often surprisingly balanced and market-maker net gamma relatively small compared with S&P futures liquidity. (Cboe — Evaluating the Market Impact of SPX 0DTE Options) A later Cboe analysis estimated net 0DTE market-maker gamma hedging at, at most, roughly 0.2% of SPX daily liquidity during the period it examined. (Cboe — 0DTEs Decoded)

That’s a striking gap. Across the full 2025 trading year, Cboe reported that 0DTE contracts averaged 2.3 million contracts a day and made up 59% of average daily SPX options volume — well over half. (Cboe — The State of the Options Industry: 2025) High trading volume and low net dealer exposure aren’t a contradiction, though — they’re measuring different things. Volume counts total activity: how many contracts changed hands. Net gamma measures what remains after those long and short exposures within that slice of the dealer’s SPX 0DTE book offset each other — though Cboe itself notes dealers can also carry offsetting positions elsewhere, in other expirations or related products, that this kind of netting doesn’t capture.

Thousands of customers are buying and selling 0DTE options for thousands of unrelated reasons on any given day — some betting the market rises, some betting it falls, some buying insurance, some running spread trades that are close to gamma-neutral on their own. None of that is coordinated. But across enough volume and enough diversity of intent, a lot of it points in offsetting directions, and what survives that netting is exposure that may still require hedging, subject to whatever other offsets exist elsewhere in the dealer’s book. High turnover, low leftover risk.

Worth flagging: that “balanced” finding describes the period Cboe examined, not a law of how options markets always behave. A single-sided event — demand overwhelmingly tilted one direction, all at once — could in principle produce a much larger net exposure on any given day. That’s an empirical finding, not a guarantee.

What a single-sided event could look like

Worth spelling out, rather than leaving as an abstraction, because the mechanism is real and well-understood, even if it doesn’t fire the exact same way on every single bad day.

On an ordinary day, customer demand for options is genuinely diverse — some people betting the market rises, some betting it falls, some running spread trades that barely create net exposure at all. That diversity is what lets it net down to something small. It’s also the thing most likely to break down during a genuine, broad selloff.

Suppose a selloff produces sufficiently one-sided demand for downside protection. Portfolio managers who’d otherwise sit on their hands start buying puts to hedge positions they don’t want to sell outright. Retail accounts that never think about options suddenly want them too.

If that demand is large enough and not offset elsewhere in the market, it stops looking like diversified, offsetting flow and starts looking like one-directional demand, overwhelmingly for puts, arriving at once.

Someone has to write those puts. That’s the dealer’s job, the same as on any other day — be the counterparty to whoever needs one. If a flood of demand like that doesn’t net down the way ordinary daily flow does, it can leave dealers heavily net short puts — not necessarily as a rounding error in an otherwise balanced book, but as a dominant feature of it.

Worth being honest about how conditional that “if” is, though. It isn’t automatic. Dealers’ starting inventory matters — a desk that’s already long gamma elsewhere from unrelated positioning doesn’t flip the same way. Customers sell puts as well as buy them, even during selloffs. Institutional and retail flow can move on different timelines entirely; one study of the COVID crash found that institutions bought crash protection in the run-up to the decline, while a meaningful share of retail buying arrived later, closer to the recovery than the crash itself — a genuinely messier picture than “everyone panics into puts at once.” (Jackwerth — The Review of Asset Pricing Studies, “What Do Index Options Teach Us About COVID-19?”) And short puts, per the mechanics already covered, mean short gamma: as the stock keeps falling, a dealer who does end up in that position has to keep selling stock to stay hedged. That’s the exact procyclical mechanism from earlier in this piece — a real, mechanically sound possibility, not a guaranteed feature of every crash. This is the machine, not a claim that the machine runs the same way every single time.

There’s a separate, permanent fingerprint the options market carries that’s worth knowing about, even though it doesn’t prove this specific mechanism fires on every crash day. Index puts have traded at persistently richer implied volatility than equidistant calls ever since October 1987 — a structural feature of the market, not a passing quirk. Cboe’s own research ties it to the same underlying asymmetry: markets tend to fall faster than they rise, so downside insurance has stayed permanently, structurally more expensive than upside speculation, every day since the crash that created that pricing. (Cboe — Inside Volatility Trading: The Adventures of Volatility Markets) That’s real evidence of persistent, structural demand for downside protection. It’s not, on its own, evidence that dealers must go heavily short gamma during any particular selloff — those are two different claims, and it’s worth keeping them apart.

It’s also worth naming plainly: Cboe runs the exchange these contracts trade on. That doesn’t make their data wrong — the figures check out against their own published numbers — but it does mean Cboe has a commercial interest in 0DTE options looking safe, and their research shouldn’t be the only word on the subject.

Independent academics don’t fully agree with each other, either. One team of academic and central-bank researchers found that 0DTE options tend to dampen market volatility, and traced the mechanism specifically to dealer hedging: intraday shifts in market-maker hedging needs predicted stronger order-flow reversals and lower momentum, largely from longer-dated positions that had simply aged into their expiration day, not from options opened and closed same-day. (Adams, Dim, Eraker, Fontaine, Ornthanalai, and Vilkov — “Do S&P500 Options Increase Market Volatility? Evidence from 0DTEs”)

A separate team at the University of Utah found the opposite. Using the staggered rollout of weekly index options as a natural experiment, they estimated that a one-standard-deviation increase in 0DTE trading raised volatility by roughly 9% relative to its average level — and that effect held up even after accounting for dealer gamma hedging, which pointed to speculative retail trading, not dealer mechanics, as the driver. (Brogaard, Han, and Won — “Does 0DTE Options Trading Increase Volatility?”)

Both are serious academic working papers from university researchers, not blog posts — and they land on opposite conclusions studying the same broad question. Neither disagreement overturns what Cboe found about net dealer gamma being small during the periods it measured. It just means the honest picture is bigger than one exchange’s own research settling the question, and that the disagreement itself seems to run along a fault line this piece has been tracing the whole way through: dealer hedging mechanics on one side, speculative flow on the other — which is exactly the kind of distinction gross volume numbers can’t tell you.

One more scope note. Delta and gamma aren’t the whole picture near expiration — dealers also watch how a hedge shifts with changing volatility and with the pure passage of time, effects with their own names (vanna and charm, respectively). Expiration itself brings a separate risk too: if the underlying finishes at or very close to a strike, nobody on the short side knows for certain whether they’ll be assigned until after the fact, which can leave a dealer unexpectedly long or short stock overnight — a distinct hazard called pin risk. Real, and part of why 0DTE desks stay busy — but beyond what this piece covers.

So options dealers don’t automatically belong on the sell side of our equation, the way forced margin selling almost always does. Sometimes they’re an accelerant. Sometimes they’re a brake. Which one depends on a sign most of us never see, sitting on a risk desk we never think about, changing quietly while the rest of the market just sees a price.

What happens when the market just stops

Everything so far assumes the market is open. Worth asking, briefly, what happens to each actor in this series when it isn’t.

The U.S. market has a market-wide circuit breaker: if the S&P 500 falls 7% intraday — a Level 1 trigger, the lowest of three thresholds (7%, 13%, and 20%) — trading halts for 15 minutes, across the board, stocks and their options together.

That halt only applies between 9:30am and 3:25pm; a decline of the same size after 3:25pm doesn't trigger it at all, on the theory that a 15-minute pause that close to the closing bell wouldn't accomplish much.

Level 1 has only been triggered four times since the current system began in 2013. All four were in March 2020, inside a single month. (Level 2 and Level 3 have never been triggered under the current system — this piece is only ever talking about the smallest of the three.)

For an ordinary investor holding stock outright, the halt itself creates almost no new mechanical obligation — they simply can’t trade for 15 minutes. Unlike the margin borrower, leveraged ETF, or hedging dealer, nobody forces them to alter the position when trading resumes. They remain exposed to whatever price comes next, for better or worse.

A leveraged ETF has a harder problem. Its whole mechanism depends on rebalancing to a fixed daily target — and that rebalancing typically happens near the close, using the very market that’s now frozen. A halt doesn’t cancel the day’s move. It simply removes the fund’s ability to adjust its exposure while trading is suspended; when trading resumes, it has to work with whatever prices and liquidity are then available.

An options dealer faces a genuinely hard problem here, distinct from the other two — especially since, per the last section, this dealer might not be walking into the halt with an ordinary, diversified book.

If the halt is happening because of a genuine, broad selloff — the kind that gives one-sided put demand time to build — the dealer plausibly enters the halt already the heavily short-gamma dealer described a moment ago, selling stock into the decline right up until trading stops. That’s not a mechanical law, though. A halt can also be triggered by a sudden, fast-moving shock — overnight news, a single jarring headline — that hits before the slower, crowd-driven rush into puts has had time to build.

Either way, once the halt hits, the mechanics are the same: delta hedging assumes continuous access to the market — the ability to keep adjusting the stock position as the price moves. A halt breaks that assumption completely, and it’s not just the stock that freezes: options on that name typically halt at the same moment. So a dealer’s hedge is frozen at whatever size matched the last price before the halt, while the underlying risk it was hedging hasn’t gone anywhere — it’s just become untradeable.

In March 2020 specifically, the usual backup channel became constrained too: S&P futures carried a 5% overnight hard limit at the time, and they hit that limit repeatedly overnight and pre-market that month, severely constraining the instrument dealers would normally lean on when the primary market was closed. (A post-event industry task force specifically examined that limit-down experience, and CME subsequently widened it to 7%, effective October 2020. (NYSE — Report of the Market-Wide Circuit Breaker Working Group))

Then trading reopens — often via an auction, not a clean resumption at the old price. If the reopening print gaps meaningfully from wherever the stock was when it halted, a dealer whose hedge has become badly mismatched may need to rebalance aggressively into the reopening market, and gamma means that mismatch can be large, not small.

Worth noting: a single stock reopens through one exchange’s cross-auction, producing one clean print. An index like the S&P 500 doesn’t work that way — it’s a calculated composite, not a single tradable thing, and it’s disseminated using the latest available prices while its hundreds of underlying constituents are still working through their own individual reopening auctions, each at its own pace. A dealer hedging SPX exposure with futures or SPY doesn’t get one clean number to react to. They get a level that’s still resolving, messily, while pieces of the index are still catching up.

Either way, rebalancing into a reopen that can be thin and volatile is exactly the moment liquidity can be weakest — and other dealers with similarly displaced hedges may be trying to do the same thing at the same time, though not every dealer entered the halt with the same gamma sign, the same starting delta, or the same urgency to act.

Three actors, one event, three different kinds of trouble: the stockholder faces no forced action at all.

A halt can delay or impair the leveraged fund’s rebalance — if the market reopens with enough time and liquidity left in the day, the fund can often still reset its exposure; if trading stays thin or the instruments it relies on stay unavailable, it may finish the day away from its target instead.

The options dealer, especially one who entered the halt already leaning heavily one way, faces a different problem: frozen mid-hedge, with a badly displaced position that may require aggressive trading the moment the market reopens, often into exactly the kind of thin, unsettled liquidity that makes that trading most expensive.

Which raises the real question.

Margin calls have a directional bias — forced selling is the default outcome, unless someone steps in with cash first.

Leveraged ETF rebalancing has a mathematically determined direction — a formula, not a judgment call, decides which way it goes each day.

Options-dealer hedging is state-dependent — its direction isn’t fixed by a bias or a formula, but by a position nobody outside the desk can actually see, one that can flip from stabilizing to destabilizing without a single new trade.

A bias, a formula, and a state-dependent variable, each running on its own clock, for entirely different reasons. What happens when all three point the same way at once?

That’s Part 4.

Regards,
Luke Kandia
Chasing Clarity

Luke Kandia writes Chasing Clarity, a Substack covering healthcare systems, financial markets, public policy, and science communication — usually by starting with something small, like an old coin, a data point, or a footnote, and following it wherever it actually leads.

Nothing here is investment, financial, or legal advice — just a curious fellow who is currently wondering how options work. Do your own homework before you make decisions with your money.

Sources and references

1. Cboe — “Evaluating the Market Impact of SPX 0DTE Options.” Primary source for delta/gamma mechanics, long- versus short-gamma hedging, the importance of net rather than gross option flow and empirical estimates of SPX 0DTE dealer gamma relative to futures liquidity. Cboe — Evaluating the Market Impact of SPX 0DTE Options

2. Cboe — “0DTEs Decoded: Positioning, Trends, and Market Impact.” Additional empirical evidence that net 0DTE market-maker gamma hedging was small relative to SPX daily liquidity during the period studied. Cboe — 0DTEs Decoded

3. Cboe — “The State of the Options Industry: 2025.” Source for the 59% figure showing 0DTE contracts’ share of average daily SPX options volume across the full 2025 trading year. Cboe — The State of the Options Industry: 2025

4. Adams, Greg; Dim, Chukwuma; Eraker, Bjorn; Fontaine, Jean-Sebastien; Ornthanalai, Chayawat; Vilkov, Grigory — “Do S&P500 Options Increase Market Volatility? Evidence from 0DTEs.” Working paper finding 0DTE options dampen volatility via dealer hedging mechanics. SSRN — Do S&P500 Options Increase Market Volatility?

5. Brogaard, Jonathan; Han, Jaehee; Won, Peter Y. — “Does 0DTE Options Trading Increase Volatility?” Working paper (University of Utah, David Eccles School of Business), finding 0DTE trading increases volatility, attributed primarily to speculative retail activity rather than dealer hedging. SSRN — Does 0DTE Options Trading Increase Volatility?

6. Cboe — “Inside Volatility Trading: The Adventures of Volatility Markets.” Source for the persistent volatility skew in equity index options — OTM puts trading richer than equidistant OTM calls — a structural feature of the market dating to the October 1987 crash. Cboe — Inside Volatility Trading: The Adventures of Volatility Markets

7. Jackwerth, Jens — “What Do Index Options Teach Us About COVID-19?” The Review of Asset Pricing Studies, Vol. 10, No. 4, 2020. Source for the more complicated real-world timing of institutional versus retail crash-protection buying during the COVID selloff, complicating any simple “everyone buys puts at once” narrative. Oxford Academic — What Do Index Options Teach Us About COVID-19?

8. Interactive Brokers Group, Inc. — 2025 Form 10-K. Primary evidence, from a market-making firm’s own SEC filing, for automated market-making systems that reevaluate quotes many times per second and rebalance risk automatically throughout the trading day. SEC — Interactive Brokers Group 2025 Form 10-K

9. NYSE — Market-Wide Circuit Breakers FAQ. Source for current Level 1/2/3 thresholds, the 3:25pm cutoff, and halt duration rules. NYSE — Market-Wide Circuit Breakers FAQ

10. CME Group — S&P 500 Price Limits FAQ. Source confirming the E-mini S&P 500 futures overnight price limit was widened from 5% to 7%, effective October 12, 2020, partly in response to the March 2020 experience. CME Group — S&P 500 Price Limits FAQ

11. Cboe — S&P 500 Index Options Product Specifications. Source confirming SPX options are cash-settled and European-style, with no physical delivery of shares. Cboe — S&P 500 Index Options

12. ProShares — Geared (Leveraged & Inverse) ETFs FAQ. Source noting that while geared funds generally rebalance at the end of each trading day, the timing and manner can vary, and circumstances can prevent a fund from rebalancing all or part of its portfolio. ProShares — Geared FAQs

13. ProShares Trust — Statutory Prospectus (e.g., ProShares Ultra CRCL, August 2025). Standard ProShares risk-disclosure language, used across its geared fund family, explicitly stating that if a fund is unable to rebalance all or part of its portfolio — including due to a trading halt — its exposure may end up significantly above or below its stated multiple. ProShares — Statutory Prospectus

14. NYSE — “Report of the Market-Wide Circuit Breaker Working Group.” Source for the post-March-2020 review that examined the relationship between the 5% overnight E-mini S&P 500 futures limit and the cash-market circuit breaker, ahead of CME’s subsequent change to a 7% limit. NYSE — Report of the Market-Wide Circuit Breaker Working Group

Here is a confession: options are not actually that complicated. What’s complicated is that the finance industry has spent fifty years explaining them using sentences that sound like they were translated from German by someone who’d never met a human being. This piece is an attempt to undo some of that damage.

It exists as a companion to a series called The Cliff Is Moving, which is asking a fairly serious question — what actually causes markets to crash — and one of the answers turns out to run through options dealers. But you don’t need to care about market crashes to read this. You just need to be mildly curious about what happens when someone buys the “right” to purchase a stock they don’t own yet, at a price it isn’t currently trading at, and why the person selling them that right can, in the wrong moment, start behaving like a shopper who keeps grabbing more off the rack because the price is jumping, with a line stacking up behind her, only to realize the register is ringing up a bigger bill every time she reaches for the next item.

Let’s start at the beginning.

What an option actually is

An option is a contract. It gives whoever buys it the right, but not the obligation, to buy or sell a stock at an agreed price by an agreed date — right, not obligation, is the single most important four words in this entire subject, and about ninety percent of options confusion comes from people quietly forgetting them. That agreed date has a name too — it’s called expiration, and it’s exactly what it sounds like: the day the contract simply stops existing, whether anyone’s used it or not.

One standard equity option contract covers 100 shares of the underlying stock — nobody ever tells you why it’s 100 and not, say, 47, and the honest answer is “because that’s the number everyone agreed on a long time ago,” which is also the honest answer to a great many things in finance.

The agreed price has a name: the strike price, or just “the strike.” A call option is the right to buy at the strike. A put option is the right to sell at the strike. If you need a way to keep these straight: a call lets you call the stock over to your side, the way you’d call a dog. A put lets you put the stock onto someone else, the way you’d hand off a hot dish you don’t want to be holding anymore. It’s not a perfect mnemonic. It’s better than nothing, which is the bar for most mnemonics.

The price of the contract itself — what the buyer pays and the seller collects — is called the premium. Think of it as the cover charge for admission to a bet.

What that premium is actually made of

A premium isn’t one number. It’s two, added together.

The first is intrinsic value — how much the option would be worth if you cashed it in this very second. If the stock’s already past the strike in your favor, that option is what’s called in the money (ITM, if you like acronyms), and intrinsic value is real money, sitting right there. If it hasn’t gotten there yet, it’s out of the money (OTM), and intrinsic value is zero — no credit for good intentions. Sitting exactly on the strike, neither up nor down, is at the money (ATM), which is exactly as indecisive as it sounds.

The second is time value — everything else. How much time is left before the option expires, how jumpy the stock has been lately, how likely a big move still seems. Time value is the market’s way of saying “sure, nothing’s happened yet, but there’s still time for it to.”

Time value is melting away the entire time the contract exists — the mathematical equivalent of a popsicle sitting on a hot sidewalk. But who that’s good or bad news for depends entirely on which side of it you’re on. Own the option, and it’s bad news: you’re the one paying the theta bill, watching value drip away whether the stock moves or not. Sold the option, and it’s the opposite: every day that passes without enough movement lets you keep a little more of what you collected. That daily leak has an official name, theta decay, and it accelerates particularly sharply for short-dated options sitting near the money.

The number hiding inside that second number

Here’s a term that gets used constantly around options and explained almost never: implied volatility.

Implied volatility isn’t a measurement of what a stock has done. It’s the market’s collective guess, baked directly into the time value portion of the premium, about how much the stock is going to move before the option expires. Picture it as a nervousness gauge. When everything’s calm, options are cheap, because nobody thinks anything wild is about to happen and there’s no reason to charge extra for insurance against a storm nobody sees coming. When a big earnings report is two days away, or a market’s already lurching around, implied volatility spikes — and so does the premium, even if the stock hasn’t actually moved a single cent yet. You’re not paying more because the stock changed. You’re paying more because everyone suddenly agrees it might.

Whether that guess turns out to be right is a completely different question, though — and the answer to it doesn’t live in the premium at all. It only reveals itself afterward, once the stock’s actually done whatever it was going to do. That distinction matters more than it might seem right now. Hang onto it — it comes back properly a little later in this piece, at the point where it actually starts mattering.

The sensitivity of an option’s price to that guess has a name too — vega — and it’s why an option can get more expensive even when the stock hasn’t moved a cent, simply because everyone got nervous at the same time.

Why would anyone buy one of these things?

A few reasons, and they’re worth knowing before we get to the more interesting question of who’s selling them.

Leverage. Buying 100 shares of a $100 stock costs $10,000. Buying a call option that controls those same 100 shares might cost a few hundred dollars. If the stock rises, the option’s value can rise by a much larger percentage than the stock itself did — same directional bet, dramatically less capital tied up.

Think of it like renting a fast car for the weekend instead of buying one. You can’t afford the Lamborghini outright, the same way you can’t afford 100 shares of Nvidia outright. But you don’t actually want to own the car forever — you want the thrill, the speed, the feeling of it, for a little while, without the six-figure commitment. So you rent it for the weekend instead. The weekend is your expiration date.

Actually, “rent” undersells it slightly. It’s closer to a lease. Because it’s not just that you can’t afford to buy the car outright — you don’t even want to. Owning it means insurance, maintenance, and eventually the genuinely annoying task of deciding when and how to sell it. A lease skips all of that. You drive it exactly as hard as you want, and when the term’s up, you hand back the keys — however you leave it — and you’re finished. No decisions left to make.

Which is exactly what an option gives you that owning the stock doesn’t.

Own 100 shares, and even a huge win comes with homework attached — when to sell, whether to trim some off the top or ride it further, what this now-much-bigger position does to the rest of your portfolio. Hold an option instead, and the contract simply ends on its own schedule. If it’s worth something, you sell it or exercise it. If it isn’t, you don’t have to do anything at all — no phone call, no decision, no ongoing relationship with the thing. It just quietly stops existing.

Maybe you get a speeding ticket. Maybe, if you’re feeling ambitious and mildly irresponsible, you enter it in an amateur rally and actually place — the automotive equivalent of an option finishing deep in the money, tires melted, suspension shot, tank on empty, absolutely worth it. Maybe you get lucky in some other way entirely that has nothing to do with horsepower. But the honest, unglamorous truth of leasing a fast car for a weekend is that most weekends, you just drive it around a bit, feel pretty good about yourself, and give it back Monday morning more or less the way you got it. No jackpot. No rally trophy. Just the thrill you paid for, and then the car’s gone.

That’s an option, too. Sometimes it goes to the moon. Sometimes it quietly expires worthless — like all of us eventually do, though hopefully with slightly less financial paperwork involved.

Speculation. Someone convinced a stock is about to move can buy a call or a put and bet on that move without ever touching the stock itself. And we all know how bets work — if your bet wins, you get your stake back plus something on top. If it loses, well, better luck next time.

Insurance. An investor holding a large position who’s nervous about a decline, but doesn’t want to sell, can buy a protective put — think homeowner’s insurance on the house you’re actually living in. You’re not selling. You’re not going anywhere. You just want a check in hand if a tree comes through the roof, so a bad storm doesn’t have to become a life-altering event. If the stock falls, the put pays out the same way, and the roof stays fixable.

In every case, the buyer is paying for the right to act later, without ever being on the hook to actually do it. That’s the whole appeal. You get to be indecisive, professionally, for a fee.

A concrete example of leverage cutting both ways

Say a stock is trading at $100. One path: buy 100 shares outright for $10,000, and sit on it. Another path: buy one call option, strike $105, for a premium of $2.50 per share — $250 for the contract — and ride this baby to its expiration date.

Now watch what happens to each of these across a range of prices the stock might be sitting at when the option expires:

Table 1.

Two things jump out of that table, and they pull in opposite directions.

On the downside, the option buyer’s loss is flat — capped at the $250 premium no matter how far the stock falls. It could go to $85, $50, $5, it does not matter. The stock buyer, meanwhile, just keeps losing more and more in a nice depressing straight line. That’s the entire sales pitch for buying options instead of stock: your worst day is capped, permanently, the moment you pay for the ticket.

On the upside, though, there’s a catch nobody mentions loudly enough: the option doesn’t start paying off the instant the stock crosses $105. It has to clear $105 and the $2.50 you already paid, which puts the real breakeven at $107.50. Below that, you were right about the direction and you still lose money — which is one of the more annoying lessons options can teach a person. But once you’re well clear of that breakeven, the returns get genuinely silly: the jump from $110 to $115 turns a 100% return into a 300% return, and by $125 you’re looking at 700%, while the poor stock buyer is having a perfectly nice but comparatively boring +25% day.

See that Delta column sitting there? We haven’t actually talked about what delta is yet — we’re about to — but keep an eye on it. It’s behaving a little strangely, and there’s a good reason why. We’ll come right back to it.

One honest note about “riding this baby to its expiration date,” since it’s worth flagging rather than quietly hoping nobody notices: this piece is about to argue that most option holders don’t actually do that. A profitable option usually gets sold before expiration, not exercised at the buzzer — you’ll see exactly why in a bit. So why build the table around holding to the very end?

Because it’s asking a specific, deliberately narrow question, and that’s actually the whole point of building it this way. Every number in that table is pure intrinsic value — what the option is worth with precisely zero time value left, the instant uncertainty fully resolves. That’s not the worst case or the best case. It’s a clean, isolated answer to “what does this payoff actually look like, stripped of everything else” — no implied volatility, no time value, no guessing. A real trader selling early, at some different point along the stock’s actual path, could walk away with more or less than these numbers, depending entirely on where the stock was and how much time was left when they sold. The table isn’t predicting that. It’s isolating the mechanics underneath it.

The risk is not the same on both sides

This is the part that actually matters for everything that follows, so it’s worth its own section rather than a footnote.

You already saw the buyer’s half of this a moment ago — risk is capped at the cost of the premium, no matter what. Full stop, no exceptions, nothing new to add there.

The seller’s side of that same contract looks nothing like this, and it’s worth sitting with how lopsided it actually is.

Sell a call without owning the underlying stock — a naked call — and if that stock takes off, you’re on the hook to deliver shares at the strike no matter how high it’s climbed. There is no ceiling on how high a stock can go, which means there is no ceiling on how much a naked call seller can lose. None. You sold someone the right to buy 100 shares from you at $105. If the stock is at $400, congratulations: you are now running a very small, very involuntary gift shop. The customer already paid the cover charge. You still have to find the merchandise.

Sell a put, and if the stock collapses, you’re obligated to buy shares at the strike even if the stock’s worth a fraction of that. A stock can’t fall below zero, so technically that loss is bounded — but “bounded at zero” can still mean losing nearly the entire value of a large position, against a premium that might have amounted to pocket change.

Buyer: capped risk, known in advance, sleep-easy-at-night risk. Seller: open-ended risk, or risk so large it might as well be open-ended. That asymmetry is not a minor detail. It’s the entire reason the rest of this piece exists — because nobody sensible collects a small premium and then just calmly accepts unlimited downside without doing something about it.

Why would a dealer take on a position they don’t even want to win?

Here’s a genuine puzzle, and it’s worth pausing on before moving forward, because if you don’t resolve it, everything downstream will feel like a contradiction.

A retail buyer who owns a call wants the stock to blow past the strike. That’s the entire point of the trade. Now think about the dealer sitting on the other side of that exact trade — the one who sold it to them. They’re the counterparty to that exact bet, and if the stock does blow past the strike, they lose money on it, dollar for dollar, right along with the buyer’s gain. So why not do something about it, instead of just sitting there quietly eating that risk like a bystander who accidentally wandered into someone else’s wager?

They actually do something. They hedge. The moment they end up holding that risk, they trade stock to cancel the exposure out, rather than just absorbing whatever the market decides to do to it. The dealer isn’t in the business of betting on direction, they’re in the business of being available. “Dealer” and market maker mean the same thing here: someone who quotes both a price to buy and a price to sell, all day, on demand, for whoever wants to trade. The price they’ll pay is the bid. The price they’ll sell at is the ask — always a little higher. That small gap between the two is the bid-ask spread, and it’s how the dealer actually makes a living, regardless of which way any given stock ends up moving. It’s less “gambler” and more “toll booth operator.” The toll booth operator does not care which direction traffic is going. They just want the toll.

To run that business, a dealer has to take the other side of whatever a customer wants, all day, in volume — selling a call to one customer betting the stock rises, buying an identical call from a completely different customer who’d rather sell theirs than hold it. Either way, the dealer just inherited a slice of directional risk they never asked for. It’s not a bet they wanted. It’s an occupational hazard, like a barista occasionally getting splashed.

So they hedge it away — not because they have an opinion about where the stock’s headed, but because they’d rather bank the spread and go home flat every night than end each day exposed to whatever the market happened to do. That’s the whole reason the hedging described in this piece runs in what looks like the “wrong” direction to a retail eye. The retail buyer wants the exposure. The dealer holding the identical contract wants absolutely nothing to do with it, and cancels it out with an offsetting stock position. Same paper. Opposite goal.

Say a dealer sells one call — 100 shares’ worth — on a stock at $100, strike $105. If the stock stays under $105, the buyer never exercises and the dealer just pockets the premium. If it climbs past $105, the dealer may owe those 100 shares at that price, and needs to either already have them or go find them fast, at whatever the market’s currently charging.

Dealers do not enjoy scrambling. So they hedge in advance — trading some of the stock right away to offset the profit-and-loss swings the option position is already creating, well before anyone gets anywhere near actually exercising anything. The question that immediately follows is: how much stock? Not necessarily the full 100 shares — the option might never get anywhere near the strike. Somewhere between none and all of it.

Why can’t the dealer just own the stock and be done with it?

Some do, in a limited way. If a dealer sells a call and already owns the 100 shares behind it, that’s a covered call. If the stock rises past the strike and gets exercised, the dealer just hands over shares they already had. No scrambling required.

That’s the entire reason a covered call has a reputation as a relatively tame, income-generating move for an ordinary investor. You already own the stock. You collect a bit of extra cash for agreeing to sell it at a price you’d already have been happy with. If it rallies through the strike, fine — you sell at a number you’d pre-approved.

But — and this is the part the friendly sales copy tends to leave out — “covered” is not the same word as “safe.” Being covered only removes one specific problem: scrambling to buy stock at a spiking price to hand it over. It does absolutely nothing about the stock simply collapsing. If the price falls off a cliff, a covered-call writer loses almost exactly what a plain stockholder loses, cushioned by nothing but that small premium. The call was never the dangerous part of this arrangement. The stock always was.

And here’s the bit that actually matters for a dealer: even setting the downside aside, “covered” doesn’t scale. A large dealer isn’t writing one call on one stock. They’re writing options on hundreds or thousands of names, all day, in both directions, buying from some customers and selling to others. Fully covering every single position — buying the whole 100 shares behind every contract, just in case — would mean parking an absurd amount of capital in stock that might never even get exercised. Nobody runs a business that way.

So dealers don’t. They hold a partial position instead — a fraction of the full 100 shares, sized to offset the profit-and-loss exposure the option itself creates. Not because they’re bracing for delivery day, but because every dollar the stock moves is already making or costing them money on the option they hold, right now, whether it ever gets exercised or not. The stock is there to cancel that live exposure out, not to prepare for some future handoff.

Which leaves an obvious question hanging: how much is enough? Too little, and the dealer’s still bleeding money on every move. Too much, and the “hedge” has quietly become a brand-new bet in the opposite direction, which rather defeats the purpose.

What the dealer actually needs is a number. Some running estimate of exactly how exposed they are, right this second, so they know precisely how much stock to hold. Not zero. Not the full hundred. Somewhere in between — and constantly moving.

That number has a name. Delta.

More on exactly what it means and how it works in a moment — but first, a quick detour, because a term’s been doing quiet work for the last few paragraphs and it’s worth pinning down before going any further.

A quick word on “investor,” before we get there

One term’s been doing quiet, slightly misleading work throughout the last few paragraphs, and it’s worth pausing to fix before moving on. Calling the covered-call writer an “investor” might sound odd. Isn’t the investor the one buying things — paying a premium for a protective put, say? Isn’t selling more of a trader’s move?

Not really. “Investor” was never about buying versus selling. It’s about why you’re holding a position, and for how long. An investor is acting on a belief about underlying value, usually over a longer horizon. A trader is acting on a short-term price move, and doesn’t much care about the underlying value at all.

Buying and selling cut straight across that line rather than tracking it. The cleanest proof is short selling. Someone who shorts a stock because they’ve genuinely researched it and believe the company is overvalued is investing — they’ve simply expressed that belief by selling first and buying back later, instead of the usual order of operations. Someone shorting the same stock overnight purely because of a headline is trading, even though the mechanics look identical from the outside.

Applied to options: the person writing a covered call against stock they already own and plan to keep either way is behaving like an investor, even though the transaction on paper is a sale. The person who bought that same call as a two-week bet is behaving like a trader, even though their transaction is a purchase. Same contract. Same moment. Opposite ends of the actual distinction, and neither one is decided by which side of the trade they happened to be standing on.

The distinction that actually matters for the rest of this piece isn’t investor versus trader, or even buyer versus seller. It’s scale and role: one person holding one position, of whichever kind — versus a dealer standing on both sides of thousands of them simultaneously, managing the net of an entire book rather than reasoning about any single contract the way a person naturally does.

Worth being clear about where that leaves the dealer, since it’s a natural next question: not on either side of the investor/trader line, but off it entirely. An investor holds a view on value. A trader holds a view on price direction. A dealer holds neither — they’re not betting on anything, they’re just providing liquidity and pocketing the spread, indifferent to which way the stock actually goes. That’s not a third example of investor or trader. It’s a genuinely different role, which is exactly why the rest of this piece stops asking “investor or trader” the moment a dealer walks into the picture, and starts asking something else instead.

Delta

Back to that number from before the detour — delta. Remember that Delta column, sitting there in the leverage table even earlier than that, quietly doing something we promised to explain? Here’s that promise, kept, properly this time.

Delta measures how much an option’s price — and therefore how much money the dealer’s currently making or losing on it — moves for every $1 move in the stock. A delta of 0.40 means: for every dollar the stock moves, the option’s value moves about 40 cents. Loosely, you can also treat it as “roughly a 40% chance this finishes in the money,” though that’s a convenient shorthand, not the technical definition, and math professors will glare at you for saying it out loud in front of them. A delta near 0.90 means it’s very likely already headed there. A delta near 0.10 means it’s a long shot. Right at the strike, delta sits around 0.50 — the genuine coin flip, could land either way.

One contract represents 100 shares, so a delta of 0.40 means the option behaves like about 40 shares of stock. Sell that call, and you’ve taken on roughly −40 shares of exposure — if the stock rises, you lose the way you would if you’d shorted 40 shares outright. So the dealer, determined to stay neutral rather than accidentally place a directional bet, buys about 40 shares to cancel it out.

That’s the hedge. Simple enough. Right up until the stock actually moves, at which point it stops being simple at all.

Say the stock climbs from $100 toward $105. As it does, the option’s sensitivity to the stock keeps growing — delta rises, maybe from 0.40 to 0.50. The dealer’s old hedge is now the wrong size. They were offsetting 40 shares’ worth of exposure; they now need something closer to 50.

So they buy 10 more.

Remember that Delta column from the leverage table earlier — the one that didn’t creep gradually the way 0.40-to-0.50 just did, but jumped straight from 0 to 1 right at the strike? Here’s what was actually going on. That table was priced at expiration, with zero time left on the clock. With no time remaining, there’s no more uncertainty left to price — the option’s either finished in the money or it hasn’t, full stop, like a coin that’s already landed. So delta stops being a gentle slope and becomes a cliff of its own: 0 below the strike, 1 above it, nothing gradual about the crossing. (Right at the exact strike, delta is technically undefined — the 0 → 1 is the cliff immediately surrounding that single point, not a defined value sitting on it.) Worth hanging onto that detail — it turns out to be the mechanical root of why options expiring “today” behave so differently from options expiring next month, which this piece gets to eventually.

Gamma

That rate of change — how fast delta itself moves as the stock moves, and therefore how much rebalancing the dealer has to keep doing — is gamma.

If delta is the current reading on the exposure gauge, gamma is how fast the needle is swinging.

This is the point where an options desk stops being a static bet and starts behaving like a live, twitchy participant in the market, minute by minute. A covered-call investor can sit on their hands. Nobody’s forcing them to trade every time the stock ticks. A dynamically hedged dealer, trying to stay close to delta-neutral — net delta hovering somewhere near zero — very often can’t afford to sit still. And which direction they end up trading in depends on something called the sign of their gamma.

Short gamma is fundamentally a dealer protecting itself against a contract it actually sold — the same $105 call from our running example. Having sold that call, the dealer’s exposure behaves like a stock short — they lose money as the stock climbs toward, and through, the strike, the same way someone who’d shorted the stock outright would. So the hedge chases that exposure in whichever direction matches it: buying more stock as the price rises toward a call they’ve sold, selling more stock as the price falls toward a put they’ve sold.

Which is exactly what happened a moment ago. They sold a call. As the stock rose toward the strike, their hedge had to grow with it — buying more stock into a market that was already rising. That’s not a coincidence. Overall, their aggregate gamma exposure — the sum of every option on their book, weighted by how much each one actually contributes — was negative. A dealer with negative net gamma is short gamma, and short-gamma hedging tends to run with the market: stock falls, they sell; stock rises, they buy. That’s procyclical — it can add fuel to whichever direction the market’s already headed. It’s also extremely common whenever customer demand for options is heavy and one-sided, like calls in a hot stock or protective puts ahead of a scary earnings report — someone has to write those contracts, and that someone frequently ends up short gamma for their trouble.

Worth being precise here, because it isn’t as simple as counting contracts sold versus bought. Gamma isn’t spread evenly across an options book. It concentrates heavily near the money, and can explode there specifically as expiration closes in — meaning a small handful of near-the-money contracts close to expiring can carry more total gamma than a much larger stack of far-out-of-the-money or long-dated ones, which tend to carry very little regardless of how close to expiration they are. What actually determines the sign is the sum, weighted properly, not the raw scoreboard of who sold more paper than they bought.

The put side runs the same way, mirrored. A dealer who sells a put has agreed to buy the stock at the strike if the holder wants out — insurance for the buyer, a slow-building liability for the dealer. As the stock falls toward that strike, the odds of actually being forced to buy climb, and the dealer starts behaving as though they already own stock they never asked for — losing money as the price drops, same as any reluctant owner would. So they hedge by selling stock, and the further the price falls, the more of that hedge they need. Same underlying logic as the call side, just facing the opposite direction.

Long gamma is the mirror image of the whole arrangement. A dealer whose aggregate gamma exposure nets out positive — say, a desk whose ongoing customer flow has left it net long options, or one deliberately holding options as a position of its own — hedges against the market instead of with it: stock falls, they buy; stock rises, they sell. That’s a stabilizing flow. It leans against the move rather than piling onto it.

On the call side, a long-gamma dealer who owns the call is hedged with a short stock position offsetting that positive delta. As the stock climbs and delta rises, that short position needs to grow to keep pace, so they sell more stock — selling into strength, which leans against the rally instead of adding to it. On the put side, a dealer who owns a put benefits as the stock falls, so they buy stock to keep pace — buying into weakness, the mirror image of selling into strength. Either direction, a long-gamma dealer leans against whatever the market’s currently doing, instead of following it.

So the whole distinction really comes down to one thing: whether the dealer’s aggregate gamma nets out negative or positive. Negative is short gamma, and it can amplify moves. Positive is long gamma, and it can dampen them. And crucially, that sign isn’t locked in for the day just because nobody’s traded another contract — the stock’s own price (traders call this spot, short for the spot price, the price right now) can drift through a strike, time can pass, a position near expiration can shed most of its gamma within the hour, and the composition of what’s driving the book’s net exposure keeps shifting even while the dealer’s own options positions sit perfectly still. Which is exactly how a single desk can be an accelerant at 10 a.m. and a brake by 2 p.m., without anybody deciding to change a single thing.

There’s a special case of this that deserves its own name: pin risk. When a stock’s price hovers right on a strike as expiration approaches, gamma is at its most extreme, and the dealer’s hedge becomes hypersensitive to even the tiniest moves. Dealers can end up buying just above the strike and selling just below it, which can create something like a gravitational pull toward that price — one reason stocks sometimes seem to “pin” to a strike right at the close on expiration Friday. It’s not a crash. It’s the mechanical reason the last hour can feel like a pinball machine.

Why bother hedging at all, if the option’s loss is already capped?

Fair question. A retail buyer’s loss on an option is capped at the premium, full stop — that was the whole point of the leverage table earlier. A long-gamma dealer holding the same option has that identical cap. So why bother trading the stock at all? Why not just let the thing run its course and either cash in or lose the premium, like everybody else?

Because for a dealer, the hedge isn’t purely defensive. It can be its own little profit engine, entirely separate from whatever the option ends up worth at expiration — a strategy with the delightful name gamma scalping.

Work it through with numbers. A dealer owns a call, delta 0.40, on a stock at $100, strike $105. To stay neutral, they short 40 shares.

The stock ticks up to $102. Gamma pushes delta from 0.40 to 0.55. The hedge is now too small, so the dealer sells 15 more shares — at $102.

Then the stock drifts back down to $100. Delta falls back to 0.40. They only need to be short 40 shares again, so they buy back those same 15 shares — at $100, two dollars cheaper than they sold them for.

Sold 15 shares at $102. Bought them back at $100. Profit: $30. None of that came from the option itself. It came purely from rebalancing the hedge as the stock wobbled — sell high, buy back low, over and over, every time gamma nudges the target.

You might reasonably ask why this can’t just be done with any stock, no option required, since “buy low, sell high” is every trader’s entire life ambition anyway. The honest answer is: it can’t, not with this kind of precision, because nobody can reliably tell when an ordinary swing has topped out or bottomed. Delta removes most of that guesswork on the size side — stock ticks up, delta rises by a calculable amount, and the dealer knows exactly how many shares would bring them back to neutral. It doesn’t tell them the perfect moment to trade — that’s a separate call, how tightly to chase the target versus wait for a better fill — but it tells them precisely where neutral sits at every instant. The option isn’t just an excuse to trade the stock. It’s the instruction manual for how much to hold, continuously updated.

And this ties straight back to theta and implied volatility from earlier, finally cashing in that promise. Remember the distinction between the market’s guess, priced in up front, and how things actually turn out? Here’s where it matters. That “how things actually turn out” number has a name — realized volatility — and it’s simply what the stock genuinely did, measured after the fact. Owning the option costs money every single day through theta decay, whether the stock moves or not. Gamma scalping is the attempt to earn that back. If realized volatility ends up beating the implied volatility that was baked into the premium — if the stock genuinely swung around more than the market guessed it would — the scalping profits can outrun the theta bleed, and the whole position turns a profit even if the option itself expires completely worthless. That takes some doing, mind you: the stock has to oscillate enough to generate those scalping gains without ever breaking through the strike in a way that leaves the option worth something at the end. It’s possible. It’s just not the base case. If the stock barely moves, realized volatility falls short of what was priced in, the scalping gains don’t cover theta and transaction costs, and the position loses money regardless of how carefully it was hedged.

Exercise, assignment, and what actually happens at the end

Everything up to now has occasionally mentioned an option “getting exercised,” as if that’s the default, satisfying conclusion to the story. For most individual holders, it genuinely isn’t.

If a call is deep in the money with real time still left before expiration, exercising it is usually the wrong move — it only captures intrinsic value and throws away whatever time value remains, on top of requiring the actual cash to buy the shares. Selling the option instead captures the intrinsic value and the leftover time value, which is almost always worth more. So most holders looking to bank a profit don’t wait around for expiration and don’t exercise at all. They just sell the contract, the same way they’d sell a stock they’ve decided they’re done with.

Who’s buying it? Could be a dealer, could be another investor — it doesn’t matter to the person selling. Their sale offsets the option they originally bought, and their position is gone. If the buyer happens to be a dealer who was already short that identical option, the purchase closes the dealer’s short too — same stock, same strike, same expiration, opposite direction, canceling out exactly the way described a moment ago. If not, the dealer just ends up holding a fresh position of their own, which they’ll then need to hedge like any other.

At expiration, options in the money by as little as a single penny are normally exercised under a process called exercise by exception — the clearinghouse exercises them by default unless someone specifically submits instructions saying not to — for any position that hasn’t already been closed out beforehand. “Sufficiently in the money” doesn’t mean some meaningful cushion. It means $0.01, unless someone actively opts out. But for most of an option’s actual life, selling it is how the profit gets realized.

One wrinkle worth knowing about: not every option gets to make this choice at every moment. Most individual stock options are American-style, meaning they can be exercised any day up to expiration, whenever the holder feels like it. Options on broad indexes — like the S&P 500 — are typically European-style, meaning they can only be exercised on the expiration date itself, no earlier. It doesn’t come up often in practice, since most people sell rather than exercise anyway, but it’s the reason index options behave slightly differently around the edges than the single-stock kind, and it’s worth knowing the label exists before you stumble across it somewhere and assume you’ve missed something.

How a dealer’s position actually closes

It’s worth being precise about what “buying it back” really means, because it isn’t a handoff between two specific people, the way returning a borrowed lawnmower would be.

Every listed option trade clears through a central clearinghouse — the Options Clearing Corporation, for U.S. equity and index options — which steps in as the counterparty to both sides the instant a trade executes. The buyer’s contract is technically with the clearinghouse. The seller’s contract is technically with the clearinghouse. Neither side is ever actually exposed to the other person directly, which is a genuinely elegant piece of financial plumbing nobody appreciates until they think about it for five minutes.

And because every $105 call on a given stock, expiring on a given date, is completely interchangeable with every other identical contract, the clearinghouse doesn’t bother tracking who traded with whom. It just tracks a running net position per account: how many of this exact contract are you short, how many are you long?

Worth being exact about the word “identical” there, because it’s doing all the work. Short and long aren’t two different kinds of contract — they’re just direction, on the exact same one. Selling that $105 call makes you short it. Buying that exact $105 call — same strike, same stock, same expiration date — makes you long it. A dealer who’s short that call cancels the position by buying that same call back, full stop. Not a put. Not a call at a different strike. Not one expiring a different week. The identical contract, opposite direction.

So when a dealer who’s short one call buys that identical call — from anyone at all, not necessarily the original customer — the two offset each other. Short one, buy one: back to zero. The dealer doesn’t have to track down the person they originally sold it to. They only have to buy the same option: same stock, same call or put, same strike, same expiration date. One position points one way, the other points the other way, and together they cancel. Nothing needs to be handed back to the original buyer, because listed option contracts are interchangeable with every other contract having those exact same terms.

What actually happens when someone gets assigned

Closing a position by buying it back is the tidy outcome. Sometimes it isn’t tidy — someone holds an option all the way to being genuinely exercised, and the clearinghouse has to actually make the swap happen, not just cancel two entries against each other. Worth walking through, because the mechanics answer a question that trips almost everyone up the first time they think about it: if a put’s strike is $105 and the stock’s crashed to $20, where do $105 shares even come from, when nothing on earth is worth $105 a share anymore?

Nowhere. There’s no such thing as “$105 stock.” There’s just ordinary stock, worth whatever it’s actually worth on the open market, and $105 is simply the fixed price two specific parties agreed to transact at, months earlier, before any of this happened.

Here’s the swap, in full. The holder of the put — the person who bought the insurance — delivers 100 real shares. The writer, the one who sold it and is now assigned, delivers $105 × 100 = $10,500 in cash. Both sides happen simultaneously, arranged through the clearinghouse — not a staggered “pay first, receive later” sequence, a single swap. Where did the holder’s 100 shares come from? If they already owned them — the classic protective-put scenario, insuring a position they’d been sitting on the whole time — those get delivered directly. If they didn’t already own any, the mechanics depend on their broker’s own policies: they might buy 100 shares fresh on the open market at $20 specifically to have something to deliver, immediately pocketing the $85-a-share gap as profit for having correctly bet the stock would fall — or, depending on the brokerage, exercising without owning shares can simply leave them holding a fresh 100-share short position instead, which carries its own ongoing risk rather than an instant, locked-in profit. Either way, the $20 market price and the $105 strike price never touch each other directly. They’re separate transactions that happen to be resolved by the same person, around the same moment.

And once that swap’s complete, the writer genuinely owns 100 shares they just paid $10,500 for — worth $2,000 on the open market. What they do next is entirely up to them, like any other stockholder: sell immediately and lock in an $8,500 loss, or hold and let the stock’s next move decide whether that number gets better or worse.

None of this happens on the writer’s own schedule, though. Once assigned, both deliveries — the shares, the cash — are required to settle within one business day. T+1, in the jargon: trade date, plus one. It’s not a target or a suggestion. It’s automatic, processed by the clearinghouse, and neither party gets a vote on the timing.

Which raises the obvious question: what if someone genuinely can’t hold up their end? Doesn’t have the cash, doesn’t have the shares, and the clock runs out anyway?

For most people this scenario basically can’t happen, because your own broker won’t let you exercise a put in the first place unless your account already has the shares, or enough buying power to go get them. The broker stands between you and the clearinghouse specifically to keep this from becoming a real problem. And the reason an individual option holder normally doesn’t have to worry about whether the stranger on the other side can perform is that the obligation gets absorbed into the clearing system itself: the clearinghouse stands between its member firms, holds enormous amounts of margin and a dedicated default fund specifically sized to manage exactly this kind of failure, and dramatically reduces the odds that any one party’s shortfall becomes anyone else’s problem. It’s not an infinite guarantee — the system has its own rules for genuinely extreme scenarios — but for the ordinary fail-to-deliver, the shortfall becomes the failing party’s broker’s problem, not the other side’s.

So the honest summary: central clearing and broker-level risk controls are specifically built to keep two individual traders from ever having to sort out a shortfall between themselves. The one real cost of things going sideways is designed to land on whoever caused it — not on the other side of the original trade.

Did the dealer actually make money on the trade?

Here’s where it gets genuinely interesting, because closing a position isn’t the same question as profiting from it — a distinction that trips people up constantly.

Take the call from the running example, strike $105, and follow it all the way to $115. The dealer sold that call and pocketed $2.50 a share up front — $250 for the contract. As the stock climbed from $100 to $115, the dealer was buying stock in pieces along the way to stay hedged: a little more at $102, more at $105, more still as delta kept climbing toward 1.0 deep in the money. By $115, they’re holding close to the full 100 shares, bought at a blend of prices somewhere in between — not all at once, and not at the top, because nobody’s that lucky.

Then the position closes. The dealer buys the call back at roughly its intrinsic value — (115 − 105) × 100 = $1,000 — and sells the accumulated stock, now worth $115 a share.

Three cash flows to net together: the $250 collected up front, the $1,000 paid to close, and whatever the stock hedge actually made or lost along the way.

Here’s the part worth being careful about, because it’s easy to get backwards. This dealer sold the call — they’re short gamma, the same dealer from the “which direction the hedge runs” section earlier. Short-gamma hedging means buying as the stock rises and selling as it falls — the classic pattern of buying into strength and selling into weakness. Any single trade in that sequence might individually show a profit or a loss; the real problem is what repeated rehedging after adverse moves does to the position as a whole, not that every single rebalance is a guaranteed loser. The $250 premium is what compensates for taking on that pattern of risk in the first place, priced in advance based on how much movement was expected.

So the honest way to think about it: delta hedging removes most of the directional bet. It does not remove volatility risk. A short-gamma dealer collects theta but pays for actual movement through the hedge. If realized volatility — how much the stock genuinely moved, over the actual time period involved — comes in below what was priced into the option, that trade can be profitable. If realized volatility comes in above what was priced in, the hedge can end up costing more than the dealer collected. Real-world spreads, gaps, and discrete rebalancing then make the result messier still. That’s the piece of logic worth carrying forward from here.

So yes: a dealer can genuinely lose money on a trade like this, even though it’s exactly the kind of trade their entire hedging apparatus exists to manage. That’s not the theory failing. It’s the gap between the idealized, continuous version of hedging and the discrete, cost-bearing version that actually exists in the real world. Which is precisely why market making is a statistical business rather than a guaranteed one — the edge shows up across thousands of trades, not necessarily any single one, and a fast, violent move is exactly the kind most likely to land as a loss on its own.

A few things worth knowing before you go

Options have their own bid-ask spreads, and they can be nasty ones. An illiquid, far-out-of-the-money option might have a wide gap between what you can buy it for and what you could immediately sell it back for — meaning you can be “right” about the stock and still lose money simply from the cost of getting in and out. Liquid, heavily traded options (think large, popular stocks or major indexes) tend to have tight spreads. Obscure ones do not.

Early assignment is a real, if occasional, hazard — mostly around dividends. If you’ve sold a call on a stock that’s about to pay a healthy dividend, the buyer sometimes has an incentive to exercise early to capture that dividend for themselves, which can catch a careless seller off guard. It doesn’t happen constantly, but it happens, and “I didn’t think that could happen” is a sentence a lot of first-time option sellers eventually say out loud. Worth knowing, too: assignment isn’t “the original buyer comes back and finds you.” When someone exercises, the clearinghouse randomly picks a member firm carrying a matching short position, and that firm then allocates the assignment among its own customers — randomly or first-in-first-out, depending on the broker. Anyone holding a short position in that contract is at risk the whole time it’s open, regardless of who they originally sold it to.

Most options don’t expire worthless — and they don’t get exercised, either. That’s a popular myth, and it’s worth correcting properly rather than repeating it. The exact numbers shift depending on the year and the dataset, but the shape of the story doesn’t: exercise is rare, somewhere in the single digits to low double digits. Expiring worthless is a real minority — meaningfully less than half. And the actual majority of contracts, by a wide margin, get closed out early instead: sold back into the market before expiration ever arrives, exactly the way this piece already described a while back. Most options don’t have a dramatic ending at all. They just get sold, the position closes, and everyone moves on to the next thing. The base rate is quieter than the myth, and worth paying more attention to than the excitement.

None of this is investment advice, obviously. It’s just the stuff a friend who happened to know this material would mention before letting you go trade real money on it.

Rent or lease. With an option to buy.

Remember the leased Lamborghini from the start of this piece? Here’s the part that might actually be the most useful thing in here.

A conventional closed-end car lease that includes a fixed purchase option — the kind where, at the end of the term, you can buy the car outright for a price agreed on before you ever signed anything — already contains a real option. Not a metaphor for one. An actual one, sitting quietly inside a transaction most people have signed without anyone ever calling it that.

The lease itself is what you pay to use the car during the term — closer to rent than to a premium, since it’s mostly covering the car’s expected depreciation, not the purchase right specifically. Unlike a listed call, the value of that embedded option usually isn’t printed anywhere as its own line item. But it’s there. Buried inside the deal is a genuine right, not an obligation, to buy at a fixed price, whenever the term runs out.

If similar cars are selling for more than your buyout price by then, you exercise — buy at the agreed number, pocket the difference in value. If they’re selling for less, you just hand back the keys and walk away, the same as letting a worthless call expire. Nobody makes you buy a car that’s now worth less than what you agreed to pay for it.

Buyout price is the strike. End of the lease is expiration. The purchase-option clause is the call. A market value above the buyout is in the money. Below it, you walk.

So if you’ve ever leased a car with a buyout price at the end — or a boat, or furniture, or anything else sold that way — you already understood a call option. You just didn’t have the word for it yet.

None of this is a person watching a screen

At a modern electronic market-making desk, this isn’t a trader glancing at delta once an hour and deciding to place a trade. Risk systems recalculate delta and gamma on every trade and on essentially every meaningful tick of the underlying — sub-second, not hourly — and an automated hedging layer can fire the rebalancing trade on its own, the moment net delta drifts past a threshold a human set in advance. Not a fixed schedule. A trigger.

Humans set the rules — how tight to keep the band, how aggressively to chase the market versus wait for a better fill, when to override the system entirely in a genuine crisis. But the moment-to-moment reality is a machine acting on its own, continuously, the instant its threshold is crossed.

That’s the mechanical foundation the rest of the Cliff Is Moving series builds on. With the basics covered — what an option is, why people buy them, why dealers hedge them, and what delta and gamma actually measure — Part 3 picks up from exactly here and asks the real question this piece was written to set up: what happens to market stability when that hedging machine can, depending on the day, run in either direction?

Regards,

Luke Kandia
Chasing Clarity

Luke Kandia writes Chasing Clarity, a Substack covering healthcare systems, financial markets, public policy, and science communication — usually by starting with something small, like an old coin, a data point, or a footnote, and following it wherever it actually leads.

Nothing here is investment, financial, or legal advice — just a curious fellow who is currently wondering how options work. Do your own homework before you make decisions with your money.

Sources and references

1. FINRA — “Trading Options: Understanding Assignment.” Source for the roughly 7% exercise rate and the description of OCC’s random assignment process, from firm to firm and then customer to customer within a firm. FINRA — Trading Options: Understanding Assignment

2. Options Industry Council (OCC) — “Options Exercise” FAQ. Source for exercise-by-exception and the $0.01 in-the-money threshold used by OCC absent contrary instructions. OIC — Options Exercise FAQ

3. Options Industry Council (OCC) — “Options Assignment” FAQ. Additional detail on random clearing-member selection and broker-level allocation to individual customers. OIC — Options Assignment FAQ

4. Options Industry Council (OCC) — “Basics” FAQ. Source for American-style versus European-style exercise, including that all exchange-traded equity options are American-style while most index options are European-style. OIC — Basics FAQ

5. OCC — Equity Options Product Specifications. Source for T+1 settlement of exercised equity options. OCC — Equity Options Product Specifications

Part 1 ended with a question I couldn’t yet answer: how would anybody actually see this happening, in real time, across the whole market? That question needs its own investigation, and it’s coming.

But before I could chase it, I had to finish mapping the machine itself.

Margin calls, it turns out, are only one gear in it.

One collapsing stock isn’t necessarily enough

Suppose Sandisk is 10% of a leveraged $100,000 portfolio. Sandisk drops 50%.

If everything else remains unchanged, the whole portfolio falls only 10% × 50% = 5%.

A 5% portfolio loss might be entirely survivable if the account began with plenty of excess equity above its maintenance requirement. But for an investor already close to that boundary — as we saw in Part 1 — even a loss that small may be enough to trigger a call. The size of the shock and the size of the consequence aren’t the same thing; the second depends entirely on how much room the account had to begin with.

Now change the portfolio. Sandisk. Micron. Nvidia. AMD. Broadcom. A semiconductor ETF. Other high-beta technology stocks.

Perhaps Sandisk is still only 10%. But perhaps 70% of the collateral is falling together.

The investor owns many ticker symbols. Economically, he owns one correlated trade.

That changes the important question from:

Which stock has fallen enough to create margin calls?

to:

Which leveraged investors own the same correlated collateral—and what happens when that collateral falls together?

China had already demonstrated the mechanism

Jiangze Bian, Zhiguo He, Kelly Shue and Hao Zhou obtained something we don’t have for the United States: account-level data.

Their study of China’s 2015 stock-market crash examined leverage, holdings and actual trading by brokerage-financed and shadow-financed investors.

They found direct evidence of leverage-induced fire sales. More constrained investors sold more aggressively, while stocks disproportionately held by constrained leveraged investors experienced greater selling pressure and abnormal declines. (Bian, He, Shue & Zhou — NBER)

That gives us an extraordinarily useful concept: distance to the financing boundary. How much farther can this portfolio fall before its leverage constraint becomes binding?

And investors don’t necessarily wait for formal liquidation. An investor approaching a known financing cliff can start selling before he reaches it. Then common holdings transmit the disturbance.

Investor A owns X and Y. X collapses. A needs to reduce risk. A sells Y. Y falls. Investor B owns Y and Z. B now needs to reduce risk. B sells Z.

X → Y → Z

No new bad news occurred at Y. No new bad news occurred at Z.

The market manufactured the transmission itself.

China provides direct empirical evidence that the mechanism operated in that market. It does not prove Chinese leverage distributions or liquidation thresholds can be transplanted numerically into America.

But the U.S. system potentially adds another mechanism: the financing boundary itself can move.

The underlying research has since been expanded and published: Bian, Da, He, Lou, Shue and Zhou, “The Drivers and Implications of Retail Margin Trading,” Journal of Finance (2026), which subsumes the original NBER working paper cited above. (Bian, Da, He, Lou, Shue & Zhou — Journal of Finance)

Then came July 2026

On July 29, the S&P 500 fell 1.52%, ending at 7,316.15. The Nasdaq fell 1.74%, and technology and semiconductor shares were among the major sources of weakness. (Reuters — Wall Street closes down sharply after Fed holds rates unchanged, July 29, 2026)

That’s unpleasant. But it isn’t March 2020. And that’s important.

A broad index decline does not tell us what happened inside a particular leveraged collateral basket.

Nasdaq’s official SOX history shows that semiconductor stocks had suffered much deeper cumulative losses over the preceding weeks than the broad market. The Philadelphia Semiconductor Index closed at a record 14,634.72 on June 22, 2026 — and by late July had fallen to roughly 10,447, a decline of just over 28% in about five weeks. On July 29 alone, the SOX fell further still, a single-day move several times the size of the S&P 500’s 1.52% decline that same day. (Nasdaq Global Index Watch — SOX History)

A customer’s actual technology portfolio can therefore be falling 10%, 15% or more while the S&P falls 1½%.

Archegos Capital Management was Bill Hwang’s family office — a private vehicle for his own wealth. As a qualifying family office, Archegos sat outside the SEC’s normal investment-adviser registration regime. Separately — and this is the part that actually mattered — Hwang used a derivative called a total return swap to build enormous concentrated positions in a handful of stocks, ViacomCBS and Discovery among them, without ever directly owning the shares. Because that exposure was synthetic rather than direct ownership, it stayed outside the public equity-holding disclosures investors normally rely on to see who owns what.

And because the swaps were spread across several banks, each bank could only see its own slice of the exposure — no single lender realized how much collateral risk the whole system was carrying.

By March 2021, Archegos had built roughly $160 billion of stock exposure on top of about $36 billion of portfolio value. When those stocks fell, the banks issued margin calls simultaneously. Archegos couldn’t meet them, and the banks began liquidating the same underlying stocks at the same time to protect themselves — the ones who sold fastest limited their damage, the ones who hesitated didn’t. Total bank losses topped $10 billion, Credit Suisse alone lost roughly $5.5 billion, and Hwang was later convicted on fraud and market-manipulation charges. For the full story, see the SEC’s own account of the case. (SEC — SEC Charges Archegos and its Founder with Massive Market Manipulation Scheme)

That is exactly the distinction Archegos taught the financial system in institutional form:

the index isn’t necessarily what breaks the leveraged structure.

The leveraged structure’s actual collateral is.

Archegos was not an ordinary FINRA customer-margin account, and its exposures do not belong in the customer-margin statistic discussed here. But Credit Suisse’s postmortem documented the mechanism: concentrated leveraged exposures, margin demands, forced unwinding and billions of dollars of losses. (Credit Suisse / SEC — Archegos Report)

Which leads to the next problem.

Margin investors aren’t the only actors who can become mechanical sellers.

Leveraged ETFs can arrive at the same trade

Before going further, it’s worth being precise about what an ETF actually is.

An ETF — an Exchange Traded Fund — is a fund whose shares trade on a stock exchange just like a stock. When you buy a share of a plain ETF, like one tracking the S&P 500, you own a small slice of a fund that actually holds the underlying stocks — Apple, Microsoft, Nvidia, and hundreds of others, in the right proportions. Your share is a claim on that basket.

Leveraged and inverse ETFs work differently, and the difference is the whole point of this section.

A 3× fund can’t simply “hold three times as much stock” — it doesn’t have three times the cash. Instead, these funds mostly build their exposure using derivatives: swaps and futures contracts.

A futures contract is an agreement to buy or sell something at a set price on a future date. Crucially, you don’t pay the full value of what you’re buying — you post a fraction of it, called margin, and the contract gives you exposure to the entire notional value. That’s leverage, built into the instrument itself.

One terminology trap worth flagging immediately: futures margin isn’t the margin loan from Part 1. Nobody has lent the trader the rest of the contract’s value. Futures margin is collateral — a performance bond posted to guarantee that losses on the contract can be paid — not a down payment on borrowed purchasing power. Same word, genuinely different mechanism.

A total return swap works differently but rhymes with the same idea: the fund enters into a contract with a bank, where the bank agrees to pay the fund whatever return a stock or index generates, in exchange for a financing fee. The fund never touches the actual shares. The bank on the other side will generally hedge its own exposure somehow — potentially with the underlying shares, futures, options, offsetting derivatives, or other positions already on its book — which is exactly the dealer-hedging mechanism described later in this piece, just wearing a different hat.

So a 3× leveraged ETF isn’t built by tripling ordinary stock ownership. It’s built out of instruments — futures and swaps — that are themselves engines of leverage: a small amount of posted capital controlling a much larger notional exposure. The fund uses these tools to reach its 3× target without needing 3× the cash on hand.

Holy handgrenade, indeed.

And this isn’t a hypothetical risk sitting in a prospectus footnote. It has already wiped out a fund’s investors — twice, in two different ways. More on that later in this series.

Leveraged and inverse ETFs generally target multiples of a benchmark’s daily return. A +3× fund wants approximately three times the daily benchmark return. A −3× fund wants approximately three times the inverse daily return.

Because the objective resets daily, exposure has to be rebalanced. ProShares explicitly describes geared funds as rebalancing exposure to their daily target. (ProShares — Geared ETFs FAQ)

Part 1’s cliff worked a specific way: a threshold sat still until a broker decided to move it, and the investor was fine right up until the moment his equity ratio crossed that line. A leveraged ETF doesn’t have a threshold like that — it doesn’t wait for a breach and then react. It has a target it’s contractually required to chase back to, every single trading day, regardless of how close or far it started. There’s no “fine for now.” There’s only today’s target, and tomorrow’s target, forever, on a schedule nobody negotiates.

That rebalancing doesn’t happen gradually, spread evenly across the trading day. Funds typically wait until they can see roughly where the day is going to end — where the underlying index they’re tracking, the benchmark, is likely to close — then execute in the final 30 to 60 minutes before the market shuts. That’s the same window when index rebalances, options hedging, and other institutional close-driven orders are also landing. (Bloomberg — AI-Dominated Leveraged ETFs Are Rattling Markets)

One more term before the math. NAV stands for Net Asset Value — it’s the actual value of what the fund owns, the fund’s own money. If you read Part 1, it’s the same idea as the investor’s equity: the real stake, as opposed to the much larger position that stake controls. A leveraged fund’s NAV is always smaller than its total market exposure — that gap between the two is exactly what “3×” or “−3×” describes.

Now imagine a +3× fund.

Start of day: $100 NAV × 3 = $300 exposure.

The benchmark falls 5%. That 5% move hits the fund’s $300 of economic exposure directly — a $15 loss. But only $100 of NAV stands behind that $300 exposure, so the fund absorbs the entire $15 loss out of a much smaller base. $15 lost against $100 of NAV is a 15% NAV decline. NAV falls from $100 to $85. That’s the leverage doing its work: a 5% move on the underlying becomes a 15% move on the fund’s own capital, because three dollars of exposure sit behind every dollar of NAV.

End of day: the fund needs $85 NAV × 3 = $255 exposure to stay at its 3× target.

But its actual exposure, which moved with the market itself rather than with the leveraged NAV, tells a different story: $300 × 0.95 = $285. The market only fell 5%, so the position built to track it only fell 5% too.

So the fund needs to close the gap — $285 of current exposure down to $255 of target exposure — by getting rid of $30 per $100 of starting NAV. It does that by going back to the bank or exchange that set up the swap or futures position in the first place and unwinding part of it.

Worth flagging right away: this isn’t necessarily $30 of actual shares hitting the market, the way the margin investor in Part 1 sold real stock to repay a real loan. What this $30 actually is — and why that distinction matters — is explained shortly.

Now consider a −3× inverse fund.

Start of day: $100 NAV × −3 = −$300 exposure (a short position).

The market falls 5%. That 5% move hits the fund’s −$300 short exposure directly — and since the fund is betting against the market, a falling price is a gain: $15. That $15 gain lands on a base of only $100 of NAV — a 15% gain. NAV rises from $100 to $115.

End of day: the fund needs $115 NAV × −3 = −$345 exposure to stay at its −3× target.

But its actual short exposure, having moved with the market itself rather than with the leveraged NAV, tells a different story: −$300 × 0.95 = −$285. Same logic as the bull fund — the market only moved 5%, so the position sized to track it only moved 5% too.

So the fund needs to close the gap the other way — from −$285 to −$345 — by adding roughly $60 of short exposure per $100 of starting NAV. It does that the same way the bull fund does: by going back to the bank or exchange on the other side of its swap or futures position and asking for more.

Worth being precise about what that bank is doing when it agrees. It isn’t taking a view that the stock is going to fall. It’s hedging its own side of the contract — the bank on the other side may offset it by holding the actual cash shares, using index or single-stock futures, entering an offsetting derivative, or netting the position against other exposures it already carries. Which method it uses depends on the bank’s own book. What doesn’t depend on the bank’s opinion is that it hedges at all: the fund is the one with the directional mandate, and the bank is mechanically covering its own exposure regardless of what it personally believes about the stock. That’s exactly why the resulting hedge adjustment can be relatively price-insensitive, whatever instrument it ends up moving through. And notice what that means: whichever instrument the fund uses to get its exposure, the hedge on the other end can still land in the same underlying stock market everyone else is trading in.

During a falling market: the leveraged bull reduces long exposure while the leveraged bear increases short exposure.

Different mechanisms. Same directional pressure.

Think about what that means on a bad day. The +3× fund’s $30-per-$100 sell order lands in the closing window — which, counterintuitively, is normally one of the deepest liquidity events of the entire trading day. NYSE’s own data shows its closing auction alone regularly moves hundreds of millions of shares and tens of billions of dollars daily. But depth isn’t the same thing as balance. A sufficiently large, one-sided imbalance can still overwhelm the liquidity available on the other side of that deep pool, especially when it’s landing on top of whatever other stressed selling is already concentrated into that same closing window. The fund didn’t cause the day’s decline. But it’s adding mechanical, price-insensitive selling directly into a moment where an enormous amount of other benchmark-sensitive trading is also converging. (NYSE — Behind the Record Volumes)

The mirror image is just as real for the inverse side. A −3× fund having a bad day is a day the market rallied — so its rebalancing buy order lands in the final hour of a rising session, adding fresh buying pressure exactly when momentum is already building. Bullish and bearish leveraged products tied to the same underlying can end up rebalancing in the same direction on the same afternoon, for opposite reasons.

Here’s the part that trips almost everyone up the first time: shouldn’t the bear fund be a buyer on a down day — swallowing up whatever the bull fund is dumping? It seems like they should cancel out.

They don’t. Walk through both funds on the same falling day, using the numbers just worked out above. The market falls 5%.

The +3× fund needs to shrink its position — sell $30 per $100 of starting NAV, as already shown.

The −3× fund actually profits on a falling market, since it’s betting against it. Its NAV grows, from $100 to roughly $115. But to stay at 3× that larger number, its short position has to grow too — from −$285 to −$345.

And here’s the key: making a short position bigger is itself a sell order.

You sell more of the thing to increase how short you are. So the −3× fund isn’t buying anything on this day. It’s also selling — roughly $60 worth, per $100 of starting NAV.

The inverse fund doesn’t show up to catch what the leveraged bull is dropping. It shows up to sell more of it.

Both are selling. For different reasons. Different mechanisms, same directional pressure — that’s the whole trick of it.

And on a rally, the same logic runs in reverse: the bull fund becomes a buyer to grow its position, and the bear fund becomes a buyer too, to shrink its now-losing short. Both sides can end up leaning the same way at the close, regardless of which direction the market moved that day.

And remember, none of this necessarily means $30 or $60 of Apple or Nvidia stock changing hands directly in the cash market. These funds mostly operate through the futures and swaps described above — so the buying or selling can take the form of futures contracts getting unwound or created, or swap notional getting adjusted with a bank, which can then trigger that bank’s own hedging desk to buy or sell the real shares. Direxion describes these instruments as central components of leveraged and inverse ETF implementation. (Direxion — Understanding Leveraged & Inverse ETFs)

The better description is: close-sensitive directional flow. That flow can migrate into cash stocks through dealer hedging and index arbitrage.

One honest caveat belongs here. The $30 and $60 figures above describe the mechanical rebalance a fund’s own math calls for — before accounting for investors creating or redeeming shares of the fund itself. Those flows matter. A 2014 Federal Reserve study found that capital flows into and out of leveraged ETFs substantially reduce the amount of rebalancing the fund actually needs to do, particularly on days when the underlying benchmark moves by a large amount — and showed theoretically that sufficiently large flows can eliminate the rebalancing need entirely. (Federal Reserve — Ivanov & Lenkey, “Are Concerns About Leveraged ETFs Overblown?”) So the $30 and $60 describe the pressure the structure wants to create — not necessarily the final amount that actually has to trade once real-world fund flows are netted against it.

This isn’t hypothetical. It’s happening in South Korea right now.

In April 2026, Korea’s Financial Services Commission approved the country’s first single-stock leveraged and inverse ETFs — 2× and −2× products tracking the daily return of one company at a time, rather than a broad index. Eligibility rules limited the pool to stocks large and liquid enough to qualify. Only two companies met the bar: Samsung Electronics and SK Hynix. (Korea Herald — Korea to debut single-stock leveraged ETFs with Samsung, SK hynix)

Korea had already seen hints of this mechanism before its own domestic products even existed. Korean investors could already trade overseas-listed leveraged products tied to Samsung and SK Hynix. On March 3, before the domestic launch, a UBS trading-desk estimate found that rebalancing tied to those existing products accounted for as much as 60% of SK Hynix’s entire turnover in the final hour of trading alone, on a day the stock plunged more than 10%. (The Korea Herald — How leveraged chip ETFs magnified Korea’s market swings)

Then, on May 27, sixteen domestic ETFs launched, alongside two leveraged ETNs, putting the mechanism directly onto Korea’s own exchange. What happened next is the mechanism just described, playing out in real time — and this time, Korean researchers put actual numbers on it.

The rebalancing isn’t merely theoretical here. Researchers could finally estimate it.

Korea’s own Capital Market Institute calculated the rebalancing demand implied by these funds’ actual assets under management and each day’s stock movement, then compared that estimated cash-equity demand against real trading turnover. Averaged across the sample, estimated rebalancing came to 1.6% of Samsung Electronics’ daily trading value and 2.1% of SK Hynix’s — generally within a 1–3% and 1–4% range, and rising further on the most volatile days, since rebalancing size scales with how much the stock moved that day. (KCMI — Market Trends and Implications Around the Launch of Single-Stock Leveraged and Inverse ETFs) That’s an estimate built from AUM and price data, not a direct tally of executed trades — but it’s real quantification of a mechanism that, until now, this piece could only describe conceptually.

And the leveraged products didn’t just add to trading in the underlying stocks — in Samsung’s case, they eventually traded more than the stock itself.

By early June, average daily turnover in Samsung’s leveraged ETFs reached 3.7 trillion won, nearly three times the 1.25 trillion won traded in Samsung shares directly. That raises what industry participants call the “wag the dog” risk: the possibility that activity in a derivative product becomes large enough to influence the price of the asset it’s supposed merely to track. The turnover numbers are real; whether the ETFs had actually begun driving Samsung’s price, rather than just trading heavily alongside it, is the open question the phrase is naming. (The Korea Herald — Samsung leveraged ETFs outtrade underlying shares)

Then, in late July, it stopped being a slow-burning concern and became a crisis — though not for the reason you’d first guess.

On July 28, Korean and Japanese semiconductor stocks sold off hard on news that had nothing to do with leveraged ETFs: a massive Chinese memory-chip IPO, reports of new Chinese lithography capability, and jitters over circular financing arrangements in the AI infrastructure buildout. Samsung fell more than 13%, SK Hynix more than 14%, and the Kospi dropped 10.84% to 6,023.66, triggering a circuit breaker.

The next day, July 29, SK Hynix reported its second-quarter results: the most profitable quarter in the company’s history — and still short of what an AI-charged market had priced in. The stock fell as much as 15% intraday before closing down 9.6%. The Kospi fell another 5.98%, to 5,663.24 — the first time circuit breakers had been triggered in both the Kospi and the Kosdaq on two consecutive trading days.

Roughly 864.5 trillion won — on the order of $595 billion — was wiped off Kospi market capitalization across those two sessions alone (600.3 trillion won on the 28th, 264.2 trillion won on the 29th).

That evening, an emergency “F4” meeting convened at 6 p.m.: the Deputy Prime Minister, the Bank of Korea governor, and the heads of both the Financial Services Commission and Financial Supervisory Service, all in the same room. Analysts pointed to panic selling compounding with mechanical rebalancing in the single-stock leveraged products as having widened the decline on top of the news itself. (Bloomberg — Korean Stocks Plunge 16% in Two-Day Burst of Retail Selling; Seoul Economic Daily — F4 Meeting Convened Urgently After 864 Trillion Won in Market Value Evaporates)

The index kept falling on July 30. Then, on July 31, the Kospi staged its largest single-day percentage gain in history — up 17.91%, to 6,595.45 — after strong U.S. tech earnings eased fears about AI infrastructure spending and semiconductor sentiment snapped back internationally. (KED Global — Samsung, SK Hynix shares jump nearly 30%; Kospi posts record daily gain)

A market that can lose roughly $595 billion in two sessions and then post its largest percentage gain in history two sessions later is exactly the kind of market in which mechanically procyclical flows matter most. They don’t have to cause the original move to amplify it — and amplification, not causation, is the claim this piece has actually been making.

One important caveat, and it’s the same one this piece has tried to hold onto throughout.

KCMI’s own research is explicit that a few weeks of post-launch data, arriving alongside a global semiconductor selloff that was happening for entirely separate reasons, makes rigorous causal attribution difficult. The institute’s own conclusion is that ETF rebalancing may have played a partial role in amplifying the swings — not that it caused them. KCMI also found a genuine counterforce worth including precisely because it complicates the story: after the first week, Korean retail investors themselves tended to trade counter to the rebalancing direction — buying more as prices fell, buying less or selling as prices rose — which the institute notes may have partially offset the procyclical pressure from the funds. That’s a meaningfully different, more defensible claim than “the ETFs caused this,” and it’s the one worth holding onto: researchers can now estimate the size of these rebalancing trades, regulators and the central bank are on record worried about their amplifying effect, and the trades are large enough to be measurable against real trading volume. Whether they caused any specific day’s move is a separate, harder question — and nobody with access to the real data claims to have settled it yet.

Regulators moved fast once the pattern became visible. The Bank of Korea itself warned on July 5 that concentrating leveraged products in just two stocks that already dominate the Kospi could deepen market skewness and amplify volatility. (The Korea Times — BOK joins warnings over Samsung, SK hynix leveraged ETFs) On June 22, South Korea’s Financial Supervisory Service governor Lee Chan-jin publicly expressed regret over how quickly the products had been approved. On July 16, regulators announced a ban on any new single-stock leveraged ETF launches and a plan to raise the minimum required deposit from about $6,800 to roughly $20,300 — though the deposit increase didn’t take effect immediately; it was later moved up to July 31 from an original early-August date, and a separate expansion of the minimum trading unit that had originally been scheduled for November was still being discussed as of mid-July, not yet in force. Following the late-July crisis, regulators went further still, announcing that they would pursue investor-specific position limits, with a 20%-of-portfolio cap given as the proposed benchmark. (Korea JoongAng Daily — Gov’t to halt new Samsung, SK hynix leveraged ETFs over volatility concerns; Korea JoongAng Daily — Korea raises cash deposit requirement for leveraged Samsung and SK hynix ETFs early; Korea JoongAng Daily — Gov’t scrambles to crack down on single-stock leveraged ETFs with Kospi in free fall)

This is not a stress scenario built to illustrate a mechanism. It’s the mechanism, unfolding on an actual exchange, quantified by actual researchers, with an actual central bank warning about it and actual regulators intervening mid-crisis.

Margin calls can create mechanical selling. So, it turns out, can the daily rebalancing of a financial product most people have never thought twice about buying.

But leveraged ETFs aren’t the only actor showing up to this trade without ever picking up the phone to place a discretionary order.

Part 3 looks at the other one: options dealers, and why their hedging can be either an accelerator or a brake, sometimes changing from one to the other in the same afternoon.

Regards,

Luke Kandia
Chasing Clarity

Luke Kandia writes Chasing Clarity, a Substack covering healthcare systems, financial markets, public policy, and science communication — usually by starting with something small, like an old coin, a data point, or a footnote, and following it wherever it actually leads.

Nothing here is investment, financial, or legal advice — just a curious fellow who is currently wondering how margin calls work. Do your own homework before you make decisions with your money.

Sources and references

1. Bian, Jiangze; He, Zhiguo; Shue, Kelly; Zhou, Hao — “Leverage-Induced Fire Sales and Stock Market Crashes,” NBER Working Paper 25040. Account-level evidence from China’s 2015 crash showing leverage-induced selling and transmission through common holdings. NBER — Leverage-Induced Fire Sales and Stock Market Crashes 1a. Bian, Jiangze; Da, Zhi; He, Zhiguo; Lou, Dong; Shue, Kelly; Zhou, Hao — “The Drivers and Implications of Retail Margin Trading,” Journal of Finance (2026). Peer-reviewed, expanded version of the NBER paper above, adding evidence that margin-induced trading contributes to shock spillovers across stocks. Journal of Finance — The Drivers and Implications of Retail Margin Trading

2. Reuters (via CNBC) — “Wall Street closes down sharply after Fed holds rates unchanged,” July 29, 2026. Source for the S&P 500’s 1.52% decline to 7,316.15, Nasdaq weakness and technology/semiconductor stress on July 29. CNBC — July 29, 2026 market close

3. Nasdaq Global Index Watch — Philadelphia Semiconductor Index historical data. Primary market-data source for the June–July 2026 SOX drawdown. Nasdaq — SOX Historical Data

4. Credit Suisse / SEC — “Report on Archegos Capital Management.” Credit Suisse’s investigation of the Archegos failure, documenting concentrated leveraged exposures, margin issues and the resulting approximately $5.5 billion Credit Suisse loss. SEC — Archegos Report 4a. U.S. Securities and Exchange Commission — “SEC Charges Archegos and its Founder with Massive Market Manipulation Scheme,” April 27, 2022. Background source explaining what Archegos was, how total return swaps allowed Bill Hwang to build concentrated, largely invisible positions across multiple banks, and how the March 2021 collapse unfolded. SEC — SEC Charges Archegos and its Founder

5. ProShares — “Geared (Leveraged & Inverse) ETFs” FAQ. Primary issuer explanation of daily investment objectives and daily rebalancing toward a fund’s leverage target. ProShares — Geared ETFs FAQ 5a. Bloomberg — “AI-Dominated Leveraged ETFs Are Rattling Markets,” 2026. Reporting on how leveraged-ETF rebalancing concentrates into the final minutes of the trading session, and how bullish and bearish products tied to the same underlying can end up rebalancing in the same direction. Bloomberg — AI-Dominated Leveraged ETFs Are Rattling Markets 5b. NYSE — “Behind the Record Volumes: A Hidden Opportunity.” Source for the closing auction’s role as the largest single liquidity event in U.S. equities, and Q1 2026 volume figures (605.5 million shares, $43 billion daily). NYSE — Behind the Record Volumes 5c. Ivanov, Ivan T.; Lenkey, Stephen L. — “Are Concerns About Leveraged ETFs Overblown?,” Federal Reserve FEDS Working Paper, December 2014. Finds that capital flows into and out of leveraged ETFs substantially reduce actual rebalancing demand, particularly on large-move days. Federal Reserve — Are Concerns About Leveraged ETFs Overblown?

6. Direxion — “Understanding Leveraged & Inverse Exchange Traded Funds.” Issuer documentation explaining daily leveraged objectives and the use of swaps, futures and other derivatives to obtain geared exposure. Direxion — Understanding Leveraged & Inverse ETFs 6a. Korea Herald — “Korea to debut single-stock leveraged ETFs with Samsung, SK hynix,” April 21, 2026. Source for the approval of Korea’s first single-stock leveraged and inverse ETFs, the 2×/−2× structure, and the eligibility rules that limited the pool to Samsung Electronics and SK Hynix. Korea Herald — Korea to debut single-stock leveraged ETFs 6b. KCMI (Korea Capital Market Institute) — “Market Trends and Implications Around the Launch of Single-Stock Leveraged and Inverse ETFs,” July 2026. Primary research source for the estimated rebalancing figures (1.6% of Samsung Electronics’ and 2.1% of SK Hynix’s daily trading value, calculated from AUM and price data rather than directly observed executions), the finding that retail investors traded counter to the rebalancing direction after the first week, and the institute’s own explicit caveat that causal attribution of market volatility to these products is not yet possible from available data. KCMI — Market Trends and Implications 6c. The Korea Herald — “How leveraged chip ETFs magnified Korea’s market swings.” Source for the UBS trading-desk estimate that rebalancing tied to overseas-listed leveraged products accounted for 60% of SK Hynix’s turnover in the final hour of trading on March 3, 2026 — before Korea’s own domestic single-stock products existed. The Korea Herald — How leveraged chip ETFs magnified Korea’s market swings 6d. The Korea Herald — “Samsung leveraged ETFs outtrade underlying shares; Tesla’s never have.” Source for the “wag the dog” data point: average daily turnover in Samsung’s leveraged ETFs (3.7 trillion won) exceeding turnover in Samsung shares themselves (1.25 trillion won). The Korea Herald — Samsung leveraged ETFs outtrade underlying shares 6e. Bloomberg — “Korean Stocks Plunge 16% in Two-Day Burst of Retail Selling,” July 29, 2026. Source for the corrected July 28/29 chronology: the July 28 selloff driven by unrelated news (a major Chinese memory-chip IPO, lithography and AI-financing concerns), and SK Hynix’s July 29 earnings miss as the specific catalyst for the second circuit-breaker day. Bloomberg — Korean Stocks Plunge 16% in Two-Day Burst of Retail Selling 6f. BigGo Finance — “Kospi Triggers Circuit Breakers for Second Straight Day; F4 Emergency Meeting Convened as Economic Chiefs Apologize Before Parliament,” July 29, 2026. Source for the first-ever back-to-back Kospi/Kosdaq circuit-breaker event and the F4 emergency meeting. BigGo Finance — Kospi Triggers Circuit Breakers for Second Straight Day 6g. Yahoo Finance / Bloomberg — “South Korea Holds Emergency Meeting as 864 Trillion Won Leaves Its Stock Market.” Source for the two-day market-cap loss figure and the July 29 Kospi close of 5,663.24. Yahoo Finance — South Korea Holds Emergency Meeting 6h. KED Global — “Samsung, SK Hynix shares jump nearly 30%; Kospi posts record daily gain.” Source for the July 31, 2026 record single-day Kospi gain (+17.91%, to 6,595.45) and the AI-sentiment catalyst behind it. KED Global — Kospi posts record daily gain 6i. Korea JoongAng Daily — “Gov’t scrambles to crack down on single-stock leveraged ETFs with Kospi in free fall.” Source for the post-crisis 20%-of-portfolio individual holding cap and the F4 meeting participants. Korea JoongAng Daily — Gov’t scrambles to crack down 6j. Korea JoongAng Daily — “Gov’t to halt new Samsung, SK hynix leveraged ETFs over volatility concerns,” July 16–20, 2026. Source for the July 16 regulatory announcement. Korea JoongAng Daily — Gov’t to halt new Samsung, SK hynix leveraged ETFs 6k. Korea JoongAng Daily — “Korea raises cash deposit requirement for leveraged Samsung and SK hynix ETFs early.” Source for the corrected effective-date timeline: the deposit increase was accelerated to July 31 from an original early-August date, and the minimum trading-unit expansion, originally scheduled for November, was still under discussion as of mid-July rather than immediately in force. Korea JoongAng Daily — Korea raises cash deposit requirement early 6l. The Korea Times — “BOK joins warnings over Samsung, SK hynix leveraged ETFs,” July 5, 2026. Source for the Bank of Korea’s direct warning about concentration and volatility amplification. The Korea Times — BOK joins warnings

I started with what seemed like a simple question:

How many margin calls were triggered by the July selloff?

I couldn’t answer it.

FINRA’s public statistics showed customer margin debit balances falling from $1.502 trillion in June 2026 to $1.417 trillion in July—a reported decline of:

$84.847 billion

FINRA warns that monthly changes can partly reflect firms changing their reporting methods, so that figure cannot simply be called $84.847 billion of forced deleveraging. And because the statistic is a month-end balance rather than a flow measure, it cannot tell us how much borrowing was repaid voluntarily, how much was extinguished through forced sales, or how much new borrowing occurred at the same time. (FINRA — Margin Statistics)

But trying to understand what could have happened underneath that number led somewhere much more interesting.

It led to the cliff.

The cliff

Suppose an investor holds $120,000 of securities, financed by $58,000 of their own money and $62,000 borrowed from their broker.

Their equity — the portion of the account that actually belongs to them — is $58,000. Their equity ratio, equity as a percentage of the total account value, is 58,000 / 120,000 = 48.3%.

Their brokerage requires that their equity ratio stay at or above 35% — "skin in the game," as it were. That 35% is called the maintenance requirement, and the broker sets it on its own terms. It isn't calculated from this investor's account, and it doesn't move just because the investor's equity ratio moves. It's simply the floor the broker has decided to enforce — and, as we'll see, the broker can move that floor whenever it wants.

If this investor’s equity ratio falls below the maintenance requirement, they get a margin call: They have to deposit cash, deposit more securities, or sell part of the position to bring the ratio back up. If they don’t, the brokerage can sell the securities for them — and it doesn’t have to ask first, or wait for any particular deadline. FINRA’s own disclosure rule is explicit on this point: firms aren’t required to notify a customer before selling, and even if a firm has given a customer a specific date to meet a call, it can still sell before that date arrives without further notice. In practice, many brokers informally allow a couple of days when a position is stable — but that’s a courtesy, not a right, and it can evaporate the moment the position keeps falling. (FINRA Rule 2264)

Right now the investor’s equity ratio is 48.3%, well above the 35% the broker requires. They’re fine.

Then the market falls.

Table 1 — The market hurts the investor

The value of the investor’s securities declines from $120,000 to $100,000. The broker’s $62,000 loan doesn’t. It stays fixed until the day it’s actually paid off. So the investor’s equity falls from

$58,000
to
$100,000 − $62,000 = $38,000

and the investor’s equity ratio falls to

$38,000 / $100,000 = 38%.

The securities fell ($120,000 − $100,000) / $120,000 = 16.7%.
The investor's equity ratio fell to 38%. But in dollar terms, their equity fell by a steeper 34.5% — from $58,000 to $38,000, since ($58,000 − $38,000) / $58,000 = 34.5%.

That is leverage doing the damage.

At a 35% house maintenance requirement, the investor is still okay. Their required equity is 0.35 × 100,000 = $35,000. They still have $38,000 in equity. So the investor has a $3,000 cushion.

Then the broker moves the cliff

For the investor’s account let’s hammer out the variables:

A = $100,000 (how much the entire Account is valued at after the market downturn)
D = $62,000 (how much Debt or loan was floated by the brokerage)
E = A − D = $38,000 (the investor’s Equity or “skin in the game”)
C = mA − E (the margin defiCiency or exCess) where m is the broker’s house maintenance requirement (in our example case - 35%).

At a 35% house requirement:

C=(0.35)(100,000)−38,000=−$3,000

The negative number means the investor is $3,000 above the required equity level. The investor is okay.

Now let’s see what happens when the brokerage tightens the requirement. The value of the securities hasn’t changed.

But the broker is getting worried about their loan. So they raise how much “skin in the game” the investor has to have, in order to stay in the game.

Table 2 — What the moving cliff forces the investor to do

At 35% maintenance requirement, even after a market move, the investor had $3,000 excess equity or “skin in the game.” But after the broker moved the maintenance requirement to 40%, suddenly they are $2,000 short.

Now if they simply added $2,000 cash all would be forgiven as their equity would be $40,000 / $100,000 = 40% maintenance requirement.

However, usually investors that have taken out brokerage loans, don’t have any excess cash kicking around. So they have to sell some of the securities they bought using borrowed money, and use the proceeds to reduce the loan.

How much?

If the account is deficient (C is a +positive number) and securities are sold to repay the margin loan, the minimum securities Sale required is:

S(min) = C/m = A − E/m

So in our example above, if the maintenance percentage moved from 35% to 40%

Remember that

C= mA − E
or
C=(.4)(100,000) - (38,000) = 40,000 - 38,000 = $2,000 deficiency

Back to the minimum security sales formula

S(min) = C/m = $2,000/0.4 = $5,000 min sale of securities

Why does a $2,000 deficiency require $5,000 of selling?

Because the investor’s dollar equity does not rise when securities are sold. The securities are being sold to repay the loan.

By repaying the loan, the investor shrinks both sides of the leveraged balance sheet.

If $5,000 of securities are sold:

A=100,000-5,000=95,000 - the new account value

and the $5,000 of proceeds are used to repay margin debt:

D=62,000-5,000=57,000 - the new broker loan amount

The investor’s equity is still:

E=A-D
E=95,000-57,000
E=$38,000

But now that same $38,000 represents:

38,000/95,000 = 40% = to the maintenance equity percentage

of the remaining account.

That is the mechanism.

The investor does not cure the deficiency by increasing their dollar equity. They cure it by shrinking the leveraged position until their existing equity becomes a large enough percentage of what remains.

Plain and simple: the broker is protecting its loan.

Across the tightening sequence 35% → 40% → 45% → 50%, the required securities sales total $5,000 + $10,556 + $8,444 = $24,000.

So the investor’s securities position shrinks from $100,000 to $76,000, while his margin debt falls from $62,000 to $38,000. His equity remains $38,000.

But remember where he started.

Before the market decline: $58,000 of investor equity.

After the market decline: $38,000 of investor equity.

Then the brokerage raises the house requirement and forces $24,000 of securities into the market.

That distinction is crucial.

The investor was already falling toward the cliff.

Then:

The cliff moved. Closer to the investor.

FINRA and the Rules

This isn’t merely a hypothetical power we’ve given the broker for our example.

FINRA’s ordinary maintenance floor for long margin stocks is generally 25%, but brokerage firms can impose higher house requirements. They can also liquidate positions when an account no longer complies. (FINRA — Know What Triggers a Margin Call)

FINRA explicitly warns that firms can increase house requirements without advance written notice. A change in the firm’s house requirement can therefore create a margin call even if the investor has made no new trade and the securities have not fallen another dollar. (FINRA Rule 2264; FINRA — Know What Triggers a Margin Call)

And that produced the sentence that changed the direction of this entire investigation:

The investor can be falling toward the cliff while the brokerage simultaneously moves the cliff toward the investor.

No conspiracy is required.

Different brokerages can be looking at the same volatility, concentration, liquidity, correlation and gap risks.

Their risk systems can independently reach similar conclusions at approximately the same time.

What is rational risk management for each brokerage individually can therefore become destabilizing collectively.

Because now we add price impact.

One investor’s forced sale can push prices lower.

Those lower prices can destroy still more equity in that same leveraged account — or in entirely different leveraged accounts holding the same securities.

Those accounts may then approach their own maintenance requirements.

Some may be forced to sell.

So the sequence becomes:

investor loss > forced selling > lower prices > new losses in the same or other leveraged accounts > more forced selling

Or, stripped to its essentials:

deficiency → selling → lower prices → larger deficiency → more selling

That’s our first genuine feedback loop.

And notice what happened.

The investor took the loss first.

Then the market took the forced selling.

Now that forced selling can create the next investor loss.

That is the feedback loop.

And once I understood it, the obvious next question was:

Can anybody actually see this happening in real time?

How many margin calls were issued?

How large were they?

How many accounts were liquidated?

How much stock was actually forced into the market?

FINRA’s public monthly statistics can’t answer those questions.

But here’s the strange part:

somebody is keeping score.

And that is where the next investigation begins.

Regards,
Luke Kandia
Chasing Clarity

Luke Kandia writes Chasing Clarity, a Substack covering healthcare systems, financial markets, public policy, and science communication — usually by starting with something small, like an old coin, a data point, or a footnote, and following it wherever it actually leads.

Nothing here is investment, financial, or legal advice — just a curious fellow who is currently wondering how margin calls work. Do your own homework before you make decisions with your money.

Sources and references

1. FINRA — “Margin Statistics.” Source for FINRA’s monthly customer margin-debit balances, including the June and July 2026 figures, and FINRA’s caution concerning month-to-month reporting variations.
   FINRA — Margin Statistics

2. FINRA — “Know What Triggers a Margin Call.” Explains maintenance requirements, higher brokerage house requirements, margin calls and brokerage liquidation rights; also provides FINRA’s worked example showing that a margin deficiency can require substantially more securities to be sold than the dollar amount of the deficiency.
   FINRA — Know What Triggers a Margin Call

3. FINRA Rule 2264 — Margin Disclosure Statement. Requires disclosure that firms may increase house maintenance requirements without advance written notice and may liquidate securities without first contacting the customer or waiting until a previously stated deadline.
   FINRA Rule 2264

4. FINRA Rule 4210 — Margin Requirements. FINRA’s underlying margin rule, including the general maintenance-margin framework governing customer accounts.
   FINRA Rule 4210

This one is longer than usual. I started with Charles de Gaulle, somehow ended up in a 1951 Federal Reserve Bulletin, and things deteriorated from there. Get ready for a ride.

This article was an accident.

Which, oddly enough, is part of the story.

A few days ago, I was discussing something I had previously written about the end of the gold-backed U.S. dollar in the comment section of a Bill Bonner article.

My thumbnail version went something like this:

By the 1960s, the United States had created more dollar claims than its gold reserves could comfortably support at the official price of $35 an ounce. France noticed. Charles de Gaulle began converting French dollar reserves into American gold. Other countries followed.

Eventually Richard Nixon closed the gold window in 1971.

Patrick, a reader of Bill Bonner’s replied that I was only “half right.”

Congress, he correctly pointed out, doesn’t print money. Congress appropriates spending. The President signs legislation. Treasury finances government borrowing. The Federal Reserve controls monetary policy.

Fair enough.

He then placed much of the blame for America’s monetary problems on Lyndon Johnson’s Great Society and the Vietnam War before turning his fire on Charles de Gaulle, whom he described, among other things, as an arrogant and ungrateful jerk.

Apparently De Gaulle was the villain in this particular telling.

I began writing a response.

And then I made the mistake, or perhaps the very good mistake, of asking another question.

How much gold did the United States actually have before De Gaulle came along?

That question ruined everything.

Because suddenly the familiar story about France calling America’s bluff in 1965 wasn’t the beginning of the story at all.

It was almost the end.

The $7 billion question

In August 1949, the United States held roughly $24.8 billion worth of monetary gold.

At the official price of $35 per ounce, that represented more than 700 million ounces.

By 1960, the American gold stock had fallen to less than $18 billion.

Something approaching 200 million ounces of monetary gold was gone.

And Charles de Gaulle hadn’t yet made his famous challenge to the dollar.

So I asked the obvious question:

Where the hell did the gold go?

That sent me into old Federal Reserve records.

And there it was.

The first full postwar year of net U.S. gold outflow was 1950.

This isn’t reconstructed history. The Federal Reserve wrote the story down in March 1951, in an article wonderfully titled The International Movement of Gold and Dollars in 1950.

For several years after World War II, the flow had been going in the opposite direction.

Europe and Asia had been devastated by war. America had the factories, machinery, food and other goods that the world desperately needed.

Other countries needed dollars to buy those American products.

And when they didn’t have enough dollars, some paid America in gold.

The American gold pile grew.

At the end of 1945, the United States held about $20.1 billion worth of monetary gold.

By August 1949:

$24.8 billion.

Then the direction reversed.

And according to the Fed, the reversal was already underway by the middle of 1949. America’s huge export surplus was shrinking, U.S. financial assistance abroad exceeded that surplus, and private American capital continued flowing overseas.

The Korean War added another jolt to international commodity and dollar flows during 1950, but it didn’t create the reversal. The direction had already changed. The Fed’s own retrospective puts the turning point in August 1949.

During 1950 alone, the United States lost roughly $1.7 billion of gold.

And the largest taker?

Not France.

Britain.

The United Kingdom bought approximately $1.02 billion of U.S. gold in 1950.

Don’t take my word for it. The Fed actually published the country-by-country shopping list.

France bought about $85 million.

The Netherlands, about $80 million.

Belgium, $58 million.

Switzerland, $38 million.

Sweden and Portugal joined in.

Canada, Mexico and others appear in the records too.

And by the second half of 1950, Britain was buying so much gold that its purchases actually exceeded its current dollar receipts. It began drawing down dollar balances it had previously accumulated.

Suddenly my neat little tale about De Gaulle looked rather incomplete.

He was fifteen years late to the party.

Back to Bretton Woods

Which produced another question.

Why in God’s name would the countries meeting at Bretton Woods in 1944 agree to make the U.S. dollar the centre of the world’s monetary system if America couldn’t possibly redeem all the dollars in existence for gold?

The answer is:

They never promised to.

That had been my conceptual error.

American citizens did not have a Bretton Woods promise allowing them to walk into Treasury with $35 and demand an ounce of gold.

Domestic gold convertibility for Americans had effectively been severed under Roosevelt in 1933–34, something I wrote about here.

And here’s another little wrinkle I had forgotten:

Bretton Woods didn’t invent the $35 gold price.

Roosevelt’s administration had established that price in 1934, after the Gold Reserve Act. The Federal Reserve’s history of the Act is here.

Bretton Woods inherited it.

The important Bretton Woods promise was international.

Other participating countries maintained exchange rates against the dollar, while the United States maintained the dollar’s relationship to gold at $35 an ounce for foreign monetary authorities. That’s the actual plumbing of Bretton Woods.

Now look at the balance sheet confronting policymakers in 1944.

At year-end, the United States held about $20.6 billion of monetary gold.

Its short-term dollar liabilities to foreign official institutions, essentially liquid dollar balances and other short-term claims held by foreign central banks and governments, were only about $3.3 billion.

These weren’t every dollar held overseas.

They were the readily accessible official dollar claims most relevant to America’s promise to convert dollars into gold.

Against those claims, America wasn’t sitting precariously on the edge of insolvency.

Quite the opposite.

It had more than six dollars of gold for every dollar of those short-term official liabilities.

No wonder they trusted the dollar.

Britain was financially exhausted.

France had been occupied.

Germany and Japan were being destroyed.

America was the industrial powerhouse of a world being blown to pieces and possessed an extraordinary share of the world’s monetary gold.

If you were sitting in New Hampshire in 1944 trying to construct a monetary system for the postwar world, the U.S. dollar didn’t merely look like a good choice.

It looked like the obvious choice.

And during the first few years after the war, the arrangement looked even better.

Countries desperately needed American products.

They desperately needed dollars to buy them.

When they couldn’t obtain enough dollars, gold flowed toward America.

Hence:

About $20 billion in American gold in 1945.

Then:

Almost $25 billion in 1949.

The system appeared to be working beautifully.

Until something changed.

Britain wasn’t calling America’s bluff

This is where I nearly made another mistake.

Calling the 1950 gold outflow a “run” on America makes for a wonderful headline.

It also isn’t really what happened.

Britain wasn’t necessarily behaving like De Gaulle would fifteen years later.

Europe was recovering.

Countries that had depleted their monetary reserves during the war were rebuilding them.

And despite Bretton Woods giving the dollar a central reserve role, central bankers still liked gold.

A fascinating 2019 paper by economists Eric Monnet and Damien Puy calls this the "Bretton Woods gold puzzle".

Why did foreign monetary authorities continue accumulating so much gold when the new international system allowed U.S. dollars to serve alongside it as reserves?

Part of their answer is wonderfully human.

Habit.

Institutional memory.

Central bankers who had spent careers under the old gold standard continued behaving as though gold really mattered.

Countries with longer histories under the prewar gold standard tended to continue accumulating more gold afterward.

In other words, dollars and gold never became the perfect substitutes for each other that Bretton Woods implicitly expected them to be.

The architects had built a new monetary system.

The central bankers brought some of the old one with them.

So Britain converting an enormous pile of dollars into gold in 1950 wasn’t necessarily shouting:

“We don’t trust America!”

It was closer to:

“Thanks for the dollars. We’ll take some gold too.”

And Bretton Woods explicitly allowed them to do it.

Which means something rather important:

American gold began leaving while Bretton Woods could still reasonably be described as a success.

The warning light kept flickering off

That made me ask another question.

If Britain had just converted more than $1 billion of dollars into American gold in a single year, didn’t anybody in Washington clear their throat and say:

Uh, guys?

Of course they noticed.

The Federal Reserve itself called August 1949 “the turning point.”

But here’s the fascinating part.

They didn’t initially see the reversal as evidence that Bretton Woods was beginning to fail.

They saw much of it as evidence that the postwar world was finally recovering.

Foreign countries had spent years desperately running down their reserves to buy American goods.

Now their economies were rebuilding.

Their exports were recovering.

Their reserves were rising.

Britain’s improved reserve position strengthened confidence in sterling.

From that perspective, gold leaving America wasn’t necessarily a warning.

It could look like success.

That is almost exactly how the Federal Reserve described what was happening at the time.

And then the signal became even harder to interpret.

Gold flowed back toward America in 1951 and 1952.

The U.S. gold stock rose from about $22.7 billion at the end of 1950 to $23.3 billion at the end of 1952.

Crisis over?

Not quite.

From roughly mid-1952 through April 1955, the Federal Reserve later calculated that an “almost uninterrupted” $1.8 billion gold outflow had occurred.

Their words, not mine.

Okay.

Warning light back on.

Except then it flickered off again.

In 1956, America’s gold stock actually rose by about $306 million, to $22.058 billion.

In 1957, it rose again, by almost $800 million, to $22.857 billion.

And in March 1957 the Federal Reserve looked at America’s position and did not sound like an institution staring into the abyss.

It calculated U.S. “net reserves”, gold minus short-term liabilities to foreign central banks and governments, at about $14 billion.

That was equivalent to roughly 100% of an entire year’s U.S. merchandise imports.

The Fed noted that this ratio was about twice as high as that of any other major industrial country.

Its conclusion?

Developments in international gold and dollar flows gave:

“promise of continued stability in international financial relations.”

Yes. Really. March 1957.

And this, I think, matters enormously.

Because it makes the policymakers of the 1950s look less stupid.

Nobody missed the gold movements.

The harder problem was that the warning light kept flickering off.

1950:

Huge gold outflow.

1951–52:

Some of the gold comes back.

1952–55:

The drain resumes.

1956–57:

Gold comes back again.

Then came 1958.

And this time the light stayed on.

America’s monetary gold stock fell from $22.857 billion at the end of 1957 to $20.582 billion at the end of 1958.

A loss of roughly $2.275 billion in a single year.

During just the first half of 1958, the Fed reported roughly $1.5 billion of renewed transfers of gold and dollars to foreign countries.

The 1958 deterioration is laid out here in the Federal Reserve’s contemporary account.

Now Washington really did begin to worry.

And there’s a lesson hiding in there.

A changing variable does not arrive conveniently labelled:

WARNING: THIS WILL DESTROY YOUR MONETARY SYSTEM IN TWENTY YEARS.

Sometimes it looks like failure.

Then recovery.

Then failure.

Then recovery again.

Until one day it doesn’t recover.

But where were all those dollars coming from?

Another question.

I had been talking loosely about America “printing too many dollars.”

That’s not quite right either.

The important problem wasn’t simply the total number of dollars inside the American economy.

It was the number of dollar claims accumulating outside the United States.

And this is where I finally had to understand something I had previously glossed over:

The balance of payments.

Imagine America sells $10 billion worth of goods and services overseas.

Foreigners therefore send $10 billion back to America.

Fantastic.

America has a trade surplus.

But during the same period:

American companies invest $3 billion building factories overseas.

The U.S. military spends $3 billion maintaining troops and bases abroad.

Washington sends $2 billion abroad in aid.

American banks lend $2 billion overseas.

American tourists and investors spend another $2 billion.

America has earned $10 billion from the rest of the world.

But it has sent $12 billion back out.

The difference is:

$2 billion of additional dollar claims accumulating overseas.

And here was something else I hadn’t appreciated.

America could have a trade surplus and still experience this problem.

It could sell more goods abroad than it imported and nevertheless send more dollars overseas overall through investment, military spending, lending, aid and other capital flows.

By the late 1950s, the Fed was describing exactly this phenomenon: countries with balance-of-payments surpluses were using substantial portions of their dollar receipts to acquire American gold. Its 1958 review is particularly illuminating.

Those receiving the dollars now had choices.

Buy American goods.

Invest the dollars back in America.

Keep them as reserves.

Or, if they were the appropriate foreign monetary authorities under Bretton Woods:

Ask for gold.

A federal deficit is not the same thing

This also cleared up another piece of confusion created by my original discussion.

Was the United States running federal budget deficits while all this was happening?

Sometimes.

But not always.

A federal budget deficit and an international balance-of-payments deficit are two different things.

Washington could theoretically balance its budget while the United States as a whole still sent more dollars overseas than it received.

American corporations could invest abroad.

American banks could lend abroad.

The military could spend abroad.

Tourists could spend abroad.

Foreign aid could go abroad.

None of that requires Washington itself to be running a fiscal deficit.

Conversely, the federal government could run a deficit while much of that money remained within the domestic American economy.

Two different ledgers.

I had mentally combined them.

In fact, the distinction is almost absurdly clear in 1960.

The federal government essentially balanced its budget that year. You can disappear into the government’s historical budget tables here, should that sound like a fun evening.

Yet America’s international payments problem was getting worse.

That year alone should have killed the simple explanation:

“Washington ran budget deficits, therefore America lost its gold.”

The actual mechanism was more complicated.

And more interesting.

Then the two lines began to cross

At this point I wanted to see the problem stripped of almost everything else.

So I compared two things:

U.S. monetary gold

and

short-term dollar liabilities to foreign official institutions.

That second measure is deliberately narrow.

It doesn’t represent every dollar held outside America.

Broader measures of foreign dollar claims were substantially larger.

I use the narrower series here because it captures short-term dollar claims held by foreign central banks, governments and other official institutions, the foreign holders most directly relevant to pressure on America’s gold-convertibility promise.

One small wrinkle:

I have referred several times to America’s August 1949 peak gold holdings of roughly $24.8 billion.

The table below uses year-end figures, so 1949 appears as $24.427 billion.

And I’ve now added several years from the middle of the 1950s.

They matter.

Because they show that the road to 1971 was anything but a straight line.

Selected year-end figures from Federal Reserve and Treasury international-capital statistics. The mid-1950s figures are particularly useful because they expose the reversals hidden by simply drawing a line from 1950 to 1960. The Fed’s contemporary short-term-liabilities tables are here, and the later Treasury series is here. Reporting categories and coverage were revised over this long period, so treat this as a pressure gauge rather than pretending the accounting definitions were frozen in amber. The 1949 gold figure is year-end; the postwar peak was approximately $24.8 billion in August. The 1971 liabilities figure is approximate because of a reporting-coverage change.

Now you can see the problem.

In 1944, there was more than $6 of monetary gold for every $1 of those official short-term dollar liabilities.

By 1949, it was more than $8.

Bretton Woods looked incredibly safe.

Then watch the cushion disappear:

1955: $3.13.

1956: $2.74.

Then something interesting happens.

1957: $2.89.

The ratio actually improves.

Only slightly.

But if you’re sitting inside the Federal Reserve in 1957, that matters.

The gold stock is rising.

The ratio is improving.

America still possesses an enormous reserve cushion.

The warning light has flickered off again.

Then:

1958: $2.38.

1960: $1.74.

1965: $1.06.

Essentially one-for-one.

And by 1971:

Twenty-six cents of gold for every dollar of those short-term official liabilities.

Now the longer story becomes much clearer.

It wasn’t simply:

Gold goes down. Dollar claims go up. Everyone ignores it for twenty years.

It was messier.

Gold leaves.

Gold comes back.

Gold leaves again.

The reserve cushion shrinks.

Then improves.

Then shrinks much faster.

Meanwhile, foreign official dollar claims keep accumulating.

By the late 1950s, the temporary recoveries were no longer enough to reverse the underlying trend.

Two things were ultimately happening:

Foreign official dollar liabilities: UP

American monetary gold: DOWN

But the path between those two endpoints was noisy enough to make complacency understandable.

And that makes what happened next even more interesting.

Then Triffin showed up

Almost exactly when that arithmetic became impossible to ignore, an economist named Robert Triffin explained the contradiction.

In 1960, he published:

Gold and the Dollar Crisis.

Eleven years before Nixon closed the gold window.

Triffin saw the trap.

The world needed growing reserves to support growing international commerce.

But there wasn’t enough newly mined monetary gold to provide all those reserves.

So the United States supplied the world with dollars.

But every additional dollar reserve held overseas was also another potential claim against America’s finite gold stock.

Stop supplying dollars and eventually the world suffers a shortage of the reserve currency it needs.

Keep supplying dollars and eventually people begin questioning whether all those dollars can really be converted into gold at $35 an ounce.

Damned if you do.

Damned if you don’t.

The Federal Reserve’s historical account now summarizes the contradiction beautifully: without additional dollar reserves Bretton Woods was unworkable; with them, it became unstable. Welcome to the Triffin dilemma.

That isn’t simply bad fiscal discipline.

That’s an architectural problem.

Triffin proposed creating a new international reserve asset that wouldn’t depend on America continually supplying more redeemable dollars. The IMF eventually did exactly that, creating Special Drawing Rights, or SDRs, in 1969.

A new treatment for the problem.

Remember that thought.

And this is where David Deutsch reared his beautiful head

My first thought was:

How could the architects of Bretton Woods possibly have failed to predict this?

And then I realized that was the wrong question.

In 1944, Bretton Woods made sense.

Perhaps tremendous sense.

The world they were trying to repair actually existed in 1944.

Their solution addressed the problems in front of them.

America had the gold.

America had the industrial capacity.

The world needed dollars.

The dollar had enormous credibility.

Fine.

The better question is:

When did they ask whether the solution still made sense?

1944?

Looks great.

1945?

Still working.

1949?

America has almost $25 billion of gold.

Fantastic.

1950?

Hmm.

Gold is now going the other way.

Still fine?

1952?

Some of the gold came back.

Maybe fine after all?

1955?

The drain resumed.

Still fine?

1957?

Gold is rising again.

The Fed still sees an enormous reserve cushion.

Still fine?

1958?

More than $2 billion of gold disappears in a year.

Still fine?

1960?

America’s gold is now below $18 billion.

Official dollar liabilities are climbing rapidly.

The original cushion has collapsed.

Still fine?

At what point does somebody stop and ask:

Will this continue to serve us like it did last year?

That was the Deutsch lesson I couldn’t unsee.

Problems are inevitable.

Problems are soluble.

But solutions are not sacred.

They solve a problem under a particular set of circumstances.

When those circumstances change, you don’t worship yesterday’s solution.

You criticize it again.

Someone did.

Triffin.

And by 1962, the argument had reached the Kennedy White House itself.

An internal memorandum from economist Carl Kaysen laid out a striking disagreement.

Treasury believed the balance-of-payments problem was manageable and worried that large structural changes might create more danger than they solved.

The Council of Economic Advisers took a much less comfortable view: some fundamental change in the mechanism of international payments was necessary.

In other words, by 1962 the argument wasn’t really whether there was a problem anymore.

It was whether the existing system could be patched successfully, or whether the machinery itself needed to change.

That is almost absurdly on-theme.

Treating the treatment

And this is where the story suddenly became uncomfortably familiar to me for an entirely different reason.

I have spent a great deal of time researching what can happen in medicine when clinicians begin treating the consequences of previous treatments.

A loop diuretic such as furosemide can cause excessive electrolyte loss, including hypokalemia. Even the FDA label says so rather plainly.

Another drug, such as potassium-sparing spironolactone, may be added, for that problem or for other cardiac reasons.

But spironolactone can push potassium in the opposite direction. Its current FDA label specifically warns that it can cause hyperkalemia and requires potassium monitoring. That label is here.

If hyperkalemia follows, yet another intervention may be required to bring potassium back down.

One intervention produces a problem.

So another intervention is added to manage it.

The second intervention can produce another problem.

At some point, somebody should ask:

Does the original treatment still make sense?

Now watch what happened under Bretton Woods.

Gold begins leaving America.

Protect the gold.

Foreign governments accumulate more dollars.

Persuade them to keep the dollars.

Private markets push gold above $35.

Defend $35.

Pressure on $35 continues.

Create the London Gold Pool.

Sell gold into the market to help keep gold at $35.

Then, in March 1968, under mounting speculative demand for gold, the London Gold Pool collapses.

So create a two-tier gold market instead: one official price for transactions among monetary authorities and another market price for everybody else.

Dollar outflows continue.

Introduce restraints on foreign lending and investment.

Try to reduce certain overseas expenditures.

Create central-bank swap lines.

Invent new international reserve assets.

There’s our SDR.

Again and again:

Treat the treatment.

The Fed’s own history of the final decade reads like a catalogue of increasingly elaborate attempts to keep the machinery running: swaps, the Gold Pool, interventions and eventually the two-tier gold market. Down the next rabbit hole, if you dare.

But here’s an important distinction.

Policymakers did reconsider the supposedly fixed variable at the centre of the system.

They reconsidered $35.

They just repeatedly concluded that changing it carried enormous political, distributional and confidence risks.

A revaluation could reward countries that had hoarded gold.

Punish countries that had trusted the dollar.

Benefit the Soviet Union and South Africa.

Encourage everyone else to rush for gold before the next revaluation.

Perhaps even trigger the very crisis they were trying to prevent.

So successive administrations repeatedly chose to defend the price and build additional machinery around it instead.

$35.

And that led to my next forehead-slapping question.

Why not just change the price of gold?

Suppose America has 500 million ounces of monetary gold.

At $35 an ounce, those reserves are officially worth:

$17.5 billion.

Now make gold worth $70 an ounce.

Same vault.

Same bars.

Same ounces.

But the official dollar value of that gold is now:

$35 billion.

And somebody presenting $70 receives one ounce of gold instead of two.

You have just cut America’s gold obligation per dollar in half.

Holy shit.

Was it really that simple?

No.

Of course it wasn’t.

But here’s the extraordinary part.

They actually examined it.

Kennedy’s public policy was emphatically to defend the $35 price.

But inside the Kennedy administration, officials nevertheless examined the alternative of increasing the official gold price.

And somebody wrote the following assessment into an official memorandum:

It would solve the short-run problem and increase world liquidity.

Then came all the reasons not to do it.

Seriously. Read the 1962 memo yourself.

The objections were substantial.

A higher gold price would reward countries already holding lots of gold.

It would benefit major gold producers, including the Soviet Union and South Africa.

It could encourage speculation.

It could undermine confidence in the dollar.

And while it might postpone America’s immediate reserve problem, it wouldn’t necessarily repair the underlying international monetary architecture.

So Kennedy defended $35.

Fine.

But then I kept digging.

Seven years later, Nixon’s advisers went considerably further.

In June 1969, Treasury Secretary David Kennedy sent President Nixon a memorandum laying out possible changes to the gold price.

One option was a modest increase of 10 to 20 percent.

The other?

A large increase to:

$70 to $100 an ounce.

I’m not paraphrasing that range.

It’s right there in the actual 1969 memorandum, in the annex discussing a large gold-price increase. Enjoy.

And the stated purpose was almost exactly the arithmetic I had stumbled into decades later:

Re-establish a strong U.S. reserve position.

Satisfy foreign official and private demand for gold.

And create a basis for maintaining gold convertibility and fixed exchange rates.

Holy shit again.

They knew the $35 relationship itself was becoming part of the problem.

There were serious objections.

If everyone believes gold will be worth more dollars next year, why hold dollars today?

Why not convert them into gold before the next revaluation?

A country that had obediently accumulated dollars could be punished while a country that hoarded gold received a windfall.

Major gold-producing countries would benefit enormously.

Congress would have to approve the change.

And if the anchor kept moving, was it really an anchor anymore?

Perfectly legitimate problems.

So simply indexing the gold price upward wasn’t some magical solution overlooked by generations of monetary economists.

The important point isn’t that policymakers never thought of changing $35.

They did.

The point is that once they examined the consequences, they repeatedly decided that defending the existing price looked less dangerous than changing it.

And each time they made that choice, another layer of machinery became necessary to keep the system functioning.

That is a subtler story.

And, I think, a more interesting one.

Then France arrives

Now, finally, we get to Charles de Gaulle.

Not in 1950.

1965.

Look back at the table.

American monetary gold:

$13.806 billion.

Short-term dollar liabilities to foreign official institutions:

$13.066 billion.

The enormous cushion that existed in 1944 had virtually vanished.

By then, the gold drain was old news.

The structural contradiction had been identified.

American policymakers were already devising increasingly complicated ways of protecting the country’s gold reserve.

On February 4, 1965, De Gaulle publicly challenged the privileged position of the dollar and called for a greater international role for gold. The IMF was writing about that speech at the time.

France then became increasingly aggressive about exercising the gold option.

And here’s where one of my own earlier drafts had the date wrong.

The spectacular Treasury memo I found is from June 1966, not 1965.

By the first half of 1966, U.S. gold sales to France were expected to total $324 million.

The memo states bluntly that French policy was to convert its dollar accruals into gold. Even with no new accruals, France intended to convert at least $34 million a month. During that first half, conversions were averaging $54 million monthly.

Here’s the actual Treasury memo. It is only a couple of pages and considerably more entertaining than most monetary history textbooks.

France was not discovering a secret.

France was looking at a weakness that had been developing for more than a decade and saying:

We’ll take the gold.

But there is an important difference between Britain in 1950 and France in the mid-1960s.

Britain’s purchases largely fit into the postwar rebuilding of reserves.

De Gaulle’s challenge was deliberate, public and political.

France did not create the vulnerability, but its challenge helped transform a long-running arithmetic problem into a much more visible political and confidence crisis.

The gentleman who responded to my original comment might still think De Gaulle was an arrogant, ungrateful, good-for-nothing chap.

Perhaps he was.

That’s a different article.

What matters here is that France was exercising a right the United States itself had promised under Bretton Woods.

The problem wasn’t that France asked for the gold.

The problem was that America had reached the point where it could no longer comfortably tolerate everybody asking for the gold.

That is a very different story.

And by then, the private-market defense of $35 had already cracked in 1968.

Nixon’s 1971 suspension of convertibility wasn’t the moment the problem appeared.

It was the moment the United States finally stopped pretending the old promise could still be maintained.

Which brings me back to AI

While I was disappearing down this ridiculous monetary rabbit hole, I was simultaneously having another conversation.

A reader named Fraser had responded to something I wrote about artificial intelligence.

He thought my writing sounded too much like AI.

Fair criticism.

I told him that I do use AI extensively when writing about subjects where I don’t possess all the underlying information.

I supply the questions and ideas.

AI helps me find material and connect dots.

Then I interrogate what comes back.

Check it.

Run it through other systems.

Read the underlying papers.

Rewrite.

Manually proofread.

Delete the damned em dashes.

Fraser described his own process.

It wasn’t terribly different.

And then he wrote something that seemed almost incidental at the time:

“At this time, the questions are almost more interesting than the answers!”

Meanwhile, without realizing it, I was conducting an experiment proving the thesis of the article he had originally commented on.

That article was called:

When Curiosity Becomes Cheap.

My argument wasn’t that humanity had suddenly acquired knowledge for free.

Most of the knowledge I found while wandering through Bretton Woods had been sitting there for decades.

The Federal Reserve recorded Britain’s enormous 1950 gold purchases at the time.

Robert Triffin published his warning in 1960.

American government memoranda documented the gold problem.

Economic historians have written entire books about it.

Researchers were still examining the “Bretton Woods gold puzzle” in 2019.

None of this was hidden.

The information existed.

What changed was the cost of moving through it.

I began with:

Who “prints” money in America?

Then:

How often has America actually balanced its federal budget?

Then:

How could France know too many dollar claims existed?

Then:

Wait.

America had almost $25 billion of gold in 1949 and less than $18 billion by 1960?

Where did it go?

Britain?

In 1950?

Why?

Where did Britain get all those dollars?

Did anybody worry when Britain took a billion dollars of gold?

Wait.

The Fed noticed, but thought it was partly evidence that Europe was recovering?

Then gold came back?

Then left again?

Then came back again?

What’s the difference between a federal budget deficit and a balance-of-payments deficit?

Wait.

America could export more than it imported and still lose gold?

Why would anyone have agreed to Bretton Woods in 1944?

How many foreign official dollar claims existed then?

Wait.

The system actually looked incredibly safe in 1944?

So when did somebody ask whether it was becoming unsafe?

Why did $35 remain fixed?

Why not revalue the gold?

And suddenly I understood more about Bretton Woods than I had ever imagined I needed to know.

And now I cannot unsee it.

Almost every piece of information was already there.

What had collapsed was the cost of asking:

Yeah. But why?

Then asking it again.

And again.

And again.

Somebody else is spending billions of dollars building the machinery that allowed me to follow that curiosity all day, sleep for a while, go run 20 kilometres, and then come back and keep asking questions.

The investors may still get killed.

AI may yet meet its pin.

But this technology just allowed an old retired IT guy in Winnipeg to wander through eighty years of monetary history because one stranger on the Internet thought Charles de Gaulle was a good-for-nothing chap.

And another stranger thought my writing sounded too much like AI.

Both made me ask another question.

For that, gentlemen:

Thank you.

Because the story I thought I knew was:

France called America’s bluff.

The story I know now is much better.

The gold had been leaving for fifteen years before France arrived.

But even that isn’t the part I find most difficult to unsee.

It is this:

Bretton Woods may have been an excellent answer to the world of 1944.

That did not make it an excellent answer forever.

And institutions don’t need to predict the future perfectly.

They need to keep asking whether the assumptions underneath yesterday’s solution remain true today.

Sometimes the warning light will even flicker off and reassure you that everything is fine.

That doesn’t make the question less important.

It makes the question more important.

So perhaps the question worth carrying out of this entire rabbit hole isn’t about gold at all.

It’s simpler.

And applicable to almost everything:

Will the solution that served us yesterday still serve us tomorrow?

Regards,
Luke Kandia
Chasing Clarity

© 2026 Lucas Kandia, All Rights Reserved.

Disclaimer This is not financial advice. The information here is for educational purposes only, and I’m not your financial advisor. While I do my best to provide accurate information, I can’t guarantee it. My opinions might change. Before you make any investment decisions, please talk to a professional advisor and do your own research. You shouldn’t rely only on what you read here.

Chasing Clarity is an independent publication supported by readers who’d rather understand what’s happening than be surprised by it. If this piece made you think, send it to someone who should be thinking about it too.

The other day I published an article called When Curiosity Becomes Cheap.

The other morning Bill Bonner published Wall Street’s Odor of Decay, arguing that the enormous amounts of capital pouring into AI may never earn an adequate return.

Reading it, I had an amusing realization.

We may both be right.

Curiosity is becoming incredibly cheap for me.

It is becoming incredibly expensive for somebody else to make it that way.

That distinction matters.

Two very different economics

When I sit down with AI and chase a question, the incremental cost to me of asking one more question is almost nothing.

I can ask:

Why?

Then:

How?

Then:

What if these two things are connected?

I can interrogate an idea, compare explanations, find weaknesses, cross disciplines and keep going.

Somebody else, meanwhile, has to build the data centres.

Buy the chips.

Generate the electricity.

Build the networks.

Train the models.

Hire the researchers.

Replace the hardware.

And raise extraordinary amounts of capital to do it.

So there are really two AI stories happening simultaneously.

One is the economics of building AI.

The other is the economics of using it.

They need not produce the same winners.

Great technology. Great investment?

We have seen versions of this before.

A technology can transform society while destroying enormous amounts of investor capital along the way.

Railroads transformed commerce even though many railroad investors lost fortunes.

The internet transformed the world despite the wreckage left by the dot-com bubble.

The telecom boom produced enormous overinvestment in fibre-optic capacity, severe financial losses, and infrastructure that remained useful long after the bubble burst.

Federal Reserve Governor Michael Barr recently used both railroads and the early-2000s fibre buildout as historical examples when discussing the possibility of AI overinvestment: the financing can go badly even when the technology eventually proves enormously useful.

The people financing a technological revolution are not necessarily the people who ultimately capture its value.

The fate of the investors and the fate of the technology are two different questions.

AI may be the next great test of that distinction.

Bill’s pin

Bill may be entirely right about the bubble.

Valuations can become absurd.

Capital expenditure can outrun any plausible future cash flow.

Companies can buy from one another, flattering revenues across an entire sector while convincing investors that spending itself is evidence of sustainable demand.

Eventually somebody asks the awkward question:

Where is the return on all this capital?

And the pin arrives.

We may be about to get a remarkable real-world test.

Anthropic, the company behind Claude, raised $65 billion in May at a $965 billion post-money valuation.

Days later, it confidentially submitted a draft S-1 for a proposed IPO. Anthropic has not announced an offering price, the number of shares to be sold, or an IPO valuation.

But the Financial Times reports that some investors expect Anthropic could float as early as October at a valuation of $2 trillion or more.

And the assumptions required to get there are extraordinary.

Reuters, citing sources familiar with Anthropic’s internal financials, reports that bankers and investors examining an IPO are already looking as far ahead as a company projection of roughly $190–200 billion in revenue in 2028.

Anthropic has not publicly confirmed that figure as formal guidance.

The projection may prove accurate.

It may prove wildly optimistic.

Either way, it reveals how much future growth is being contemplated when investors discuss a valuation on this scale.

Heavy spending on compute, training and talent means today’s margins tell only part of the story. Much of the valuation case depends on continued extraordinary growth and substantially better economics later.

That doesn’t mean a $2 trillion valuation is necessarily wrong.

It means an enormous amount of the future has already been pulled into today’s price.

Two trillion dollars.

Before its shares have even begun public trading.

Claude may be an extraordinary tool.

I use it myself.

That tells me almost nothing about whether Anthropic at $2 trillion would be an extraordinary investment.

And that is precisely the distinction.

A $2 trillion valuation is not $2 trillion

It is worth pausing here because the number itself can be misleading.

If Anthropic were valued at $2 trillion in an IPO, nobody would hand Anthropic a cheque for $2 trillion.

Only some portion of the company would be offered. The offering price, considered across all of the company’s outstanding shares, could imply an overall value of roughly $2 trillion even though only a fraction of those shares actually changed hands in the offering.

Once public trading begins, the market price can move that implied valuation higher or lower.

No extra trillion dollars has to materialize.

The important question is:

Can Anthropic eventually earn enough money to justify that valuation?

Not:

Is Claude useful?

Those are entirely different questions.

Value created is not value captured

And this, I think, is the missing piece.

Value created is not the same as value captured.

AI could create trillions of dollars of economic value for the people and businesses using it without the companies building it ever managing to collect trillions of dollars from them.

Suppose an AI system saves a physician hours of research.

Suppose it lets a small company perform work that once required five additional employees.

Suppose it helps a scientist find a connection buried across three different literatures.

Suppose it allows someone like me to chase questions through medicine, economics, physics, history and finance for the price of a monthly subscription.

Real economic value has been created.

But who gets it?

And how much can the AI provider charge for it?

That is the question investors should care about.

If strong competition persists among Anthropic, OpenAI, Google and others, capabilities could keep improving while prices to users fall.

But that outcome is not guaranteed.

If one AI provider becomes the default interface through which millions of people work, research, code, communicate and perhaps even navigate the internet itself, the economics could look very different.

Distribution matters.

Switching costs matter.

Proprietary data may matter.

Deep enterprise integration may matter.

Scale may matter enormously.

A dominant provider could build durable advantages and capture a very large share of the value it creates.

The point is not that users must capture the surplus.

It is that technological usefulness alone does not tell us who ultimately will.

If competition remains fierce, though, the other possibility is just as real:

Models get better.

Prices fall.

Capabilities spread.

Margins get squeezed.

The technology becomes more useful while simultaneously becoming harder to monetize at the level implied by enormous valuations.

The user can win enormously while the investor loses.

Somebody else is paying for my curiosity

This is the part I find fascinating.

Yesterday I argued that AI is radically reducing the cost of pursuing a question.

Today Bill reminded me of the other side of that equation.

Curiosity is becoming cheap precisely because somebody else is spending a fortune to make it cheap.

That is not a contradiction.

It is the business problem.

And, of course, these investors are not doing this as charity.

Venture funds, technology companies, cloud providers and other strategic investors are spending these billions because they expect to capture value somewhere — through subscriptions, enterprise contracts, cloud computing, infrastructure demand, market share, ecosystems or future profits.

Some may succeed spectacularly.

Others may discover that the value they helped create escaped somewhere else.

That possibility matters.

Hundreds of billions of dollars may be spent constructing a technological infrastructure that eventually provides consumers with astonishing capabilities for $20, $50 or $100 a month.

That can create enormous economic value.

It can still produce miserable financial returns for some of the people who supplied the capital.

And Anthropic at a hypothetical $2 trillion valuation makes the distinction wonderfully clear.

Claude could eventually create enormous economic value.

Doctors could use it.

Scientists could use it.

Programmers could use it.

Businesses could use it.

Millions of curious people could use it.

But somebody must ultimately capture enough of that value to justify the capital being committed today.

Maybe Anthropic does.

Maybe Amazon or Google captures part of it through cloud infrastructure.

Maybe companies using AI capture much of it through productivity gains.

Maybe consumers capture an enormous share through falling prices and better tools.

Maybe the value spreads across all of them.

That is the unknown.

Creating value and capturing value are two very different things.

The real question

So there are two questions we should stop confusing.

Is AI overvalued as an investment?

Quite possibly.

And:

Is AI creating enormous value for the people who use it?

Quite possibly.

Both can be true.

If strong competition between AI companies continues, an increasing share of the benefits may flow toward users rather than producers.

The producer spends billions trying to remain ahead.

The customer gets a better product.

The competitor lowers its price.

Another model catches up.

And the extraordinary capability everyone spent fortunes creating begins to look increasingly ordinary to the person paying twenty bucks a month for it.

The investor curses.

The user smiles.

After the bubble

And perhaps this is where Bill’s comparison with earlier technological booms becomes most useful.

The bust does not necessarily kill the technology.

Sometimes it clears away the financial fantasy and leaves something useful behind.

But the historical analogy has limits.

Railroad tracks could remain productive for decades.

Fibre could stay in the ground waiting for traffic that eventually arrived.

AI hardware depreciates much faster.

Today’s frontier chip can become yesterday’s hardware within a few years, and a frontier model may become obsolete even faster than the machines that trained it.

So an AI bust may not leave behind precisely the same kind of durable physical inheritance.

There may not be warehouses full of old GPUs patiently waiting to power the next revolution.

But something else can survive.

The data centres.

The power connections.

The networks.

The software.

The research techniques.

The trained engineers.

The accumulated knowledge about how to build and use these systems.

Cheaper and more widely available compute.

And perhaps most importantly, millions of people and businesses that have learned what to do with the technology.

That is what happened with the internet too.

The internet after 2000 became far more consequential than the internet investors were speculating on in 1999.

The fibre remained in the ground.

But so did something less tangible.

The protocols remained.

The engineering knowledge remained.

The developers remained.

The habits changed.

Prices fell.

New businesses appeared.

AI could follow its own version of that path.

Some of today’s companies may disappear.

Some valuations may collapse.

Some enormous investments may never earn their cost of capital.

Anthropic could become a fabulous company and still prove to have been a terrible investment at the wrong price.

The particular models and chips may become obsolete.

But the capability, complementary infrastructure, accumulated know-how and habits of use need not disappear with them.

And millions of people may continue discovering that they can investigate questions, cross intellectual boundaries and connect ideas at a cost that would have been unimaginable only a few years ago.

The bubble can burst.

The investors can lose.

And the revolution can still win.

Yesterday I wrote that curiosity was becoming cheap.

Today I realized why.

Somebody else is paying a fortune to make it so.

Regards,
Luke Kandia
Chasing Clarity

© 2026 Lucas Kandia, All Rights Reserved.

Disclaimer This is not financial advice. The information here is for educational purposes only, and I’m not your financial advisor. While I do my best to provide accurate information, I can’t guarantee it. My opinions might change. Before you make any investment decisions, please talk to a professional advisor and do your own research. You shouldn’t rely only on what you read here.

Chasing Clarity is an independent publication supported by readers who’d rather understand what’s happening than be surprised by it. If this piece made you think, send it to someone who should be thinking about it too.

Background

The International Agency for Research on Cancer (IARC) Monographs programme first classified alcoholic beverages as carcinogenic to humans (Group 1) in 1987; the dedicated alcohol monograph was published in 1988. Subsequent evaluations added colorectal cancer and female breast cancer to the list of cancers causally linked to alcohol. The established sites are now the oral cavity, pharynx, larynx, esophagus, liver, colorectum, and female breast (IARC, 2024).

Group 1 is frequently misunderstood. It is IARC’s strongest classification of evidence for a carcinogenic hazard: the evidence is sufficient to conclude that an exposure can cause cancer in humans. It does not rank the size of the risk at a particular dose. Two agents in Group 1 can produce very different risks under ordinary exposure conditions. Placing alcoholic beverages in the same evidence category as tobacco smoke or asbestos is therefore a statement about causal evidence, not a claim that the exposures are equally dangerous (IARC Monographs Q&A).

All beer, wine, and spirits contain ethanol. The cancer risk is therefore not confined to a particular beverage: it arises from ethanol, its metabolism, and related biological effects. The causal link has been established for decades, but public awareness and public-health messaging have lagged behind the science.

The French Paradox - and What Happened to the Heart-Health Story

In the early 1990s, the “French Paradox” became a compelling story: France appeared to have relatively low coronary mortality despite a diet that, by the prevailing model, should have predicted more disease. A 1992 paper by Renaud and de Lorgeril helped popularize the idea that wine might contribute to the paradox (Renaud & de Lorgeril, 1992). Red wine, and later compounds such as resveratrol, became part of a broad cultural narrative that moderate drinking could be heart-healthy.

The evidence since then has not produced a simple reversal from “alcohol is protective” to “alcohol is uniformly harmful.” The cardiovascular literature is more complicated than that. Observational studies still often find lower cardiovascular mortality among moderate drinkers than among never-drinkers. A National Academies review published in 2025 concluded, with moderate certainty, that moderate alcohol consumption was associated with lower cardiovascular mortality compared with never drinking. But an association is not proof that alcohol caused the lower mortality. Abstainer bias, former-drinker misclassification, socioeconomic differences, diet, exercise, smoking, and other lifestyle factors remain difficult to eliminate completely. Much of the evidence linking alcohol consumption to health outcomes comes from observational epidemiology, where residual confounding remains possible; genetic and Mendelian-randomization studies provide an important independent line of evidence for evaluating causality.

Genetic approaches point in a less reassuring direction. In a 2022 UK Biobank analysis, adjustment for favorable lifestyle factors attenuated the apparent observational benefit of modest drinking, while Mendelian-randomization analyses suggested increasing risk of hypertension and coronary artery disease as alcohol exposure increased (Biddinger et al., 2022). A large prospective study of roughly 500,000 adults in China likewise found that genetic evidence did not support a protective effect of alcohol on vascular disease (Millwood et al., 2019).

The practical conclusion is narrower than either slogan. It is no longer scientifically sound to recommend alcohol as a strategy for cardiovascular health. At the same time, it is more accurate to say that the apparent cardiovascular benefit remains debated than to say it has simply disappeared.

The often-quoted 2018 Global Burden of Disease analysis concluded that zero alcohol consumption minimized aggregate health loss at the population level (GBD 2016 Alcohol Collaborators, 2018). A 2022 GBD update refined that estimate by age and geography: for younger adults the theoretical minimum-risk exposure remained at or very near zero, while in some older populations the model yielded a small positive amount because possible reductions in ischemic heart disease and diabetes were weighed against other harms (GBD 2020 Alcohol Collaborators, 2022). That refinement does not imply that alcohol is cancer-protective. It illustrates that “minimum total health risk” and “zero cancer risk” are different questions.

Cancer Risk by Consumption Level

The alcohol-cancer relationship is considerably less ambiguous. A meta-analysis of light drinking - defined as up to one drink per day - found increased risk for cancers of the oral cavity and pharynx, esophagus, and female breast, but not statistically significant associations for colorectal, liver, or laryngeal cancer (Bagnardi et al., 2013). Subsequent IARC evaluations and the 2025 U.S. Surgeon General’s Advisory identify at least seven cancer sites with established causal links to alcohol: oral cavity, pharynx, larynx, esophagus, liver, colorectum, and female breast (U.S. Surgeon General, 2025).

Risk generally rises with the amount consumed. For some cancers - particularly breast, mouth, and throat cancers - increased risk is detectable at relatively low levels of consumption. The Surgeon General’s review cites pooled evidence in more than one million women showing about a 10 percent relative increase in breast-cancer risk among women consuming up to approximately one drink per day compared with women who did not drink.

Relative risk can sound larger or smaller than it feels in everyday terms, so absolute risk matters too. The 2025 Surgeon General’s Advisory modeled the lifetime risk, to age 80, of developing any of the seven alcohol-related cancers using an Australian cohort of 226,162 people.

Among women, estimated risk rose from about 16.5 percent for those consuming less than one drink per week to 19.0 percent at one drink per day and 21.8 percent at two drinks per day. Among men, the corresponding estimates were about 10.0, 11.4, and 13.1 percent.

These are population estimates, not personal predictions, and the reference group was very light drinkers rather than lifetime abstainers. They illustrate an important point: at low exposure, the absolute increase in individual risk may be modest, but modest is not the same as zero.

Canada’s 2023 Guidance on Alcohol and Health communicates risk as a continuum: 1-2 standard drinks per week are categorized as low risk, 3-6 as moderate risk, and 7 or more as increasingly high risk. Lower consumption generally means lower alcohol-attributable risk; not drinking avoids additional alcohol exposure.

A note on standard drinks. Definitions vary by jurisdiction. In Canada, one standard drink contains 13.45 g of pure alcohol; in the United States it contains 14 g. WHO uses 10 g in some international surveillance instruments. Unless otherwise noted, quantities in this article follow the definition used by the cited study or guidance.

What Happens When People Cut Back or Stop?

A 2022 Korean cohort study of approximately 4.5 million adults found that reductions in alcohol consumption were associated with lower subsequent risk of all cancers and alcohol-related cancers, particularly among people who began at heavier levels of consumption (Yoo et al., 2022). The study was large, but it was observational: drinking behavior was self-reported, follow-up was relatively short for cancer development, and changes in alcohol use may have occurred alongside other health changes. The authors also observed patterns consistent with the “sick quitter” problem - some people may stop drinking because illness is already developing.

IARC’s 2024 evaluation provides the more conservative summary. Compared with continued drinking, there is sufficient evidence that reduction or cessation lowers the incidence of cancers of the oral cavity and esophagus. For the other established alcohol-related cancers, evidence for risk reduction after cutting back or stopping was limited or inadequate at the time of review (IARC Handbook 20A, 2024). That does not mean reduction has no benefit for the other sites; it means the direct evidence for reversal of risk is not yet equally strong across all seven cancers.

How Alcohol Causes Cancer

Several biological pathways are supported by the literature, but they are not all established with equal precision. The best-supported mechanisms can be described without assuming that every molecular step has been settled.

Acetaldehyde and DNA damage

Ethanol is metabolized to acetaldehyde, a reactive compound that can bind to DNA, form adducts, interfere with DNA repair, and contribute to mutagenesis. This is one of the best-established pathways linking alcohol to cancer. Genetic variation matters: variants that reduce aldehyde dehydrogenase 2 (ALDH2) activity, which are common in East Asian populations, can produce much higher acetaldehyde exposure after drinking and substantially alter cancer risk (Im et al., 2022).

Oxidative stress and inflammation

Alcohol metabolism can generate reactive oxygen species. These molecules can damage DNA, proteins, and lipids and contribute to inflammatory signaling. IARC and the U.S. Surgeon General identify oxidative damage as another major pathway in alcohol-related carcinogenesis.

Hormonal effects

Alcohol can alter circulating sex-hormone concentrations, including estrogen. That pathway is biologically relevant to breast cancer, particularly hormone-sensitive disease. The high-level relationship is well supported; the exact contribution of altered synthesis, aromatization, hepatic metabolism, and other pathways in humans is less completely resolved. That distinction matters: the evidence supports a hormonal mechanism without requiring an overly specific molecular story.

Facilitating exposure to other carcinogens

Alcohol can act as a solvent and can alter tissue and intestinal permeability, facilitating exposure to other carcinogens. This is especially important for tobacco smoke and helps explain the strong combined effect of alcohol and smoking on cancers of the mouth, pharynx, larynx, and esophagus. IARC also identifies changes in intestinal permeability and the gut microbiome as relevant mechanistic pathways. Folate disruption remains a plausible additional mechanism, but it is less central to the causal case than acetaldehyde and oxidative damage.

These mechanisms are consistent with the fact that alcohol can cause cancer in several distinct tissues. They also explain why beverage type does not remove the risk: beer, wine, and spirits all deliver ethanol (IARC Evidence Summary Brief, 2025).

Beyond Cancer: Important, but a Separate Question

Alcohol affects more than cancer risk. Those effects matter in their own right, but they are distinct from the cancer evidence. They are best considered separately rather than presented as if every non-cancer effect necessarily amplifies carcinogenesis.

Alcohol-associated liver disease

Persistent heavy alcohol use can produce a spectrum of alcohol-associated liver disease, from steatosis to alcohol-associated hepatitis, fibrosis, cirrhosis, and hepatocellular carcinoma. Not every “regular drinker” develops these conditions, and progression depends on dose and pattern of drinking as well as sex, genetics, metabolic disease, diet, and other liver disease. Steatosis can improve substantially with abstinence; established cirrhosis is much less reversible (AASLD guidance). The terminology matters too: contemporary hepatology increasingly uses “alcohol-associated” rather than “alcoholic” liver disease to reduce stigma and improve clinical precision.

Nutrition and thiamine

Chronic heavy alcohol use is associated with thiamine and other nutritional deficiencies through reduced intake, impaired absorption, altered storage, and other mechanisms. Severe thiamine deficiency can cause Wernicke encephalopathy and, if untreated, persistent Korsakoff syndrome (Thomson, 2000). These are well-established complications of chronic heavy use; they should not be generalized to occasional or low-level drinking.

Fluid balance

Ethanol can produce an acute, transient diuretic effect, particularly while blood alcohol concentration is rising, but the magnitude varies considerably by dose, beverage composition, accompanying fluid intake, and the person’s starting hydration state (Jones, 1990). Controlled rehydration studies also show that alcohol content and electrolyte content can materially change net fluid retention (Desbrow et al., 2015). It is therefore too broad to say that one or two drinks inevitably cause clinically meaningful dehydration. Fluid and electrolyte disturbances become more relevant with heavier intake, poor oral intake, vomiting, illness, or other vulnerability.

Public Health Messaging Has Shifted, but Not in a Straight Line

The causal evidence long preceded widespread public understanding. In 2018, the American Society of Clinical Oncology issued a formal statement describing alcohol as an established, modifiable cancer risk factor and calling for greater oncology-community engagement (LoConte et al., 2018). Canada’s 2023 guidance then moved away from a simple “moderation is safe” message toward a continuum-of-risk framework.

WHO/Europe stated in 2023 that no safe amount of alcohol consumption can be established for cancer and health, emphasizing that a threshold at which carcinogenic risk suddenly “switches on” has not been demonstrated (WHO, 2023). The phrase is easy to misread. “No safe level” does not mean that one drink creates a large individual risk. It means that current evidence has not identified a positive exposure below which cancer risk is demonstrably zero.

In January 2025, the U.S. Surgeon General issued a dedicated advisory on alcohol and cancer and called for cancer-risk warnings on alcoholic-beverage labels (U.S. Surgeon General, 2025). The 2025-2030 Dietary Guidelines for Americans, published in January 2026, changed the headline framing to: “Consume less alcohol for better overall health” (Dietary Guidelines for Americans, 2026). That change was not simply a tightening of guidance, however. The new edition also removed the previous sex-specific daily limits and was silent on the established alcohol-cancer relationship. The American Association for the Study of Liver Diseases criticized those omissions, arguing that the guidelines had become less specific even as their headline message became more cautionary (AASLD, 2026).

Ireland offers another example of the uneven trajectory. Regulations signed in 2023 require alcohol labels to carry health information including warnings about cancer, liver disease, and pregnancy. Implementation, originally scheduled for May 2026, was later deferred until 3 September 2028 (S.I. No. 423/2025).

Public-health advocates sharply criticized the delay; Alcohol Action Ireland argued that intensive industry opposition had weakened implementation of a measure already enacted into law (Alcohol Action Ireland, 2025). Taken together, these developments show that public-health messaging has shifted toward clearer recognition of alcohol-related cancer risk, but policy has not moved in a single direction.

What the Evidence Supports

• Alcoholic beverages are a known human carcinogen, with causal links established for at least seven cancer sites.

• Cancer risk generally increases with dose, and increased risk is detectable at relatively low consumption levels for some cancers, particularly female breast cancer and several upper aerodigestive cancers.

• Beer, wine, and spirits do not differ in the fundamental carcinogenic exposure: all contain ethanol.

• No positive alcohol-consumption threshold has been established below which cancer risk is demonstrably zero. This is a statement about risk thresholds, not a claim that low-dose absolute risk is large.

• Reducing or stopping alcohol is associated with lower cancer risk in observational data; direct evidence of risk reduction is strongest for oral-cavity and esophageal cancers, while evidence for the other established sites is less complete.

• The cardiovascular literature remains contested. Some observational evidence associates moderate drinking with lower cardiovascular mortality, while genetic studies weaken the case for a causal protective effect. Alcohol should not be recommended as a cardiovascular intervention.

• Alcohol has important non-cancer effects, particularly with heavier or chronic use, including alcohol-associated liver disease, nutritional deficiency, cardiovascular effects, dependence, and acute injury risk. These harms strengthen the case for informed decision-making but should not be conflated with the cancer evidence.

For an individual deciding whether to drink, the relevant question is not whether alcohol is “good” or “bad” in the abstract. It is what level of risk is acceptable in the context of that person’s age, health, medications, family history, social circumstances, and preferences. The role of evidence is to make that tradeoff visible.

On cancer specifically, the message is unusually clear: less alcohol means less exposure to a known carcinogen. At very low levels, the absolute increase in risk may be small. Small, however, is not zero. Good risk communication should be able to say both things at the same time.

Regards,

Luke Kandia
Chasing Clarity

This is a public-facing review of published scientific literature. It does not present original research and has not undergone formal peer review. It is intended for informational purposes and does not constitute medical advice. Individual decisions about alcohol consumption should be discussed with a qualified healthcare professional where appropriate.

Sources and Further Reading

• International Agency for Research on Cancer (2024). IARC Handbooks of Cancer Prevention, Volume 20A: Reduction or Cessation of Alcoholic Beverage Consumption.

• International Agency for Research on Cancer (2025). Alcohol: A Major Preventable Cause of Cancer. IARC Evidence Summary Brief No. 6.

• International Agency for Research on Cancer. IARC Monographs Questions and Answers: classification describes carcinogenic hazard, not the magnitude of risk at a particular exposure.

• Office of the U.S. Surgeon General (2025). Alcohol and Cancer Risk: The U.S. Surgeon General’s Advisory.

• Bagnardi V, et al. (2013). Light alcohol drinking and cancer: a meta-analysis. Annals of Oncology 24(2):301-308.

• Yoo JE, Han K, Shin DW, et al. (2022). Association Between Changes in Alcohol Consumption and Cancer Risk. JAMA Network Open 5(8):e2228544.

• Im PK, Yang L, Kartsonaki C, et al. (2022). Alcohol metabolism genes and risks of site-specific cancers in Chinese adults: An 11-year prospective study. International Journal of Cancer 150:1627-1639.

• GBD 2016 Alcohol Collaborators (2018). Alcohol use and burden for 195 countries and territories, 1990-2016. The Lancet 392:1015-1035.

• GBD 2020 Alcohol Collaborators (2022). Population-level risks of alcohol consumption by amount, geography, age, sex, and year. The Lancet 400:185-235.

• Renaud S, de Lorgeril M. (1992). Wine, alcohol, platelets, and the French paradox for coronary heart disease. The Lancet.

• Biddinger KJ, et al. (2022). Association of Habitual Alcohol Intake With Risk of Cardiovascular Disease. JAMA Network Open 5(3):e223849.

• Millwood IY, et al. (2019). Conventional and genetic evidence on alcohol and vascular disease aetiology: a prospective study of 500,000 men and women in China. The Lancet 393:1831-1842.

• National Academies of Sciences, Engineering, and Medicine (2025). Review of Evidence on Alcohol and Health.

• Canadian Centre on Substance Use and Addiction (2023). Canada’s Guidance on Alcohol and Health.

• World Health Organization Regional Office for Europe (2023). No level of alcohol consumption is safe for our health.

• LoConte NK, et al. (2018). Alcohol and Cancer: A Statement of the American Society of Clinical Oncology. Journal of Clinical Oncology 36(1):83-93.

• American Association for the Study of Liver Diseases. Alcohol-Associated Liver Disease guidance.

• Thomson AD. (2000). Mechanisms of vitamin deficiency in chronic alcohol misusers and the development of the Wernicke-Korsakoff syndrome. Alcohol and Alcoholism 35(Suppl 1):2-7.

• Jones AW. (1990). Excretion of alcohol in urine and diuresis in healthy men in relation to age, dose and time after drinking. Forensic Science International 45(3):217-224.

• Desbrow B, et al. (2015). Manipulations to the alcohol and sodium content of beer for postexercise fluid replacement. International Journal of Sport Nutrition and Exercise Metabolism 25(3):262-270.

• U.S. Department of Health and Human Services and U.S. Department of Agriculture (2026). Dietary Guidelines for Americans, 2025-2030. 10th Edition. January 2026.

• Government of Ireland (2025). S.I. No. 423/2025: Public Health (Alcohol) (Labelling) (Amendment) Regulations 2025.

• American Association for the Study of Liver Diseases (2026). AASLD Raises Concern Over Removal of Evidence-Based Alcohol Guidance in 2025-2030 Dietary Guidelines for Americans.

• Alcohol Action Ireland (2025). Consequences of alcohol labelling delay will be felt by ordinary Irish people every day. 23 July 2025.

Bill Bonner wrote something today that got me thinking.

His argument, stripped down, is that artificial intelligence has no privileged access to truth. Large language models are built from human words, human ideas, human judgments and human mistakes. AI may be clever, but it is still standing on a mountain of material produced by us.

Fair enough.

But I think that misses something important.

Perhaps profoundly important.

Medicine is divided into specialties. Science into disciplines. Universities into departments. Research into journals. Economics, history, engineering, biology, physics, politics — each develops its own vocabulary, experts, literature and assumptions.

The knowledge may exist.

The connection between two pieces of knowledge may exist.

But nobody necessarily sees it.

Because nobody is looking at both.

AI doesn’t particularly care which department owns the information.

And that may change far more than we realize.

The old cost of curiosity

I grew up with the World Book Encyclopedia.

If something caught my attention, that was where I started.

World Book could take me surprisingly far. But eventually I would reach the edge of the entry.

Then came the library.

The library could take me farther still, assuming I knew what to look for and could find the right book.

Eventually, though, serious curiosity often ended at a textbook.

And textbooks were formidable things.

Six hundred pages. Eight hundred pages. Dense terminology. Long chapters establishing concepts you might need eventually but didn’t particularly care about now.

Somewhere inside that book might be the answer to the question that had originally fascinated you.

But there was a substantial admission price:

Time. Background knowledge. Patience. Access. The ability to understand the terminology. The ability to figure out which parts mattered.

And quite often, the question simply died.

Not because you stopped being curious.

Because pursuing it cost too much.

That is what AI has changed for me.

I can begin with the interesting part.

Why?

Then:

How?

Then:

What would happen if...?

Then:

Wait. If that’s true, wouldn’t this other thing happen too?

And suddenly we are six levels deeper than the original question.

Sometimes much deeper.

I have had more than a few conversations that started with an innocent question and ended, much later, with me staring at the screen thinking:

Holy shit.

The remarkable thing is not merely that AI supplied information.

It removed friction.

The textbook hasn’t disappeared

There is an obvious danger here.

Sometimes the boring parts matter.

Sometimes the chapter we would rather skip contains the definition, mathematical constraint, methodological detail or inconvenient evidence that destroys our beautiful idea.

AI cannot make rigor optional.

If anything, easier hypothesis generation makes rigor more important.

Eventually you may still have to read the paper. Check the data. Understand the experimental design. Look at the primary source. Do the math.

But the route to that material has changed.

Instead of:

curriculum → facts → understanding → interesting question

we can increasingly work like this:

interesting question → explanation → missing concepts → targeted reading → better question

That is a very different way of learning.

The textbook becomes less of an obstacle course and more of a tool.

Instead of starting at page one, we can start near the intellectual frontier of the particular question that interests us, then work backward into the foundations we actually need.

And, crucially, we can interrupt.

A book cannot respond when paragraph three makes you say:

“Hold on. Why?”

AI can.

Then I thought about a cardiologist

A few days ago I was with an older relative at her cardiologist.

She had been taking a medication that was leaving her very tired in the morning. He switched her to something else, and we discussed how she was doing.

At some point I asked whether he might be willing to read a paper I have been working on about diuretics, dehydration and the difficulty of determining volume status in older patients.

He said he would love to.

But he barely has time for his family.

Never mind finding time for a paper on a complicated subject he hadn’t expected to encounter that day.

And looking at him, I believed him.

He is younger than I am.

He looked exhausted.

Think for a moment about the information burden carried by a physician like that.

Patients. Medications. Blood tests. Echocardiograms. Guidelines. Drug interactions. Referrals. Follow-ups. Chart notes. New research. Old research. Administrative work.

And behind all of it, the knowledge that the decisions being made aren’t academic.

There are human lives attached to them.

Then somebody like me wanders into the office and says:

“Here. I think there may be something interesting in this paper.”

The problem isn’t that he doesn’t care.

The problem is bandwidth.

Now imagine giving that physician an AI system capable of reading the paper first.

Not deciding whether it is correct.

Not practicing medicine for him.

Not stamping the hypothesis VERIFIED.

Simply saying:

Here are the five claims most relevant to cardiology.

Here is the evidence offered for each.

Here are the assumptions.

Here are the weak links.

Here is where the argument agrees with or challenges existing literature.

Here are the strongest counterarguments.

Here are the three sections worth reading yourself.

Here are the primary papers behind them.

Now he can ask:

“Why does the author think this?”

Then:

“Show me the evidence involving patients over 80.”

Then:

“What about those with chronic kidney disease?”

Then:

“Show me the strongest evidence against the hypothesis.”

Then:

“What would we have to measure to determine whether this is actually happening?”

He hasn’t verified the hypothesis.

He hasn’t reproduced the investigation that created it.

He doesn’t need to.

AI has compressed the route sufficiently for him to decide whether the destination is worth examining.

Instead of rebuilding the argument from raw material, he gets to begin with a better-formed hypothesis and interrogate it.

The hypothesis itself must still survive contact with reality.

That may require prospective observation, better measurement, an experiment, replication or years of accumulated evidence.

AI doesn’t remove that.

What it removes is much of the excavation required merely to get there.

That isn’t replacing his judgment.

It may be restoring his ability to use it.

Compression may matter as much as intelligence

This is where I think much of the conversation about AI goes wrong.

We keep asking whether AI is intelligent.

Whether it has judgment.

Whether it knows truth.

Whether it can replace professionals.

Those are important questions.

But perhaps another capability matters just as much:

compression.

That changes the economics of thought.

The physician doesn’t need another machine confidently telling him what to do.

He needs help getting 40 relevant pages down to four paragraphs — with the important qualifications intact — so that his scarce attention can be spent where it has the greatest value.

There is an instructive historical precedent.

The printing press didn't make Aristotle's ideas any truer.

It didn’t remove error, superstition, bad ideas or outright nonsense from human thought.

What it changed was the cost of reproducing ideas.

And once copying became cheaper, other things became easier too. Texts could be compared. Arguments could be checked against other arguments. Ideas could travel farther. More people could enter the conversation. Knowledge that had once been geographically or institutionally separated could increasingly be placed side by side.

The printing press made ideas cheaper to reproduce.

AI makes them cheaper to traverse.

It changes the cost of navigating, interrogating and recombining knowledge in something approaching real time.

The printing press didn’t improve the stock of human knowledge.

It transformed the flow of it.

AI may be doing the same thing at another layer: making knowledge dramatically easier to navigate, interrogate and recombine.

Bill looks at the accumulated corpus on which AI rests and quite reasonably sees the trash heap mixed in with the treasure.

But perhaps he is looking at the stock and missing the flow.

He can be entirely right about the trash heap and still miss the revolution.

The important change may not be what knowledge AI contains.

It may be what becomes possible when the cost of moving through, comparing, questioning and recombining that knowledge collapses.

A historian can cross-reference economics.

An engineer can wander into physiology.

A retired businessman can start asking questions about renal function, blood rheology, endothelial shear stress, monetary history, nuclear physics or whatever else happens to catch his attention that morning.

The traditional barriers remain.

Expertise still matters. Evidence matters. Experience matters.

And some knowledge is not written down at all.

The tacit judgment that comes from years of practice — knowing what looks wrong, what matters, what usually fails, what doesn’t quite fit — does not compress nearly as easily as text.

AI can help a non-specialist reach the edge of a question much faster.

It cannot instantly confer twenty years of experience standing at that edge.

The walls have become much lower.

They have not disappeared.

The economics of curiosity

Somewhere in thinking about all of this, one sentence stopped me:

What happens to human curiosity when the cost of pursuing a question collapses?

That may be the thing.

For most of human history, curiosity was expensive.

Every question carried a tax.

You needed access to information, then you needed to find it, understand its language, determine what mattered, locate the next source and perhaps find an expert willing to explain what you still didn’t understand.

Most questions never survived the process.

They evaporated somewhere between:

“That’s odd...”

and:

“I wonder why?”

AI has dramatically reduced that tax.

Not eliminated it.

Reduced it.

And when the cost of pursuing curiosity falls, more questions survive.

More questions surviving means more connections.

More connections mean more hypotheses.

More hypotheses mean more opportunities to discover that something everyone has been looking at separately may belong together.

But there is an important asymmetry.

AI has made verification cheaper too.

A literature search that once required days can sometimes be done in hours. Claims can be traced to sources. Conflicting papers can be found more quickly. Several explanations can be compared. A model can even be asked to attack its own conclusion.

That is real progress.

But hypothesis generation has become cheaper even faster.

We can now ask:

“Could A explain B?”

a hundred times before lunch.

Many of those ideas can also be rejected quickly.

But establishing that one of the survivors actually reflects reality may still require careful measurement, primary evidence, specialist expertise, experimentation, replication or time.

So the important change isn’t that verification remains expensive while curiosity becomes free.

It’s that the ratio has shifted.

The cost of generating plausible explanations is collapsing faster than the cost of proving them.

That creates abundance.

And a new bottleneck.

AI can produce an elegant bridge between two islands that aren’t actually connected.

It can also faithfully summarize five papers that all share the same mistaken assumption.

Asking five AIs the same question may help reveal whether one model misrepresented a source.

It does not magically turn five models trained on overlapping human knowledge into five independent experiments.

Consensus is not the same thing as truth.

So discernment matters enormously.

Is this connection interesting?

Is it merely plausible-sounding?

What evidence would change my mind?

What observation could prove this wrong?

Are my sources actually independent?

When do I stop asking the machine and measure something in the real world?

Those may become some of the most valuable questions of the AI age.

For generations, some of the friction in learning acted as an accidental brake. You couldn’t race very far ahead of your understanding because sooner or later the textbook, the mathematics or the laboratory stopped you.

AI removes much of that natural braking mechanism.

That is wonderful for curiosity.

But it may mean we need to develop deliberate brakes where friction once supplied them accidentally.

Not barriers to asking questions.

Barriers to becoming prematurely certain of the answers.

AI doesn’t eliminate human judgment.

It may increase the demand for it.

The machine can generate possibilities almost without limit.

Someone still has to decide which possibilities deserve to survive.

The human part

There is another reason I don’t think this ends with AI doing our thinking for us.

AI needs something from us.

Lived experience.

The odd observation.

The thing that bothered us.

The memory of an exhausted cardiologist.

The question that didn’t fit.

The instinct that two things somehow belong beside each other.

I can feed an AI those experiences.

And something interesting happens.

It can hold the pieces together and look for structure.

I supply the dots.

It helps draw lines between them.

Then I look at the picture and decide whether the lines make sense.

That leads to another question, which becomes another dot, then another connection.

The process begins to look less like:

AI thinks → human receives

and more like:

human notices → AI connects → human judges → AI investigates → human notices something new

Round and round.

That is a very different relationship.

Stranded Human Curiosity

And this is where the potential becomes enormous.

There must be millions of people carrying questions they never pursued.

Not because they lacked intelligence.

Not because the questions were foolish.

Because they weren’t doctors, physicists, economists, engineers or academics.

Because they didn’t know the terminology.

Because they didn’t know where to start.

Because nobody had time to explain it to them.

Because opening an 800-page textbook seemed like a ridiculous commitment merely to satisfy a suspicion that something didn’t quite add up.

We often talk about stranded capital, stranded assets and stranded talent.

I wonder how much stranded human curiosity has accumulated through history.

Questions were noticed and abandoned, connections glimpsed but never pursued. Nurses, mechanics, farmers, patients, technicians, businesspeople and grandparents understood one small corner of the world but lacked the vocabulary, access, credentials or time to follow the question into another.

We have no way to measure how much potentially useful thought died there.

AI may not democratize expertise.

Expertise takes work.

Experience still takes time.

Reality does not waive its standards merely because we can ask better questions faster.

But AI may democratize something that comes before expertise.

The ability to begin an investigation.

That matters.

Because now someone can start with four words:

This doesn’t make sense.

And keep going.

They can ask the next question.

And the next.

And the next.

They can discover what terminology they were missing, find the field that owns the problem, encounter the counterargument and learn enough to realize their original idea was wrong.

Or, occasionally, discover that the odd little thing they noticed really does deserve another look.

Perhaps the coming awakening isn’t that millions of people discover artificial intelligence.

Perhaps it is that millions of people discover their own ability to investigate things.

People who spent their lives thinking:

“I’m interested in this, but I’m not a doctor.”

“I’m not a scientist.”

“I wouldn’t know where to begin.”

may suddenly realize that beginning is no longer the hard part it once was.

And somewhere far downstream from the innocent question they originally asked, they may find themselves staring at the screen thinking:

Wait. I can actually think about this.

Bill may be right that AI possesses no divine source of truth.

It doesn’t.

Neither do we.

But that may not be the most important thing happening.

The revolution may not be that machines have acquired judgment.

It may be that ordinary human curiosity has suddenly acquired extraordinary leverage.

The stock of human knowledge remains messy, contradictory, brilliant, foolish, inspired and wrong.

Just as it always was.

What has changed is our ability to move through it.

To interrogate it.

To connect it.

To challenge it.

And to follow a question far enough that another question appears.

David Deutsch argues in The Beginning of Infinity that the growth of knowledge has no obvious endpoint: explanations lead to new problems, and answers make previously impossible questions possible.

AI doesn’t change that process.

What it may change is how many more people can participate in it — and how quickly they can reach the next question.

We have no idea yet what happens when millions of people discover that they can afford to take their own curiosity seriously.

But I suspect we’re barely on page one.

Regards,

Luke Kandia

Luke Kandia writes Chasing Clarity — one man’s search for a better understanding of the world, both small and large, around him.

September 24, 1974. Inside Fort Knox.

There’s a photograph of that day that still gets reprinted every time this story resurfaces, including in the coverage of Trump and Musk’s audit push last year — Mary Brooks, Director of the US Mint, arm raised, grinning up at a wall of stacked gold bars floor to ceiling, a congressman peering up beside her. What makes it remarkable isn’t really what’s in it. It’s that it exists at all. Photography had never once been permitted inside Fort Knox before that day. Brooks told the small group of reporters and congressmen filing in beside her, “We’ve never done this before and we’ll probably never do it again.” She was almost right. It took 43 years for anyone to be let back in.

I want to tell you what actually happened that day, and why — because it turns out the entire modern tradition of “let’s go check if the gold is really there” traces back to one specific, never-substantiated claim from a single man. But first, some context.

Last week, I wrote about the silver that quietly vanished from Canadian and American pocket change in the 1960s, and about the day in 1971 when Richard Nixon severed the dollar from gold entirely.

Then I wrote about what Canada did to its own gold reserve after that — sold every ounce of it, across thirty-six years and six governments, while sitting on some of the richest gold-bearing ground on the planet. Canada’s vault is verifiably, provably empty. Zero tonnes. We checked. Everyone agrees.

Today I want to tell you about the other vault. The one that’s supposed to still be full.

Because here’s the thing that stopped me cold when I went looking: nobody has actually verified that it is. Not recently. Not thoroughly. Not in fifty-two years.

What’s supposed to be there

The United States officially holds 8,133 tonnes of gold — the largest declared reserve of any single nation on Earth, more than double the holdings of Germany, the next-largest official holder. It’s split across four locations:

• Fort Knox, Kentucky — the big one, roughly 147.3 million troy ounces, about 4,580 tonnes.

• West Point, New York — about 54.1 million ounces.

• Denver, Colorado — about 43.9 million ounces.

• The Federal Reserve Bank of New York — a smaller slice, about 13.4 million ounces of Treasury-owned gold, though the vault as a whole is famous mainly because most of what’s actually stored there belongs to other countries, including roughly a third of Germany’s entire gold reserve, kept in a chamber more than eighty feet below Manhattan.

Together, Fort Knox, West Point, and Denver hold about 95% of the gold the US government says it owns.

Worth pausing on that fourth location, because it raises an obvious question: why would other countries keep their gold sitting in someone else’s basement in the first place?

A few real reasons, and none of them are naive. The New York Fed’s vault, carved into Manhattan bedrock in the 1920s, was and largely still is one of the most secure, purpose-built gold-storage facilities on Earth — cheaper for a country to rent space in an existing world-class vault than to build and defend an equivalent of its own.

A huge share of this gold moved to New York during and after the Second World War specifically because European vaults were genuinely at risk, either from invasion or postwar instability, and a country that hadn’t been fought over on its own soil looked like the safer bet. Under Bretton Woods, with so much international trade and settlement running through the dollar, having gold physically in New York also made the accounting simpler — no need to ship bars across an ocean every time a balance needed adjusting. And for much of the Cold War, storing gold with a close, militarily and financially stable ally functioned as a kind of insurance against a home country’s own political instability.

None of that was foolish at the time. But it depended entirely on trust that never had to be tested — and once someone actually tested it, the arrangement started to look different. That's exactly what happened in 2013, when Germany announced a plan to bring home 674 tonnes of gold: all 374 tonnes it held in Paris, plus a portion — 300 tonnes — of a much larger holding in New York. It wasn't a full withdrawal. London's holdings were left untouched from the outset, and so was the bulk of what sat in Manhattan. The move finished in 2017, three years ahead of schedule. Germany's own central bank never claimed anything was missing. It simply wanted its own eyes on more of the gold that officially, legally, had always belonged to it — while consciously choosing to leave roughly a third of its total reserve exactly where it already was. That's the third still sitting in New York today. Keep that instinct in mind — it shows up again later in this piece, closer to home.

It's not entirely settled history, either. As recently as January 2026, German politicians and the Bundesbank's own former head of research began renewing calls to bring home what's left, citing concerns about the unpredictability of U.S. politics under President Trump and the risk of assets abroad being frozen. Not everyone in Germany agrees it's wise — the Ifo Institute's president has warned that pulling the rest out could inflame diplomatic tensions for its own sake. Either way: the same question that got partially answered in 2013 is still, quietly, being asked again.

This is an almost unimaginable pile of wealth at today’s prices. On the government’s own books, though, it’s still priced like it’s 1973.

The price that time forgot

The Treasury values its gold, by law, at $42.22 per ounce — a figure fixed in 1973 and never once updated since.

At that statutory price, America’s entire gold reserve is worth about $11 billion.

At the market price of gold in 2026 — north of $4,000 an ounce — the same reserve, all 261.5 million ounces of it, is worth north of $1 trillion. Fort Knox alone, the single largest piece of it, accounts for roughly $600 billion of that on its own.

Read that gap again. The United States government is carrying its single largest physical asset on its own books at roughly 1% of its actual market value, and has been for over fifty years. Nobody revalues it. Nobody’s required to. It just sits there, frozen at a Nixon-era number, quietly understating the balance sheet of the most powerful economy in the world.

That alone would be a strange enough story. It gets stranger.

When was it last actually checked?

This is the part that surprised me most, and I want to be precise about it, because the internet is full of overheated claims in both directions.

Fort Knox, the flagship vault, has had exactly one comprehensive independent physical audit in its entire history — in 1953, under President Eisenhower. That’s it. That’s the only time anyone outside a small circle of officials did a full accounting.

In September 1974, President Ford allowed a small group of congressmen and journalists into the facility for what was billed as a verification — but by most accounts it covered only a small fraction of the bars, not anything close to a full count. That’s the visit behind the photograph up top, and it’s worth knowing why it happened at all, because it wasn’t a routine oversight exercise.

Accounts of the specifics vary depending on the retelling: some describe him alleging that British intelligence sources had told him the gold was gone; others describe a claim that the Rockefeller family had organized an Army convoy to move it out through Mexico and on to Switzerland for private sale.

What’s consistent across every version is the evidentiary basis behind it: none.

No documents, no named corroborating source, nothing beyond one former official’s public allegation. It was enough anyway. The claim spread through the press, Representative Philip Crane started pressing Treasury Secretary William Simon on where the gold actually was, and Simon’s answer was to invite Congress in to look for themselves. Nine members of Congress and a few dozen journalists got their limited tour that September; the GAO ran a partial audit afterward, covering roughly a fifth of the bars; Representative John Rousselot delivered the verdict that’s basically held ever since: “I think it’s there.”

A smaller, similarly limited visit happened again in 2017, when Treasury Secretary Steve Mnuchin toured the facility with Kentucky lawmakers — a photo opportunity more than an audit, and not independently verified either. Nobody has been let in for anything resembling a comprehensive count since.

West Point and Denver get folded into the same annual process as Fort Knox: an internal Treasury Inspector General review that inspects the locks and seals on the storage compartments to confirm nothing’s been physically tampered with. The Inspector General has stated that, by the end of fiscal year 2008, every deep-storage compartment across all three Mint facilities had been physically inventoried at least once, with official joint seals applied afterward — a real, if one-time, accomplishment. Since then, the annual reviews have checked those seals for signs of tampering rather than reopening the compartments to recount or reweigh anything inside.

The New York Fed is the one partial exception. In 2012, the Treasury Inspector General and the US Mint actually pulled 367 physical samples and sent them to an independent lab for purity testing. Nearly all of them came back within industry tolerance of the recorded purity. That’s a real, if narrow, external check — the only one of its kind at any US gold facility in the modern era. Since then, even the New York Fed has gone back to lock-and-seal checks only.

So the honest ledger looks like this: one facility, tested once, fourteen years ago, on a sample basis. Three others — holding 95% of the gold — verified only by checking that the padlocks haven’t been cut, going back to 1974 for anything more.

Why “checking the locks” isn’t actually enough

Here’s the part that should worry a careful reader more than any of the dates above: even a full weighing, done properly, wouldn’t necessarily catch the thing that’s actually gone wrong with gold elsewhere in the world.

Tungsten has a density of 19.25 grams per cubic centimetre. Gold’s is 19.32. That gap is small enough that a bar with a tungsten core and a genuine gold shell weighs correctly, feels correctly, and passes a basic scratch or surface test — it looks solid gold straight through, because on the outside, it is.

This isn’t a hypothetical.

In September 2012, a Manhattan dealer named Ibrahim Fadl bought four 10-ounce bars for roughly $70,000, grew suspicious, and drilled into them — finding gray tungsten cores under the gold.

Similar bars turned up in England around the same time, and a German refiner, W.C. Heraeus, reported receiving a tungsten-filled 500-gram bar from an unnamed bank in 2010. A 2012 Reuters investigation traced the pattern back further, to counterfeit 400-ounce “good delivery” bars first discovered upon arrival in Hong Kong in 2009, having been shipped from the United States.

The private bullion market took that seriously. Sealed, tamper-evident packaging with certified assay cards, serial-number tracking against a mint’s own database, ultrasonic testing (gold and tungsten transmit sound at very different speeds), and neutron activation analysis all became standard tools specifically because “it weighs right and looks right” turned out not to be good enough.

Which is exactly why the checks Fort Knox actually relies on are the wrong checks for this particular risk. Confirming a seal hasn’t been cut tells you nobody has opened the compartment. It tells you nothing about what was inside a bar the last time anybody looked closely enough to know for certain — and the only method that could tell you that, sampling and lab-testing the actual metal, has been done exactly once, at exactly one of the four facilities, in 2012. Fort Knox, holding well over half the country’s entire reserve, has never had that test run on it at all.

But the deeper problem was never really which test would catch a fake bar. It’s that nobody with the power to order any test — the right one or the wrong one — has ever insisted on it long enough to make it happen. Which brings us to the one moment in fifty-two years when the loudest, most powerful people in the country actually tried.

“All the gold is there”

To be fair to the people currently in charge of it: Treasury Secretary Scott Bessent has repeatedly and publicly stated that the government conducts a private internal audit every year. In February 2025, he told Bloomberg the most recent one was dated September 30, 2024, and said plainly, “all the gold is there.” He’s repeated versions of that assurance since, without visiting Kentucky himself, while offering to arrange a visit for any senator who asks. Steve Mnuchin said much the same during his 2017 tour. Nobody with access to the vaults has publicly claimed anything is missing, and there is no credible evidence that it is.

I want to say that plainly, because this isn’t a “the gold is gone” article. It almost certainly isn’t. Stealing thousands of tonnes of bullion out of the most secured storage facilities in the country, undetected, across five decades, would be one of the great unsolved crimes in human history, and Occam’s razor says that didn’t happen.

But “I’m sure it’s fine” and “we verified it’s fine” are two completely different sentences, and for over half a century, America has only ever offered the first one.

Three days up. Three weeks down.

This isn’t a new complaint. Ron Paul was raising it in Congress years ago, and Germany’s own repatriation back in the early 2010s already showed the confidence issue doesn’t stay hypothetical forever.

But watch how fast it moved in 2025, because the speed is the whole point.

It started with a post on X on February 16, suggesting Elon Musk — then heading the Department of Government Efficiency — take a look at Fort Knox. Musk replied, “Surely it’s reviewed at least every year?” Senator Rand Paul jumped into the same thread: “Nope. Let’s do it.” Senator Mike Lee joined in with the line that’s already in this piece — a decade of trying to get into Fort Knox and being told, flatly, “you can’t come to Fort Knox, because you can’t.” Musk kept going from there, posting a string of claims with no evidence behind them at all: “Who is confirming that gold wasn’t stolen from Fort Knox? Maybe it’s there, maybe it’s not. That gold is owned by the American public!” He topped it off with a meme implying the gold was already gone.

Three days later, on February 19, President Trump backed the effort publicly, telling reporters aboard Air Force One: “We’re going to go into Fort Knox to make sure the gold is there. If it’s not, we’re going to be very upset.” A casual social-media exchange had become a presidential commitment in seventy-two hours.

Representative Thomas Massie introduced the Gold Reserve Transparency Act that year, which would mandate a genuine independent audit every five years. It still hasn’t passed.

And then, just as fast as it had spun up, it stopped. Treasury Secretary Scott Bessent told the public the gold was already accounted for, citing that internal September 2024 report. Trump and Musk went quiet on the subject almost immediately. Musk left DOGE not long after — departure that was always coming, since he was a temporary government employee, but it also removed the one person who’d actually been driving the demand. The story simply faded from the news cycle. As former investment banker James Rickards put it a year later: Trump announced plans to audit, Musk offered to video it, Bessent said it’s all there, and the two most powerful men in the country who’d just spent a week publicly questioning whether America’s gold reserve was real stopped talking about it entirely.

It isn’t fully dead. Trump brought it up again in a May 10, 2026 interview, saying he still wants to “knock on the door” of Fort Knox. Then, three weeks later, the story flared up a third time — and this round has a detail almost too fitting to be believed. On May 27, 2026, the FBI raided the Virginia home of David Rush, a former senior CIA official, and seized 303 gold bars worth roughly $40 million, along with $2 million in foreign currency and 35 luxury watches. Rush is accused of stealing public money — allegedly requesting gold and currency through his own agency in amounts that couldn’t be fully accounted for, on top of separately defrauding the government of about $77,000 through a falsified military-leave claim. Trump seized on it immediately, posting “TIME TO AUDIT FORT KNOX” on Truth Social that weekend, arguing that officials “steal a lot” and that Americans deserve to know the reserves are actually intact. Advocacy groups pushing the Gold Reserve Transparency Act cheered the renewed attention and urged Congress to finally move on it.

There’s no public evidence connecting Rush’s gold to Fort Knox at all — his bars appear to be his own, illegally acquired, nothing more. But sit with the shape of the moment anyway: the government’s answer to “can we trust that officials aren’t quietly walking off with gold that isn’t theirs” was, once again, a public demand to audit its own unaccounted gold — prompted not by a governance process, but by a single arrest that had nothing to do with Fort Knox at all. As of this writing, still no date has been set. Massie’s bill is still stalled. It’s the same three-day spark, the same eventual silence, for the third time in fifteen months

Massie’s and Lee’s asks, underneath all the noise, were both asking for the same modest thing Germany simply went and got for itself back in 2013: not proof anything is wrong. Just eyes on it.

That pattern — from casual tweet to presidential backing in three days, and from presidential backing to total silence in about three weeks — is, in its own way, the most interesting part of the whole story. It shows exactly how fast institutional urgency can appear, and exactly how little it takes to make it disappear again.

This is the same failure. It just wears a different coat.

I keep coming back to David Deutsch’s distinction between static and dynamic institutions — the idea that what separates a system that stays healthy from one that quietly rots isn’t intelligence, it’s whether anything inside it is built to periodically ask “is this still true?” and actually act on the answer.

Canada’s gold policy was a static system that kept re-confirming an old decision, year after year, decade after decade, without ever revisiting it — even as the world around that decision changed completely.

Fort Knox is a static system too. It’s just the mirror image. Canada’s failure was a policy that kept getting renewed. America’s failure is a question that keeps getting asked, loudly, and then dropped — 1974, 2025, and now 2026 are the same story wearing different years. Each time, public pressure built fast. Each time, it produced one brief, symbolic gesture — Mary Brooks opening the doors for cameras, Bessent citing an internal report nobody outside Treasury has seen, a Truth Social post following an arrest that had nothing to do with the vault at all — and each time that single gesture, or lack of one, was enough to close the question again until the next spark... Ford let cameras in once, in 1974. Bessent gave one assurance, once, in 2025. Trump has now demanded an audit three separate times without one ever happening. Fifty-two years, and counting, of "we already checked" — or, increasingly, of not even bothering to say that much.”

An institution doesn’t need to be corrupt to fail this way. It just needs a good enough excuse the first time, repeated often enough that repeating it starts to feel like evidence.

Three coins, three failures

If you’ve read all three of these pieces, here’s the pattern sitting underneath all of them:

The dimes — silver removed deliberately, all at once, by people who looked hard at a real problem and made a defensible, examined choice. Not a failure of correction. Just a policy.

Canada’s vault — gold removed slowly, over thirty-six years, by an institution that made one decision in 1980 and then simply never revisited it, through six governments, while the facts on the ground changed completely underneath the policy.

Fort Knox — gold that’s probably all still there, sitting behind a number that hasn’t been checked, meaningfully, since before I was old enough to vote, protected less by any actual verification than by the simple fact that nobody with the power to look has been willing to insist on it.

Three different countries, three different decades, three different failure modes. All of them come down to the same question, asked or unasked: does anyone in this system have to prove, on a regular basis, that what they’re claiming is still true?

Where the answer is yes, you get an institution that can be trusted. Where the answer is no, you get a number on a balance sheet that nobody’s actually seen the bottom of since 1974 — sitting quietly under fluorescent lights in Kentucky, worth either $11 billion or well over $1 trillion depending entirely on which fifty-year-old assumption you’re willing to keep believing.

There’s a fourth number worth asking about, too — a stranger one than any of these. It’s not about whether something exists. It’s about whether “value,” on a government balance sheet, means anything at all. But that’s its own piece.

Regards,

Luke Kandia

Luke Kandia writes Chasing Clarity, a Substack covering healthcare systems, financial markets, public policy, and science communication — usually by starting with something small (an old coin, a data point, a footnote) and following it wherever it actually leads.

Nothing here is investment, financial, or legal advice — just a retired IT guy who now can’t stop wondering what’s actually behind the fluorescent lights in Kentucky. Do your own homework before you make decisions with your money.

Sources & further reading

• Yahoo Finance — “Trump and Musk say they want to make sure the Fort Knox gold is still there” — Bessent’s exact quote citing the September 30, 2024 audit report.

• CoinTribune — “Trump renews the call for an audit of Fort Knox’s 6,000 tons of gold” — 1953 and 1974 audit history, current market valuation, Massie’s Gold Reserve Transparency Act.

• Nationwide Coins — “Origin of the Fort Knox Conspiracy Theory” — Peter David Beter’s 1974 claim, Secretary William Simon’s response, and the congressional visit that followed.

• ABC News via AOL — “Trump, Musk raise questions, without evidence, about whether nation’s gold at Fort Knox has been stolen” — an alternate account of Beter’s original claim (British intelligence sources, rather than the Rockefeller-convoy version), and the AP photo credit.

• U.S. Mint — “Inspection of Gold at Fort Knox” — the Mint’s own September 1974 press release confirming it was the first time photography had ever been permitted inside the depository.

• Associated Press — gold bars and Mint Director Mary Brooks at the U.S. Bullion Depository, Fort Knox, Kentucky, September 24, 1974.

• NBC News — “Counterfeit gold bars discovered in New York City” — the September 2012 Manhattan tungsten-filled bar discovery.

• Reuters (2012) and Bullion Trading LLC — “Greatest Gold Counterfeiters in History” — density comparison between gold and tungsten, the 2009 Hong Kong “good delivery” bar reports, and detection methods including ultrasonic testing.

• The Washington Times — “Gold rush fizzles: Trump drops plan to check on Fort Knox inventory” — Bessent’s assurances, congressional access disputes, 1974 audit details.

• Yahoo News — “Trump Revives Fort Knox Audit Push, Wants to Knock on the Vault Door” — 2026 status update, Massie’s bill text.

• Yahoo News — “Elon Musk Finds DOGE’s Next Target: Gold Reserves at Fort Knox” — the February 16, 2025 X exchange between Rand Paul and Musk that started the audit push, Mike Lee’s contribution, and Musk’s unsubstantiated claims.

• Fox Business — “Trump backs confirming Fort Knox gold is there, as Treasury says it is” — Trump’s February 19, 2025 comments aboard Air Force One.

• BizPacReview — “Trump gives update on long overdue Fort Knox audit” — Trump’s May 2026 interview with Sharyl Attkisson, still floating a visit.

• Tavex — “Why Did Trump and Musk Suddenly Go Silent on Fort Knox Gold?” — on Musk’s departure from DOGE and the story’s fade from the news cycle.

• RealClearDefense — “What Happened to the Fort Knox Gold Audit?” — on the stalled momentum behind the 2025 audit push.

• Deutsche Bundesbank — “Bundesbank completes transfer of gold from New York” — the 2013 repatriation plan’s actual scope (674 tonnes: all of Paris, a portion of New York), completed 2017.

• Mining.com — “Calls grow for Germany to repatriate US-held gold” — the renewed January 2026 push to repatriate Germany’s remaining ~1,236 tonnes from New York.

• IDN Financials — “German economists urge withdrawal of gold reserves from the US” — current holding percentages (over half in Frankfurt, 37% New York, 12% London) and the Ifo Institute’s caution against repatriation.

• CMI Gold & Silver — “Inside the Gold Vault: How NY Fed Protects US & Global Gold” — facility-by-facility breakdown of US gold holdings and storage locations.

• Fox Business — “How much gold reserves does the United States hold?” — exact per-facility holdings figures, including the 54.1 million ounces at West Point.

• Fox Business — “The Fed Has 6,200 Tons of Gold in a Manhattan Basement — Or Does It?” — details of the 2012 New York Fed purity-testing audit.

• U.S. Treasury, Bureau of the Fiscal Service — Status Report of U.S. Treasury-Owned Gold — official per-location gold totals.

• U.S. Mint — Fort Knox Bullion Depository — official current holdings figures and statutory gold valuation.

• David Deutsch, The Beginning of Infinity: Explanations That Transform the World (2011) — the static-versus-dynamic-institution framework used throughout this series.

• Bitcoin News — “Trump Demands Fort Knox Audit After CIA Official Seized With $40M in Gold Bars”

• Yahoo Finance — “Trump wants Fort Knox ‘physically audited’ after stunning $40M arrest”

• Access Newswire — “Sound Money Defense League Applauds President Trump’s New Call for Fort Knox Gold Audit”

• CoinTribune — “Trump renews the call for an audit of Fort Knox’s 6,000 tons of gold” (already in your list — verify it’s the same one, since it also covers the May 2026 Rush story)

Many years ago, in one of my earliest diaries, I wrote down an idea that had been rattling around in my head.

I imagined all human knowledge as a gigantic book.

Everything we had discovered. Everything we understood. Physics, biology, mathematics, medicine, astronomy, engineering, philosophy — all of it.

And I wrote that humanity had barely flipped page one.

I don’t remember having some elaborate argument for it. It was simply an intuition: that despite everything human beings had discovered, what we knew must be infinitesimal compared with what remained to be discovered.

Decades later, I started listening to David Deutsch’s The Beginning of Infinity. I had cataloged my entire library at home on ChatGPT, and asked, “based on my books, what book would be a good recommendation?” David’s book topped the list.

And initially my reaction was: what the hell am I listening to?

Seriously. Deutsch seemed to be taking me all over the place. Epistemology. Physics. Mathematics. Evolution. Computation. Politics. Art. The Enlightenment. Karl Popper. Good explanations. Bad explanations.

I couldn’t understand why I was listening to this book.

But I kept listening.

And chapter by chapter, something strange began happening. Ideas that had seemed disconnected started connecting. Then they started connecting to things I had thought about myself.

Eventually I realized that Deutsch was building an argument around an idea I had once scribbled, in vastly simpler form, into a diary.

Except there was an astonishing difference.

My imaginary book of knowledge was gigantic.

Deutsch’s might have no last page.

The Things We Don’t Know We Don’t Know

We normally think about ignorance as missing information. There is something to know, and we don’t know it yet.

What is dark matter? How did life begin? Is there life elsewhere in the universe? How does consciousness arise?

Those are unanswered questions. Presumably there are explanations for them, and perhaps someday we’ll discover those explanations.

But there is a much more profound kind of ignorance.

There are questions we haven’t asked yet. Not because nobody has thought hard enough about them, but because the knowledge required even to formulate the question doesn’t exist yet.

Imagine bringing someone from 1750 into the present. Show them a smartphone.

You could tell them that this little piece of glass and metal communicates almost instantaneously with billions of other devices around the planet. That it determines its position by receiving exquisitely timed signals from machines orbiting Earth. That it can access much of humanity’s accumulated knowledge, translate languages, show moving pictures transmitted invisibly through the air — and even hold something resembling a conversation with an artificial intelligence.

Now explain how it works.

Where would you begin? Electricity? Electromagnetism? Atoms? Electrons? Semiconductors? Quantum mechanics? Relativity? Binary computation? Information theory? Satellites? Integrated circuits?

The problem isn’t merely that our visitor from 1750 doesn’t know how a smartphone works. Much of the conceptual structure required to understand the explanation doesn’t exist in his world yet.

That realization changes the metaphor of my imaginary “knowledge” book. Page two of that book may contain questions that cannot even be asked by someone still living on page one.

Knowledge Creates Ignorance

That sounds paradoxical, but I think it’s one of the most beautiful things about human progress.

Every major discovery doesn’t simply eliminate ignorance. It creates new ignorance.

Before we understood microorganisms, nobody was asking how bacterial resistance to antibiotics evolved. Before we understood DNA, nobody was asking whether a particular sequence of nucleotides could be edited. Before Einstein, engineers weren’t worrying about relativistic clock corrections in satellite navigation systems. Before quantum mechanics, nobody was asking how quantum states might be exploited for computation.

The questions become possible only after previous questions have been answered. An explanation opens a door. Behind it are ten more doors. Open those, and perhaps there are a thousand.

Knowledge doesn’t merely fill blank spaces on a map. Knowledge enlarges the map.

And suddenly my old diary metaphor seems inadequate. Perhaps humanity hasn’t simply turned the first page of an enormous book. Perhaps turning pages somehow causes new pages to appear.

Then Socrates Met Hermes — In a Dream

This was where Deutsch really got me.

Late in The Beginning of Infinity, in a chapter he calls “A Dream of Socrates,” Deutsch imagines Socrates visited in his sleep by a god claiming to be Apollo, who turns out to be Hermes. By then I had gone from wondering why I was listening to this strange book to repeatedly thinking: oh my.

The dialogue crystallizes something Deutsch had been building toward the whole time — an idea he inherited from Karl Popper: there is an enormous difference between possessing the correct answer and understanding why it is correct. A pronouncement from a god doesn’t count as knowledge just because it’s true.

It only becomes knowledge once it can be criticized, tested, and shown to survive — or fail to survive — that scrutiny.

An oracle can give you an answer. An authority can give you an answer. A textbook can give you an answer. Today, an artificial intelligence can give you an answer. But an answer isn’t necessarily an explanation. And without explanation — without the possibility of criticism and correction — we haven’t necessarily acquired the kind of knowledge Deutsch is talking about.

That idea seems especially important now. We have built machines capable of placing an extraordinary amount of accumulated human knowledge within conversational reach. I can ask a question today and receive in seconds an answer that might once have required days in a university library. That is astonishing.

But Socrates would still have questions. Why? How do you know? What explanation accounts for it? What competing explanation exists? What observation would expose an error? What problem does this explanation solve? And perhaps most importantly: how could we be wrong?

Those questions are not obstacles to knowledge. They are the machinery that produces it.

We Are Not at the End of Discovery

There is a recurring temptation in every age to believe that the really big discoveries have already been made.

Look around. We split the atom. Mapped the human genome. Detected gravitational waves. Photographed the environment surrounding a black hole. Built machines that can converse in natural language. Sent spacecraft beyond the planets. Surely we’re getting pretty far into the book.

But imagine someone in 1750 making the equivalent assessment. Newton had explained gravity. The planets obeyed mathematical laws. Telescopes had transformed astronomy. Human anatomy was being systematically investigated. The Enlightenment was underway. From inside that world, their accumulated knowledge must have seemed enormous.

And yet they hadn’t discovered electromagnetism, evolution by natural selection, germ theory, relativity, quantum mechanics, DNA, plate tectonics, nuclear physics, computers, or the expansion of the universe. They didn’t know about galaxies beyond our own. They didn’t know the age of Earth. They didn’t know the universe had a history. They didn’t know that every cell in their bodies contained a molecular information system inherited across billions of years.

They weren’t merely missing answers. They were missing entire categories of questions.

Why should we imagine that we’re different? What concepts will exist in 2300 that are literally impossible for us to formulate today? What phenomena surround us right now that we don’t recognize as phenomena? What assumptions do we consider obvious that future generations will find astonishing?

We cannot know. That’s the point.

The Beginning of Infinity

That is why Deutsch’s title finally landed so powerfully for me. The Beginning of Infinity. Not the middle. Certainly not the end. The beginning.

Human beings possess something extraordinary: the capacity to create explanatory knowledge. We encounter problems. We propose explanations. We criticize them. We discover errors. We construct better explanations. And those explanations reveal new problems.

There is no obvious reason to believe that process is approaching completion. Quite the opposite — the more we learn, the more enormous the frontier appears.

So, decades after writing that sentence in an old diary, I think I’d change it slightly.

I once imagined knowledge as a giant book and humanity as having barely flipped page one. Now I think that gives us too much credit. Because somewhere beyond the edge of what humanity presently understands are not merely answers waiting to be discovered. There are questions waiting to be invented. There are concepts for which we don’t yet have words. There are explanations that would be incomprehensible to us without discoveries that haven’t happened yet. There are pages we cannot read because they haven’t been written.

And perhaps the most extraordinary thing about knowledge is this: every page we turn writes the next one.

We have barely turned page one. And there may never be a final page.

Regards,

Luke Kandia

David Deutsch is a British theoretical physicist at Oxford, widely regarded as the father of quantum computing. He is also the author of The Fabric of Reality and The Beginning of Infinity.

Luke Kandia writes Chasing Clarity — one man’s search for a better understanding of the world, both small and large, around him.

I promised you this one.

The other day I wrote about silver disappearing from Canadian and American coins in the 1960s, and about Nixon severing the final link between the U.S. dollar and gold in 1971.

Canada, as it turns out, had started backing away from gold much earlier.

In January 1929, with gold flowing out of the country and pressure building on the Canadian dollar, Ottawa effectively stopped honouring the old gold-standard rules. Canada was still officially on gold, but in practice the government had begun preventing the normal conversion and export mechanisms that made a gold standard a gold standard.

Then came the Depression.

In October 1931, Canada made the break formal by banning gold exports without government permission. Britain had abandoned gold a month earlier.

So Canada was effectively off gold by 1929 and formally off it by 1931 — forty years before Nixon closed the American gold window.

But leaving the gold standard did not mean Canada stopped caring about gold.

Far from it.

Canada kept one of the world’s largest sovereign gold reserves for decades.

Then it sold almost all of it.

Slowly. Across nine prime ministerships. While sitting on top of some of the richest gold-bearing rock on Earth.

Today Canada is the world’s fourth-largest gold producer. We mine roughly 200 tonnes a year.

Our official gold reserve? Essentially zero.

That is where this story gets interesting.

First, a strange detour through the price of gold

When Canada imposed its gold-export embargo in 1931, it didn’t simply vacuum every ounce coming out of Canadian mines into a government vault.

The reality was more practical. Ottawa took control of the movement of gold. Exports required federal approval. The Royal Canadian Mint became the country’s principal refiner. Gold still left Canada in large quantities because selling it abroad generated something the country badly needed: foreign exchange.

Only part of that gold accumulated as a Canadian monetary reserve. Even in the 1930s, Canada often seemed less interested in owning the metal than in owning what the metal could buy internationally.

That instinct turns out to matter later.

During the Second World War, the controls became much tighter. Gold and foreign exchange were managed centrally through the Foreign Exchange Control Board.

Then, after the war, something genuinely odd happened.

In September 1951, the International Monetary Fund relaxed its opposition to countries allowing newly mined gold to be sold into private markets at prices above the official monetary price of US$35 an ounce.

Think about that. The Bretton Woods system ultimately rested on the U.S. dollar being convertible into gold at $35 an ounce for official purposes. And yet the IMF was acknowledging that private markets could value the same gold differently.

Canada moved almost immediately. Beginning October 1, 1951, Canadian miners could sell newly produced gold into those premium markets rather than through the official channel, although doing so meant giving up certain government mining assistance.

And the premiums were real. The IMF reported gold changing hands for roughly $38 to $39 an ounce in some markets shortly after the policy change.

Same metal. Different price.

But why didn’t everyone just arbitrage it?

Because $35 gold wasn’t available to everyone. The United States wasn’t running a retail window where you could show up with $35, buy an ounce, cross the street, and sell it for $39. The official $35 price belonged to the international monetary system — governments, central banks, official settlements. Private buyers couldn’t simply acquire unlimited official gold and flip it into the higher-priced private market.

So the gap wasn’t really an information problem. It was an access problem. The rules themselves kept the two markets apart.

London makes the distinction especially clear. Its bullion market had been closed since 1939. When it finally reopened on March 22, 1954, the first fixing was 248 shillings and sixpence — about US$34.79 at the prevailing exchange rate.

Almost exactly the official $35 parity.

Canada kept liberalizing too, removing its remaining restrictions on gold transactions in 1956.

Then the strain started showing. In 1960, private demand briefly pushed the London price toward $40 an ounce. Now the world’s most important bullion market was sending a very public message: maybe $35 wasn’t the right number anymore.

The response was extraordinary. In 1961, eight central banks — the United States, Britain, West Germany, France, Italy, Belgium, the Netherlands, and Switzerland — created the London Gold Pool. They pooled physical gold. The Bank of England used it to intervene in the London market. If demand pushed the price too high, they sold gold. If demand weakened, they could buy it back.

For several years, it worked. Early on, the Pool was even a net buyer.

Then confidence in the dollar began to crack. By 1967 and early 1968, central banks were pouring enormous quantities of physical gold into the market simply to hold the price near $35.

In March 1968, they gave up. The Gold Pool collapsed. London temporarily closed. When it reopened on April 1, gold fixed at $38 an ounce.

From that point on, the split became explicit: official monetary gold could still be valued at $35. Private gold would trade wherever the market wanted.

Three years later, Nixon closed the gold window altogether.

So the real story isn’t that policymakers somehow failed to notice a giant arbitrage opportunity for twenty years. It’s stranger than that. They knew the market could value gold differently from the official price. They built rules separating the two markets. And when the market signal became inconvenient, central banks actually used their own gold reserves to fight it.

The problem wasn’t that nobody could see reality. The system was designed so reality didn’t automatically get to rewrite the official number.

Keep that idea in mind. We’re going to see it again.

Canada had a mountain of gold

At its peak in the 1960s, Canada’s official international reserves held roughly 1,023 tonnes of gold. That’s about 33 million ounces. At about 12.4 kg or 27.4 lb, a London Good Delivery bar would weigh about 400 ounces. So that would be about 82,500 of those London gold bars.

At the August 7, 2026 gold price of roughly US$4,336 an ounce, that pile would be worth about US$143 billion.

For comparison, Canada held about US$96.5 billion in liquid foreign-currency securities and deposits as of June 30. So the old gold reserve, simply sitting there untouched, would today be worth nearly 50% more than Canada’s liquid foreign-currency portfolio.

Now, an important caveat. Canada did not simply throw the sale proceeds into a furnace. The money was reinvested into interest-bearing reserve assets that earned returns for decades. So saying Canada “lost $143 billion” would be wrong. A proper comparison would have to calculate the total return on the assets Canada bought with the proceeds and compare that with the return on gold. As far as I can find, nobody has published that calculation.

So the honest claim is narrower: if Canada had simply kept the physical gold, it would be worth about US$143 billion today.

Still a mountain.

How do you sell a mountain?

Slowly. That’s the interesting part.

On December 11, 1979, Finance Minister John Crosbie, serving in Joe Clark’s short-lived Progressive Conservative government, announced a plan to sell up to one million ounces of Canada’s roughly 22-million-ounce reserve. His reasoning wasn’t crazy — gold prices had surged, and gold had grown to almost three-quarters of Canada’s official reserves by value. Crosbie proposed rebalancing the portfolio and putting some of the money into interest-bearing assets.

Two days later, Clark’s government fell.

The policy survived. Nearly 400,000 ounces were sold during the first quarter of 1980, which straddled the transition from Clark back to Pierre Trudeau. That August, Trudeau’s finance minister, Allan MacEachen, authorized another sale of up to one million ounces. By the end of 1980, Canada had sold 1.2 million ounces and still held about 21 million.

But what had started as portfolio rebalancing was becoming something else. A continuing policy. The logic was simple: gold earns nothing, bonds earn interest, gold costs money to store, foreign securities are liquid. Why leave metal sitting in a vault when the money could be working?

Reasonable enough. Except Canada kept applying that same logic. And applying it. And applying it.

Thirty-six years later

Through the 1980s and 1990s, hundreds of tonnes disappeared from the reserve. Different governments. Different finance ministers. Same direction. By 1985, Canada was down to roughly 500 tonnes — around half its 1965 peak. The sales continued through Michael Wilson. Don Mazankowski. Paul Martin.

By the early 2000s, almost nothing remained. And still the selling continued. In December 2002, Finance Minister John Manley authorized another sale of roughly 2.6 tonnes. By 2003, Canada’s actual bullion was essentially gone. All that remained was a small stock of gold coins. Those were wound down too.

In February 2016, under Finance Minister Bill Morneau, Canada sold 21,851 ounces of gold coin, worth about $35 million at the time. Afterward, the Department of Finance reported just 77 ounces remaining.

Seventy-seven.

Canada had once held more than a thousand tonnes. Today its official holdings are recorded as essentially zero. And Canada is alone among the G7 — the United States still holds gold, and so do Germany, Italy, France, Japan, and Britain. Canada decided it didn’t need any.

And then I found a typo

This is almost too perfect. A 2026 Finance Canada briefing document says Canada’s reserve had already fallen to “just under 2 million ounces” by the end of 1979 — which would mean more than 90% of the historical reserve had disappeared before the formal sales program even started.

Except it hadn’t. John Crosbie’s own 1979 budget says Canada held 22 million ounces. An IMF historical study says the same. The Bank of Canada’s 1980 annual report shows 21 million ounces remaining after that year’s sales.

So somewhere inside a government briefing, it appears that 22 million became 2 million. Probably a dropped digit. Nothing more sinister than that. I wouldn’t build an argument about institutional competence around a typo. But in a story about inherited assumptions travelling quietly through institutions, the irony is hard to ignore.

The part that makes it sting

Canada didn’t sell its gold because we ran out of gold. Quite the opposite. Canada is now the fourth-largest gold-producing country on Earth, behind China, Russia, and Australia. Canadian mines produced about 200 tonnes in 2024, up 25% from 2015. We are exceptionally good at finding the stuff. We are exceptionally good at extracting it. We just don’t keep any as a sovereign reserve.

Now, these are two separate systems. A gold mine in Ontario doesn’t automatically ship some percentage of its production to Ottawa. If the federal government wanted gold reserves, it would generally buy gold just as it buys other reserve assets. And this isn’t an argument that Canada should have nationalized its gold-mining industry — companies like Barrick Gold and Agnico Eagle are private businesses, and nothing here requires Ottawa to own them or dictate what happens to what they produce.

The real question is simpler: why did one of the world’s great gold-producing countries decide that gold itself had no place in its reserve portfolio? Especially when the rest of the world started moving the other way. Central banks have now been net buyers of gold for fifteen consecutive years. In 2024 alone, they bought more than 1,000 tonnes — Poland, Turkey, India, China.

Canada’s bullion was already gone by 2003. But Ottawa kept liquidating the remaining gold coins until 2016 — roughly six years after central banks collectively had turned back toward gold.

Again: different circumstances. Same Canadian answer.

This is where David Deutsch shows up

I’ve been reading David Deutsch’s The Beginning of Infinity, and one of his ideas kept nagging at me while I researched this.

Deutsch distinguishes between static societies and dynamic societies. Very roughly, static societies suppress criticism and novelty in order to preserve existing ways of doing things. Dynamic societies permit criticism, correction, and the creation of new knowledge. I’m borrowing that idea and applying it to institutions.

Because here’s the tempting argument about Canada’s gold reserve: nobody ever reconsidered the policy.

Except that’s not true. They did.

That’s not stupidity. That’s not neglect. That’s a coherent argument.

But look at the question being asked. The question wasn’t really should Canada own gold? It was closer to does gold perform well under the criteria we’ve already chosen for the Exchange Fund Account?

Those criteria are the three things Lane’s speech actually names. Liquidity — can you convert the asset to cash fast, at scale, without moving the price against yourself, if a currency crisis hits on a Tuesday? Yield — does it earn interest while it sits there? Asset-liability matching — does the asset line up with what the reserve is actually for, funding short-term currency intervention, rather than just sitting there as dead weight? Gold loses on all three by its nature. It earns nothing. It’s slower to move at scale than a Treasury bill. And it doesn’t correspond to any specific liability the way a bond does. Once those are the criteria, gold starts the exam with three strikes against it.

And that leads to what I think is the real question: were the criteria themselves ever seriously reconsidered? Not the policy. The premises underneath it.

Those premises are the deeper layer nobody argued for, because nobody had to. They’re assumptions like: that a reserve’s main job is funding short-term currency intervention, not insuring against a slower, larger shock. That yield matters more than an asset carrying zero counterparty risk — gold isn’t a promise from anyone; it doesn’t depend on another government or institution honouring an obligation to still be worth something. A Treasury bond does. And that the kind of crisis worth hedging against is a currency-market wobble, not a crisis where trust in fiat currencies, or in the U.S. financial system itself, is the thing that’s failing. Lane’s speech defended gold’s performance inside that frame. It never had to defend the frame, because nobody was asking him to.

That’s harder to establish. I can’t prove that nobody, anywhere inside Finance or the Bank of Canada, ever reopened that question. Maybe someone did. But if they did, I haven’t found much public evidence of it. What we can see publicly is repeated examination of gold through essentially the same reserve-management lens. And when you keep asking the same question through the same lens, you shouldn’t be surprised when you keep getting the same answer.

That’s the institutional problem

Thirty-six years. Nine prime ministerships if you start with Crosbie’s proposal. Finance ministers from opposing parties. Governments that disagreed profoundly about taxes, spending, deficits, trade, and almost everything else. On gold reserves? Remarkable continuity.

To be fair, Finance Canada’s position is internally consistent. Gold is non-yielding. Canada wants liquid, interest-bearing foreign assets. Canada hasn’t regularly intervened in currency markets since 1998. Under that framework, holding gold isn’t necessary. That’s a perfectly respectable argument.

But there is a difference between reviewing a policy and reviewing the assumptions that determine how the policy will be judged.

What Norway did differently

Norway offers an interesting contrast. Not because Norway hoarded barrels of oil — it didn’t. And not because Canada should have copied Norway’s exact tax system, state ownership model, or investment portfolio. That’s not the lesson.

Norway did something more fundamental: it built reconsideration into the machinery of government.

Norway created a state oil company. It kept direct financial interests in petroleum production. It taxes petroleum profits at a combined marginal rate of 78%. Then it built a sovereign wealth fund designed to turn an exhaustible natural resource into something future generations could inherit. That fund is now worth roughly US$2 trillion.

But the part I find most interesting isn’t the size. It’s the mechanism. Norway publishes annual white papers to Parliament on the fund’s strategy. And an expert council now recommends comprehensive reviews of the investment framework every five years.

In other words, the institution doesn’t simply ask is our portfolio performing properly under the existing rules? It periodically drags the rules themselves back onto the table. Is this still the right mandate? Are these still the right assumptions? Are we still measuring the right things? Has the world changed enough that the framework itself needs to change?

That’s the mechanism Canada could have borrowed. Not Norway’s oil policy. Not its tax rate. Not its portfolio. The habit of periodically forcing an institution to defend not only what it is doing, but why those are still the right things to be doing in the first place.

Because without that second question, review can become strangely circular. You measure the policy against the framework. The policy passes. Then you congratulate yourself for having reviewed it. What never gets tested is the framework that determined the result.

My old dime, one more time

I keep coming back to the coin that started all this. An old Canadian dime sitting on my kitchen table. It says 10 CENTS. But the silver inside it is worth considerably more.

Canada removed silver from the dime in 1968 because the economics of using precious metal in circulating coins no longer made sense. A decision was made. The coin changed. Done.

Canada’s gold reserve disappeared differently. Not through one dramatic announcement. Not through one obvious mistake. It disappeared slowly. Sale after sale. Government after government. Review after review.

The people involved weren’t asleep. They were looking. They just kept looking through essentially the same lens.

And maybe that’s the lesson. Institutions don’t always fail because nobody is thinking. Sometimes they fail because very smart people keep answering the same question correctly — long after someone should have asked whether it was still the right question.

The United States, by contrast, still claims to hold more gold than any other nation on Earth — over 8,000 tonnes, split between Fort Knox, West Point, Denver, and the New York Fed. And here’s the part that stopped me cold when I went looking: nobody has actually verified that it’s all still there. Not thoroughly. Not independently. Not in over fifty years.

That’s next.

Regards,
Luke Kandia

Luke Kandia writes Chasing Clarity, a Substack covering healthcare systems, financial markets, public policy, and science communication — usually by starting with something small, like an old coin, a data point, or a footnote, and following it wherever it actually leads.

Nothing here is investment, financial, or legal advice — just a curious fellow who can’t stop thinking about a vault that used to have gold in it. Do your own homework before you make decisions with your money.

Sources & further reading

• BNN Bloomberg — “More mines, more ounces: A decade of gold gains turned Canada into a mining powerhouse”

• The Oregon Group — “Canada’s Gold Rush 2.0”

• Natural Resources Canada — “Gold facts”

• World Gold Council — Gold Demand Trends, Full Year 2024: Central Banks

• CBC News — “Ottawa sells off almost all its gold reserves, leaving just 77 ounces — or less”

• Global News — “Canada sells off remaining gold reserves; has just 77 ounces left” — G7 comparison of gold holdings

• Bordo & Redish — “Costs and Benefits of Exchange Rate Stability: Canada’s Interwar Experience”, Federal Reserve Bank of St. Louis

• “Foreign Exchange Control in Canada, 1939–51”, Canadian Journal of Economics and Political Science

• Bank of Canada — “Back on the Gold Standard, 1926–31”

• Bank of Canada — historical study of Canada’s gold reserves

• IMF eLibrary — Gold in the IMF, on the 1951 premium-market policy change

• Government of Yukon EMR Library — Canadian Mineral Industry, 1951

• IMF eLibrary — on Canada’s March 21, 1956 removal of remaining gold-transaction restrictions

• London Bullion Market Association — “After the War Is Over: The London Gold Market Reopens, 1954–1968”

• London Bullion Market Association — “Re-opening the London Gold Fixing after the Second World War”

• Federal Reserve History — “Gold Convertibility Ends”

• John Crosbie — 1979 Budget Speech, Government of Canada

• Bank of Canada — Annual Report of the Governor, 1980

• Bank of Canada — Timothy Lane, “Taking Precautions: The Canadian Approach to Foreign Reserves Management” (February 6, 2019)

• House of Commons of Canada — Standing Committee on Public Accounts, Meeting 31, Evidence

• Government of Canada — Briefing binder for the Deputy Minister of Finance, Standing Committee on Public Accounts appearance, February 9, 2026

• Government of Canada — Official International Reserves, July 6, 2026

• Reuters — “Gold heads for strongest week since January as US payrolls data comes into focus” (August 7, 2026)

• Norwegian Ministry of Finance — The Government Pension Fund 2026, annual white paper to the Storting (March 27, 2026)

• Norskpetroleum.no — “Norway’s petroleum history” and “The government’s revenues”

• David Deutsch, The Beginning of Infinity: Explanations That Transform the World (2011)

A couple of weeks ago, I went to the Winnipeg Fringe Festival and watched FISSION: The Death of Dr. Louis Slotin.

I had wanted to see it for a very Winnipeg reason.

Louis Slotin went to my high school.

St. John’s High School, in Winnipeg’s North End.

I found this out somewhat by accident. Some time back, I looked up St. John’s on the Internet, mostly out of curiosity, to see who else the school had produced over the decades. It turns out to be a remarkable list of names for a North End public high school. Burton Cummings, of The Guess Who, is on it. So is Monty Hall, who left Winnipeg and became famous across North America as the creator and longtime host of the game show Let’s Make a Deal. David Steinberg, one of the defining comedians and talk-show personalities of the 1960s and 1970s, is there too, though harder to verify. And further down the list, tucked in among the entertainers, was Louis Slotin.

A striking number of the names on that list — Louis, Monty, and David among them — were Jewish. That isn’t a coincidence. Winnipeg’s North End, through the early and mid-20th century, was one of the great immigrant neighbourhoods of the Canadian Prairies: a dense, overlapping mix of Eastern European Jews, Ukrainians, Poles, and other communities who had come to Canada through successive waves of migration. Slotin’s own parents were Yiddish-speaking Jewish refugees who had fled the pogroms of Czarist Russia. (Canada’s History) St. John’s drew its student body straight from those streets.

Not the same building I attended, as it turns out. More on that in a moment.

Then, today, Louis Slotin unexpectedly appeared in my reading again. Dan Denning was writing about the end of the Second World War and the possibility of a third American atomic bomb being used against Japan after Hiroshima and Nagasaki. In passing, he mentioned the plutonium core that would later become known as the Demon Core, and the scientist who died while experimenting with it.

Dan spelled his name “Louis Sloton.”

Close. But when the guy went to your alma mater, you notice these things.

His name was Louis Slotin. And his story is extraordinary.

From the North End to Los Alamos

Slotin was born in Winnipeg’s North End on December 1, 1910.

He attended St. John’s Technical High School, then entered the University of Manitoba at only sixteen. He eventually won the university’s gold medal in chemistry and physics, continued his studies overseas, earned a doctorate in London, and eventually joined the Manhattan Project.

By late 1944, he was at Los Alamos.

He became particularly skilled at working with the new and barely understood material at the heart of the atomic bomb: plutonium. (MHS Manitoba Historical Society)

This is where the connection to my own life gets interesting.

Slotin and I attended the same high school. But we did not walk the same halls.

His St. John’s was built between 1910 and 1912, a magnificent red-brick and Tyndall-stone building with a tower over its main entrance. By the 1960s it had become outdated, and a replacement school was constructed immediately behind it. The original building was demolished in 1967. (MHS Manitoba Historical Society)

There is a wonderful photograph that captures the transition.

And then there is the St. John’s I knew.

So Louis Slotin and I never walked the same hallways. But we went to the same school.

And decades before I arrived there, one of its graduates found himself standing over one of the most dangerous objects human beings had ever created.

Tickling the dragon’s tail

The Fringe play recreated Slotin’s final experiment in deliberately crude fashion.

There was a metallic sphere, split into two halves. Slotin used an ordinary flat-head screwdriver to keep the upper portion slightly separated from the lower.

Then, in the play, something dropped behind him. He turned. The screwdriver slipped.

Clunk.

The upper hemisphere dropped closed.

Even sitting safely in a Winnipeg theatre almost eighty years later, watching actors and props, it was chilling.

The historical record differs slightly from the stage version. There is no established dropped object that caused Slotin to turn. What we know is that on May 21, 1946, while Slotin was manually controlling the experiment, the screwdriver slipped.

The result was the same.

For one horrifying instant, the system shot past criticality. Los Alamos’s later technical reconstruction confirms the reflector closed much farther around the core and produced a very brief, intense burst of fission before Slotin separated it again. (MCNP / LANL Technical Report)

To understand why that killed him, you don’t need much nuclear physics.

Think of it this way.

Plutonium atoms can split — fission — releasing energy and neutrons. Those neutrons can strike other plutonium atoms, causing more fissions, which release more neutrons.

If enough neutrons escape from the material, the process dies down. If just enough remain to sustain the next generation of fissions, you reach criticality. If still more remain, the reaction begins to grow. Push it far enough beyond criticality, and it can grow extraordinarily fast.

Slotin was experimenting with material surrounding the plutonium that reflected neutrons back toward the core.

Move the reflector closer and fewer neutrons escape.

Closer.

Closer.

Closer.

Somewhere in that progression lies an invisible boundary between a reaction that fades away, one that sustains itself, and one that can suddenly run away.

Los Alamos scientists had an appropriately ominous nickname for experiments in which they approached that boundary.

Tickling the dragon’s tail.

The dragon woke up

When Slotin’s screwdriver slipped, the neutron-reflecting material suddenly enclosed much more of the plutonium core. For a fraction of a second, the rate of fission soared.

It was not an atomic bomb explosion. The room did not blow apart.

Why not? That puzzled me when I watched the play — I suspect it puzzles most people who’ve seen it, or who’ve watched Fat Man and Little Boy. If the core went critical, why wasn’t there a mushroom cloud?

Criticality — even a rapid, runaway version of it called prompt criticality — does not automatically mean a bomb explosion. Slotin’s accident produced a sharply growing chain reaction, but only for an instant. The sudden burst of energy immediately began changing the physical conditions that had made the reaction possible — the assembly heated, expanded slightly, and the geometry that had briefly allowed it to run away started working against it — and Slotin pulled the reflector off a fraction of a second later. The whole episode was over almost as quickly as it began. A nuclear weapon, by contrast, is deliberately engineered to drive its fissile material much further beyond that threshold and hold it there — for an instant longer than nature wants to allow — so that an enormous number of fission generations occur before the assembly destroys its own geometry and the reaction stops itself. Slotin’s bench got the radiation burst without the explosion. Standing an arm’s length away, that was enough.

That almost makes the event more unsettling. Nothing had to explode. An intense burst of neutron and gamma radiation passed through the people standing nearby.

Slotin was closest. He reacted immediately, pulling the assembly apart. That immediately stopped the excursion and prevented everyone else in the room from receiving still more radiation.

But for Slotin, the damage had already been done.

Radiation at sufficiently high doses damages cells throughout the body. There is no need for a conventional wound, blast, or fire. The energy is deposited directly into living tissue, disrupting cells and their ability to reproduce and repair themselves.

Slotin had absorbed a fatal dose.

He died nine days later, on May 30, 1946. He was 35 years old. (MHS Manitoba Historical Society)

A fabulous 28 min short movie is available to watch on Youtube, called “The Demon Core” about Louis Slotin and that event.

The Demon Core

And then the story gets stranger. Because this piece of plutonium had already killed someone.

Nine months before Slotin’s accident, physicist Harry Daghlian had been conducting a different criticality experiment using the same plutonium core. That experiment also went accidentally critical. Daghlian received a fatal radiation dose and died weeks later.

Then came Slotin.

Two scientists. Two criticality accidents. The same plutonium core.

Only afterward did history give it the name that now seems almost inevitable: the Demon Core.

After Slotin’s death, Los Alamos stopped conducting this kind of criticality experiment with scientists standing immediately beside the material. Subsequent experiments incorporated remote operation and much greater separation between people and the nuclear assembly. (Nuclear Museum)

But the Demon Core’s history had begun before either accident. It had been made for war.

The third shot

Hiroshima was bombed on August 6, 1945. Nagasaki followed on August 9. And the United States was preparing another plutonium core — intended for another Fat Man-type implosion weapon if Japan did not surrender.

This is the “third shot” Dan Denning was referring to.

The exact chronology of the core’s attempted shipment is surprisingly murky. Robert Bacher, the Los Alamos physicist overseeing the core, later remembered that it had been checked out and a car was waiting to take it toward the airport when Robert Oppenheimer suddenly arrived with a hold order from Washington. Bacher himself couldn’t later recall the precise date.

What we do have is a remarkable contemporary document, dated August 10, 1945. General Leslie Groves told Army Chief of Staff George Marshall that the next implosion bomb’s components could leave New Mexico within days, and that, barring delays, the weapon might be ready for use after August 17 or 18.

Marshall wrote across the bottom of that memo:

“It is not to be released on Japan without express authority from the President.”

Whatever the precise sequence of events at Los Alamos, the outcome is clear. The third core never went to Tinian.

Japan announced its surrender on August 14–15. The third bomb was never used.

This particular sphere of plutonium never reached Japan. It remained at Los Alamos. It became an experimental core.

Harry Daghlian worked with it. Then Louis Slotin worked with it.

A piece of plutonium manufactured as the heart of another weapon against Japan never destroyed its intended target. Instead, it killed two of the scientists studying it. (Nuclear Museum)

There is something haunting about that.

Back in Winnipeg

I keep thinking about the moment in the Fringe play.

A man standing over a laboratory bench. A screwdriver. Something falls behind him. He turns.

Clunk.

Again, that dropped object belongs to the theatrical version of events. History gives us only the slipping screwdriver. But dramatically, it communicated the terrifyingly small margin involved.

There was no enormous reactor surrounding Slotin. No giant control room. No red countdown clock. No mushroom cloud.

Just a 35-year-old physicist manipulating an experiment at the edge of criticality.

And a screwdriver.

Before all of that, Louis Slotin had been a kid in Winnipeg’s North End. He had gone to St. John’s.

Not my St. John’s building. His magnificent old school was gone by the time mine took its place. But for one brief period captured in that aerial photograph, both buildings stood together — one generation of St. John’s giving way to another.

Decades later, I would attend the school behind it. And nearly a century after Slotin had been a student there, I would sit in a Winnipeg theatre watching an actor portraying him make one tiny movement with a screwdriver.

The hemisphere closed. The dragon woke.

And Louis Slotin, graduate of St. John’s Technical High School in Winnipeg, had only nine days left to live.

It’s strange what catches your eye in a stranger’s writing. Dan Denning dropped one letter from a dead man’s name in passing, on his way to a bigger point about the bomb that never fell.

I only stopped because Louis Slotin and I had gone to the same school.

Not the same building. Not the same halls. The same St. John’s.

Lucas Kandia
Chasing Clarity

References

1. Manitoba Historical Society. “Memorable Manitobans: Louis Alexander Slotin.”

2. Manitoba Historical Society. “Historic Sites of Manitoba: St. John’s Technical High School.”

3. Canada’s History. “Dr. Louis Slotin and ‘The Invisible Killer.’”

4. McLaughlin, Thomas P., Shean P. Monahan, et al. A Review of Criticality Accidents, 2000 Revision. Los Alamos National Laboratory, LA-13638. Technical reconstruction of the Slotin accident.

5. Office of Scientific and Technical Information (OSTI). Technical account of the beryllium-reflector apparatus used in Slotin’s experiment.

6. National Museum of Nuclear Science & History. “Atomic Accidents.”

7. California Institute of Technology Archives. Oral history interview with Robert Bacher.

8. Dannen, Gene. “The Decision to Use the Bomb: Groves memo to Marshall, August 10, 1945, with Marshall’s handwritten hold order.”

Note on sourcing: Where the historical record is genuinely disputed — particularly the exact date the third plutonium core’s shipment was halted — the piece says so rather than presenting a single account as settled fact.

I wasn’t trying to research monetary history.

I was going through an old collection of coins.

There were Canadian dimes from 1914, 1917, 1920, 1936, 1942, 1945, 1955, 1961. A 1947 dime with the little maple leaf after the date — indicating that it was actually struck in 1948 while the Mint waited for new dies after Indian independence.[^1]

There was a beautiful 1967 Centennial set.

And then there was an American 1964 Kennedy half-dollar.

At first I was simply trying to figure out what they were worth.

Then something became obvious.

Drop one of those old Canadian dimes onto a hard surface and it doesn’t sound like a modern dime.

It rings.

The newer coin gives you a dull little clink.

The old one sings.

Why?

Silver.

And once I started arranging the coins by date, an unexpectedly large historical story appeared right there on the kitchen table.

The denominations didn’t change. Ten cents still said 10 CENTS. Twenty-five cents was still twenty-five cents. A half-dollar was still fifty cents.

But the substance from which the money was made slowly changed. Precious metal became less precious metal. Then no precious metal at all.

And that sent me down a rabbit hole extending from my pile of Canadian dimes... to ancient Rome... to Franklin Roosevelt... to Bretton Woods... to Charles de Gaulle demanding America’s gold... and eventually to Richard Nixon appearing on television in 1971 and ending the whole arrangement.

The recurring question throughout the story is deceptively simple: what happens when the promises represented by money grow faster than the valuable thing supposedly backing it?

Money that was actually worth something

For much of Canadian history, a coin wasn’t valuable merely because the government said it was valuable. There was valuable material inside it.

Canadian dimes through 1919 were sterling silver — 92.5% silver. Beginning in 1920, the silver content was reduced to 80%.

So one of the old 10-cent pieces sitting on my table contained about 1.86 grams of actual silver.

That’s why an old silver dime today is worth considerably more than the number stamped on it. The government still says 10 cents. The metal politely disagrees.

And Canada wasn’t unusual. American dimes, quarters and half-dollars were 90% silver through 1964. Then something happened very quickly.

1965: America starts taking the silver out

By the early 1960s, the relationship between the market price of silver and the face value of American coins was becoming untenable.

People began doing exactly what you would expect. They kept the silver coins. Why spend something containing valuable silver when you could spend a newer coin containing inexpensive base metal instead?

This is the classic environment for Gresham’s Law: when two forms of money circulate at the same legal value, people tend to spend the less valuable one and hoard the more valuable one.

The United States responded with the Coinage Act of 1965. Silver disappeared from newly issued dimes and quarters, which became copper cores clad in copper-nickel.

The Kennedy half-dollar was treated somewhat differently. The beautiful 1964 Kennedy half I found is 90% silver. The replacement half-dollar introduced under the 1965 legislation was reduced to 40% silver — the U.S. Mint’s own December 1965 announcement explicitly described the new coin as 40% silver, compared with the traditional 90% silver half-dollar.[^3]

Think about what had happened. The coin still said HALF DOLLAR. But the amount of silver represented by that half-dollar had been cut dramatically. And eventually even that silver disappeared.

Canada followed almost simultaneously

Canada confronted essentially the same commodity problem. The Royal Canadian Mint describes 1967 as the year the silver content of Canadian coinage was reduced from 80% to 50%, followed by the transition to nickel in 1968.[^2]

This happened right in the middle of one of the most beautiful Canadian coin issues ever produced. Canada’s 1967 Centennial set, designed by Alex Colville, featured animals instead of the usual designs: the dove on the cent, the rabbit on the nickel, the mackerel on the dime, the bobcat on the quarter, the wolf on the half-dollar, and that magnificent Canada goose on the silver dollar.

I had one of those sets sitting in front of me. It’s an almost perfect historical artifact because it was produced precisely as Canada’s metallic monetary system was changing.

And the transition wasn’t perfectly synchronized with the calendar. The Mint’s specifications show both 50%-silver and nickel quarters in 1968: production changed composition during the year rather than conveniently at midnight on December 31.[^4]

By the end of the 1960s, however, the transformation was essentially complete. Canada’s circulating money no longer needed precious metal to function as money. Neither did America’s.

And none of this was particularly revolutionary. Human beings had been experimenting with this problem for thousands of years.

Rome figured it out long before us

This is where comparisons with ancient Rome become irresistible — but also where we need to be careful.

It would be misleading to claim that one Roman coin, the denarius, simply declined continuously from nearly pure silver to 2% silver. The monetary system itself changed.

Early imperial silver was remarkably pure. Under Augustus, Roman silver coinage could approach 95–98% silver.

Over succeeding centuries, emperors repeatedly reduced either the weight of the coins, the purity of the silver, or both.

Rome also changed the monetary unit itself. Under Caracalla in the early third century came the coin we now call the antoninianus. It was tariffed at roughly two denarii while containing substantially less than twice the silver.

That is an important distinction. Rome wasn’t simply putting progressively less silver into exactly the same coin. It was also changing the relationship between denominations — effectively stretching the monetary metal farther.

Over time the antoninianus itself deteriorated spectacularly. By the later third century, nominal Roman “silver” coinage could contain only a few percent actual silver.

The denomination remained money. The silver increasingly did not.

But this needs an enormous historical warning label: Rome did not fall because somebody diluted the denarius. Roman decline had vastly more complicated military, political, demographic and economic causes. Nor does it mean modern Canadian nickel coinage is economically equivalent to a third-century antoninianus. It isn’t.

But the recurring monetary pressure is fascinating. A government has obligations. The amount it has promised to spend rises. The valuable commodity embedded in its monetary system constrains how much money it can produce. Sooner or later, somebody notices: if we put less precious metal into each monetary unit, we can make more monetary units from the same amount of metal.

Rome learned that lesson early. Modern states eventually found an even more powerful solution. They removed the metal constraint altogether.

But silver wasn’t the ultimate constraint

Gold was.

For centuries, gold played a peculiar role in monetary systems. It was simultaneously money and a constraint upon money.

Governments could print paper currency. Banks could create credit. But if holders of that paper retained the right to say “fine, give me the gold” — then eventually the paper promises had to bear some relationship to the gold available to redeem them.

That placed a limit — admittedly imperfect — on monetary expansion.

Gold standards also had very real costs. They could transmit deflation, constrain governments during crises and force painful domestic adjustments when gold flowed out of a country. The restraint worked in both directions.

And in the United States, Franklin Roosevelt dramatically changed that relationship in 1933.

Roosevelt changes the rules

During the banking crisis of the Great Depression, Roosevelt’s administration severely restricted private monetary gold ownership and required substantial amounts of monetary gold to be surrendered. It wasn’t until Gerald Ford in 1974, rescinded this decree were private citizens able to own gold bullion once again - 41 years later.

The Gold Reserve Act of 1934 went further, transferring monetary gold to the U.S. Treasury and ending ordinary domestic redemption of dollars into gold.

Then came one of the most important price changes in American monetary history. Gold had previously been officially valued at $20.67 per troy ounce. The Roosevelt administration changed the price to $35 per ounce.

Here’s where the arithmetic matters. The price of gold in dollars increased by about 69%. But that does not mean the dollar was devalued by 69%. Look at it from the opposite direction and the dollar retained about 59% of its former gold value. In other words, its gold value fell about 41%.

That’s not an accounting curiosity. It’s a substantial monetary devaluation.

And then, a decade later, that new $35 figure became the foundation of the international monetary system.

Bretton Woods

In July 1944, representatives of 44 nations gathered at Bretton Woods, New Hampshire, to design the postwar monetary order.

The resulting system was ingenious. Other major currencies would maintain exchange-rate relationships with the U.S. dollar. And the United States would maintain the dollar’s relationship with gold:

$35 = one troy ounce of gold

But there was an important catch. An ordinary American couldn’t walk into the Treasury with $35 and ask for an ounce of gold, due to the Gold Reserve Act of 1934 which disallowed normal citizens the right to convert cash to gold. And even if they could get around that act, the 1933 decree prohibiting private ownership of gold made it illegal to do have possession of gold.

But foreign governments and monetary authorities effectively could.

That distinction made the U.S. dollar the centre of the postwar financial system. Instead of every country needing enormous gold reserves to settle international accounts, countries could hold dollars. Those dollars were credible because, ultimately, the United States promised to convert eligible foreign official dollar holdings into gold at the fixed price.

The dollar was therefore not exactly gold. But internationally it was as good as gold — as long as everybody believed the United States could make good on the promise.

And for a while, it worked brilliantly. The United States emerged from the Second World War holding an enormous share of the world’s monetary gold. Europe and Japan were rebuilding. The world needed dollars. The arrangement seemed almost indestructible.

Except it contained a fundamental tension.

The Triffin dilemma

The world needed dollars to conduct international trade and accumulate reserves. How did the rest of the world get those dollars? America had to send them abroad. In practice that meant the United States had to run external deficits — buying, investing and spending abroad so that dollars accumulated in other countries.

But every dollar accumulating overseas represented a potential claim on America’s finite gold reserve.

This was the Triffin dilemma: supplying the dollar reserves the world needed tended, over time, to undermine confidence that all those dollars could remain convertible into a finite U.S. gold stock.

Economists disagreed over whether better fiscal and monetary discipline could have prolonged the system, and Bretton Woods should not be treated as mechanically doomed from the moment it was created. But the tension itself was real. More international liquidity required more dollars. More dollars meant more potential claims against America’s gold.

Eventually those claims became very large indeed. And then someone decided to test the promise.

Enter Charles de Gaulle

France had become increasingly uncomfortable with the monetary privilege enjoyed by the United States.

De Gaulle attacked the special advantage the dollar gave America — what his finance minister Valéry Giscard d’Estaing famously called America’s “exorbitant privilege.” The phrase dates to Giscard’s tenure as finance minister in the 1960s.[^5]

The objection was easy to understand. Suppose France wanted to purchase $100 million worth of American goods. France had to earn dollars or acquire them through foreign exchange. But suppose America wanted to purchase $100 million worth of French goods. America paid — in dollars. The very reserve asset everybody else had agreed to hold.

That gave the United States an extraordinary advantage. De Gaulle wanted a system with a neutral settlement asset that no individual country could manufacture: gold.

And France didn’t merely complain. France started redeeming its dollars. For gold.

France actually takes the gold

This is the part of the story that sounds like something invented later by gold enthusiasts. It isn’t. The documentation survives in the U.S. government’s own archives.

In January 1965, advisers informed President Lyndon Johnson that France intended to convert roughly $300 million of excess dollar holdings into American gold.

A few months later, Treasury officials reported another large decline in the U.S. gold stock, including hundreds of millions of dollars’ worth being supplied to France.

And on June 21, 1966, Treasury Secretary Henry Fowler wrote Johnson that French gold purchases during the first half of 1966 would total $324 million. Fowler explicitly described French policy as converting its dollar accruals into gold.

Those aren’t allegations. They’re Treasury memoranda.

France was effectively saying: “You tell us the dollar is worth gold at $35 an ounce. Fine. We’ll take the gold.”

And that’s when the conceptual problem became a physical one. Gold was leaving America’s reserves.

The world’s biggest bank run

Imagine a bank that has issued vastly more redeemable claims than the asset available to redeem every claim simultaneously. That’s fine as long as depositors remain confident. It becomes a problem if everyone wants the underlying asset at once.

Bretton Woods contained a similar vulnerability. Foreign dollar holdings had become enormous. America’s gold stock was finite. If enough countries followed France’s example, the United States faced an increasingly difficult choice: allow the gold reserves to drain away, or change the rules.

Washington spent much of the 1960s trying to avoid that choice. One important defensive mechanism was the London Gold Pool.

1968: the warning shot people forget

The United States and several European central banks cooperated to maintain the market price of gold near the official $35-per-ounce level. When private demand drove gold upward, members of the London Gold Pool could sell gold into the market — essentially defending the official price.

But by 1967–68, pressure became intense. Sterling had been devalued. Confidence in currencies was deteriorating. Demand for gold surged. Central banks found themselves selling increasing quantities into the market to maintain the official price.

Then, in March 1968, the London Gold Pool collapsed.

The system was replaced by an extraordinary compromise: a two-tier gold market. Official transactions among monetary authorities could continue at $35. The private market could trade gold at whatever price buyers and sellers established.

Think about the contradiction. Officially: gold = $35. Meanwhile, the free market was increasingly saying: no it isn’t.

The clock was running.

Meanwhile, look at the coins

This is where the chronology becomes almost cinematic. Look at what was happening during precisely these years.

1964 — The Kennedy half-dollar I found: 90% silver.

1965 — America eliminates silver from dimes and quarters. Kennedy halves fall to 40% silver.

1967 — Canada begins its final transition away from traditional silver circulation coinage.

1968 — Canada’s dimes and quarters transition through 50% silver into nickel. The London Gold Pool collapses.

1969 — The Denver Mint strikes the last 40%-silver Kennedy halves intended for normal circulation.

1970 — The U.S. Mint strikes another 4.6 million 40%-silver Kennedy halves — but only for proof and uncirculated collector sets.[^1]

1971 — Regular Kennedy halves become copper-nickel clad.[^1] And then... Nixon closes the gold window.

That’s a remarkable seven-year sequence. Silver disappears from people’s pockets while the international gold system simultaneously comes apart above their heads. Different policies. Different causes. But both represent the rapid disappearance of precious metal from the monetary architecture.

August 15, 1971

On a Sunday evening, President Richard Nixon addressed the nation on television. Among several economic measures, he announced that he had directed Treasury Secretary John Connally to suspend the convertibility of the dollar into gold.

Nixon described it as temporary. It wasn’t.

The United States would no longer honour the central Bretton Woods promise by giving foreign monetary authorities gold at $35 an ounce in exchange for dollars.

Attempts were subsequently made to construct a new system of fixed exchange rates. They failed. By March 1973, major currencies were floating against one another. The postwar monetary order no longer possessed a formal metallic anchor.

The dollar was not redeemable into silver. It was not redeemable into gold. It was money because the United States government denominated its obligations and taxes in dollars... because the Federal Reserve managed the monetary system... because America’s productive economy, political institutions and taxing capacity stood behind it... and, ultimately, because people throughout the world were willing to accept it.

Money had become primarily institutional rather than metallic.

Was Nixon doing what Roman emperors had done?

This is where the analogy can easily become silly.

No. Nixon wasn’t Nero. Lyndon Johnson wasn’t Caracalla. The Federal Reserve isn’t a Roman mint.

A modern fiat currency is a more sophisticated instrument than ancient commodity coinage, and gold standards were never free lunches either — they could transmit deflation and force brutal adjustments when gold flowed out of a country. Bretton Woods had a genuine structural flaw of its own.

So this isn’t a simple morality tale where gold is virtuous and paper is corrupt. But it isn’t a wash, either. The point isn’t that commodity money was perfect. The point is what happens to the people holding power once the physical limit is gone — and history isn’t ambiguous about that part.

The issuer of money makes promises. Those promises grow. A physical commodity constrains how far the issuer can go. And every single time that constraint became inconvenient, someone with the authority to remove it did.

Rome reduced the silver. Rome also changed the monetary units themselves. Canada reduced the silver. America reduced the silver. Roosevelt weakened the domestic gold constraint. Bretton Woods retained an international gold constraint. France tested it. The London Gold Pool tried to defend it. Nixon finally removed it.

The mechanism changes. The tension doesn’t.

What restrains the issuer?

That’s the question I was ultimately left with.

Under commodity money, there is a physical restraint. You can’t create silver that doesn’t exist. You can’t print gold. The metal didn’t care about your politics, your election cycle, or your war. It simply wasn’t there once you’d used it up.

Under fiat money, that physical restraint disappears. And I understand why economists like this. A central bank can create liquidity during a banking crisis. Governments don’t have to allow their economies to collapse merely because gold reserves leave the country.

But notice what replaced the restraint.

Not a better restraint. A promise. Independent central banks, fiscal discipline, credible government — all of it resting on the willingness of human beings holding power to restrain themselves.

History already answered the question of how well that tends to go.

Rome didn’t run out of silver because the silver disappeared on its own. Emperors debased the denarius because debasing it was easier than raising taxes or cutting spending — and because, once the physical constraint stopped being physical, there was nothing left to say no. Roosevelt didn’t reprice gold because $20.67 was scientifically wrong. He repriced it because $35 let him do more of what he wanted to do. Nixon didn’t close the gold window because gold stopped mattering. He closed it because the alternative was admitting the promise couldn’t be kept.

Every single time, the constraint wasn’t removed by accident. It was removed because someone with power found it inconvenient.

That’s the part I don’t think we’re allowed to wave away with words like “institutional discipline.” Institutions are just people. And the metal enforced discipline on people whether or not they felt like being disciplined that day. Trust doesn’t. Trust is a preference, held by whoever’s in charge, and preferences change with elections, wars, and deficits.

The printing press doesn’t stop rolling because we asked it to. It stops rolling when something outside the people running it makes it stop. For most of human history, that something was a piece of metal that didn’t care what anyone wanted.

We took that away. What we replaced it with isn’t a mechanism. It’s a hope.

And that’s why my old dime is worth more than ten cents

There is a wonderful irony sitting on my kitchen table.

One of my old Canadian dimes says 10 CENTS. Legally, that’s exactly what it is. Ten cents.

Yet nobody who understands the coin would willingly sell it for ten cents. Why? Because the silver inside it is worth many times the number stamped onto its face.

The government says: 10 cents. The silver says: try again.

For most of monetary history, those two concepts — the declared value of money and the commodity value embodied in it — were intertwined. Over the twentieth century, we progressively separated them. First for ordinary citizens. Then in circulating coinage. Eventually between governments themselves. And by 1971, the world’s dominant reserve currency no longer promised conversion into precious metal at all.

I started this exercise trying to determine whether the dull clink of one coin and the clear ring of another could tell me which one contained silver.

It turns out they were telling me something much bigger. They were marking the boundary between two eras of money. One in which value was partly contained inside the coin. And another in which value resides principally in the institutions behind it.

Sometimes history doesn’t arrive with tanks, revolutions or declarations.

Sometimes it arrives in your pocket change.

One postscript, for now. Canada’s dimes lost their silver in 1968. Canada’s dollar lost its gold nearly forty years before Nixon’s dollar did. And what Canada did with its gold after that — while the ground beneath this country kept giving up more of the stuff every single year — is a story too strange, and too infuriating, to fit in the margins of this one.

Stay tuned.

Regards,
Luke Kandia

Luke Kandia writes Chasing Clarity, a Substack covering healthcare systems, financial markets, public policy, and science communication — usually by starting with something small (an old coin, a data point, a footnote) and following it wherever it actually leads.

Nothing here is investment, financial, or legal advice. Do your own homework before you make decisions with your money.

The monetary transition at a glance

Sources & further reading

[^1]: U.S. Mint — First Striking of New Half Dollars and Mint Resumes Production of Half Dollars — the 1965 announcement of the Kennedy half’s reduction to 40% silver, and the record showing no circulating Kennedy halves after December 1969 until clad halves appeared in 1971 (1970 silver production went into collector sets only). See also the 1970 Coinage Law.

[^2]: Royal Canadian Mint — The Dime and its Colourful History — Canadian silver transition, the 1947 Maple Leaf story, and Alex Colville’s 1967 Centennial dime.

[^3]: U.S. Mint — First Striking of New Half Dollars (December 1965 announcement).

[^4]: Royal Canadian Mint — 25-cent technical specifications — documents overlapping 50%-silver and nickel quarter production in 1968.

[^5]: McKinsey Global Institute — “An exorbitant privilege” — origin of Giscard d’Estaing’s phrase.

Additional background: Federal Reserve History, “Gold Reserve Act of 1934,” “Creation of the Bretton Woods System,” and “Nixon Ends Convertibility of U.S. Dollars to Gold”; U.S. Department of State, Foreign Relations of the United States, 1964–1968, Volume VIII (Treasury and White House memoranda on French dollar-to-gold conversions, including Secretary Fowler’s June 1966 memorandum reporting $324 million in French gold purchases in the first half of that year).

This morning, Bill Bonner wrote about watching a wildfire close in on his farm in Poitou in his article, “Where there’s smoke.” He used it, as he does, as a way into a bigger question: how much smoke is there really behind the AI buildout, and is anyone plowing a firebreak before it reaches the house?

Dan Denning’s research note, tacked onto the end of that piece, put a number on it: $1.65 trillion in hidden debt at five U.S. tech giants, against $1.35 trillion they report openly. The number comes from Nikkei. It went viral within a day of publication, and it deserves more than a headline’s worth of scrutiny before anyone repeats it — including that $1.35 trillion comparator. Nikkei doesn’t publish what’s in it, and it’s larger than these five companies’ combined funded borrowings alone, so it likely blends debt with recognized lease liabilities or other financing obligations. Worth treating as a rough benchmark, not an audited “reported debt” figure, until its composition is published.

So we went and read the filings.

What Nikkei actually found

Nikkei’s reporter, Kohei Yamada, examined the financial statements of Alphabet, Microsoft, Amazon, Meta and Oracle and estimated their combined off-balance-sheet debt at $1.65 trillion for the most recent quarter — an eightfold increase in roughly four years. Nikkei is upfront that the total “includes some estimates.” There is no published line-by-line methodology. That doesn’t make the number wrong. It means the number is a well-sourced estimate, not an audited figure, and should be cited as one.

The mechanism behind it is straightforward. When a tech company signs a long-term lease for a data center that hasn’t been built yet, or commits to buy GPUs that haven’t been delivered, accounting rules don’t require that obligation to appear on the balance sheet until the asset shows up. Until then, it lives in a footnote. It’s completely legal. These are real contractual exposures, but their enforceability, timing, cancellation rights and potential loss vary substantially by instrument — a breached lease or purchase agreement and a defaulted bond create different legal remedies and contagion risk. A bond default can trigger acceleration and cross-default provisions across a company’s other debt; a commercial contract breach depends heavily on the specific termination rights, mitigation obligations and damages clauses in that contract, and doesn’t carry the same automatic chain reaction. That distinction matters for how bad a downturn would actually have to get before these obligations convert from inconvenient to genuinely destabilizing.

What we could verify ourselves

We pulled the primary SEC filings for as many of the five companies as we could, rather than taking Nikkei’s aggregate on faith.

Meta checks out almost exactly. Its most recent 10-Q discloses $182.9 billion in leases not yet commenced and $237.7 billion in non-cancelable contractual commitments — a combined $420.5 billion, which lines up almost to the dollar with the $420 billion Nikkei reported for Meta specifically. This is the cleanest confirmation in the whole chain.

Oracle is close but not identical, and the gap is likely timing, not error. Oracle’s own filing data shows $261.0 billion of additional lease commitments not on its balance sheet as of February 28, 2026 — the end of its fiscal third quarter. Nikkei’s figure of $273.3 billion is dated to the end of May, Oracle’s fiscal year-end. Given that Oracle’s backlog grew by $85 billion sequentially between its third and fourth fiscal quarters, it’s entirely plausible the lease-commitment figure grew similarly in that window, landing closer to Nikkei’s number by year-end. That backlog growth includes software and services revenue alongside data center commitments, though, so it isn’t a direct proxy for lease growth specifically — this remains our inference, not something confirmed line-by-line, but it’s the most reasonable explanation for the gap, and the direction of change is unambiguous.

Microsoft and Alphabet don’t get individual figures in Nikkei’s article at all — the piece only gives a combined cloud-backlog number for the three companies together. Separately, Microsoft’s own filing shows $196.6 billion in datacenter leases not yet commenced as of March 31, 2026, up from $92.7 billion at its fiscal year-end nine months earlier — more than double. Alphabet’s most recent 10-Q, for the quarter ended March 31, 2026, shows $75.6 billion in similar leases not yet commenced — up from $58.5 billion at the end of 2025 — plus a $9.9 billion power-purchase agreement signed in January.

Amazon doesn’t produce a comparable number at all. Its disclosure architecture bundles leases, purchase obligations and already-recognized debt into one $569.3 billion total, which means Amazon’s true off-balance-sheet-only slice can’t be cleanly extracted from what it discloses. That’s a structural difference from its four peers, not a smaller number necessarily — just an unmeasurable one from outside.

The layer underneath the headline number

The part of this story that doesn’t show up in any of the topline figures — Nikkei’s, Moody’s, or Morgan Stanley’s — is guarantee exposure.

Meta’s Louisiana data center joint venture with Blue Owl Capital comes with a residual value guarantee — Meta itself discloses an aggregate threshold of approximately $28 billion, decreasing over time, covering a defined shortfall between the facility’s fair value and that threshold under specified termination or non-renewal conditions. Moody’s contribution isn’t discovering that figure — it’s in Meta’s own filing — but flagging how this category of guarantee creates economically debt-like exposure that stays unrecognized as long as payment is judged improbable. Because Meta judged that payout “not probable,” none of it is counted in the $420.5 billion figure above. It exists, on paper, one contingency deeper than the number everyone’s citing. (Meta’s filing also discloses total maximum exposure to the joint venture — equity, future funding, leases and the RVG combined — of roughly $46 billion, for anyone who wants the fuller picture.)

Oracle’s filing discloses something similar in miniature: a guarantee of up to $2.2 billion of a lessor’s own borrowing, maturing this September. If the landlord’s financing goes bad, Oracle is on the hook for it directly — not just for its own lease payments.

This is the mechanism the Bank for International Settlements labeled “shadow borrowing” back in March, and it’s a sharper version of Bonner’s railroad analogy than the raw dollar figure conveys. Nineteenth-century railroads didn’t just borrow to build track — they issued new debt to service old debt when traffic didn’t keep pace, in what Bonner called “the dance of the doomed.” A guarantee that sits off the off-balance-sheet number is a modern cousin of the same instinct: push the exposure one layer further from view, not because anyone’s hiding anything illegal, but because the incentive to understate risk during a boom is exactly the same one it’s always been.

Nikkei’s own article draws the Enron comparison directly, with a caveat: Enron, “though fundamentally different from tech giants,” collapsed because off-balance-sheet debt was hidden behind shell companies, and low-transparency joint ventures can raise the same category of concern even when the accounting is done properly.

Oracle is the sharpest edge of this

Of the five companies, Oracle carries the most concentrated version of this risk, and it’s the one place where a company’s own SEC filing — not an analyst’s inference — puts real language behind it.

Oracle’s fiscal 2026 annual report states plainly that if it overestimates customer demand, or a key customer can’t pay, it “could be locked into multi-year commitments for excess data center space and related capital expenditures... without receiving corresponding revenue.” A separate paragraph adds that some of its customers “may be highly leveraged,” and that in certain OCI offerings, Oracle is “more concentrated among a number of large customers, which could increase these risks.”

Oracle does not name OpenAI anywhere in that risk factor. Every report tying that specific language to the OpenAI relationship — including a Bank of America analyst estimate, cited in subsequent press coverage rather than in a report we’ve read directly, that more than half of Oracle’s $638 billion backlog traces to that one customer, and S&P’s decision to cut Oracle’s credit rating to BBB-, one notch above junk, citing concentration risk — is analyst inference layered on top of a filing that speaks only in general terms. The inference is reasonable given the scale of the Stargate contract relative to everything else in Oracle’s backlog. It is still an inference, not a disclosed fact, and the distinction matters if you’re the one making the claim. The Bank of America figure in particular should be treated as a secondhand estimate until we can trace it to the original research note.

What Oracle’s structure amounts to, in plain terms: it’s taking unusually concentrated anchor-customer risk, committing capital and long-duration capacity years before the matching revenue arrives, with a significant share of expected demand believed by analysts — though not confirmed by Oracle itself — to depend on one customer. That’s a difference of degree from Meta, Microsoft, Google and Amazon, not a difference in kind — all four also build infrastructure partly for external customers and partners. Oracle’s version of that risk is simply far more concentrated, financed further ahead of the revenue that’s supposed to justify it, and resting more heavily on a single relationship than any of its four peers.

When Concentration Risk Stopped Being Theoretical

The concentration risk described in Oracle’s own filing language stopped being an abstract “could happen” scenario in the second half of 2026 — the market started pricing it in real time.

S&P Global cut Oracle’s credit rating to BBB- on July 9, 2026, a single notch above junk, citing OpenAI’s roughly 50% share of Oracle’s $638 billion backlog as the specific driver. Oracle’s stock fell to a 52-week low near $131 in the aftermath, down about 28% over the preceding month.

That wasn’t an isolated event. In late June, a New York Times report that OpenAI was weighing a delay to its IPO — reportedly to protect a valuation target rather than pursue liquidity sooner — knocked Oracle down more than 2.5% on June 26, part of a slide that left the stock over 40% below its year-to-date high, per Barchart’s coverage of the move. About two months before that, a Wall Street Journal report that OpenAI had missed its own internal revenue and user-growth targets dropped Oracle 4% in a day on April 28; SoftBank, one of OpenAI’s largest investors, fell roughly 10% on the same news in Asian trading, per CNBC’s reporting at the time.

None of these three events is proof that OpenAI can’t pay. They’re proof that Oracle’s stock now trades, in part, as a referendum on OpenAI’s ability to keep its commitments — which is exactly the concentration risk Oracle’s own filing describes in general terms, now visibly priced by the market rather than sitting quietly in a footnote.

One claim worth flagging rather than repeating as fact: some coverage has characterized Oracle’s most recent annual report as directly naming OpenAI in a disclosure about possible non-payment. We haven’t been able to independently trace that specific framing to Oracle’s own filing text — what Oracle’s filing does say, in the general concentration-risk language already covered above, doesn’t name any customer. Treat the naming claim as media characterization until it’s confirmed against the primary document.

Where that leaves the $1.65 trillion

Three different measurements, not one:

• The directly documented four-company lease subtotal, and how we got there. Summing just the lease-not-yet-commenced figures pulled directly from filings above — Meta’s $182.9B, Microsoft’s $196.6B, Alphabet’s $75.6B, and Oracle’s $261.0B — comes to roughly $716 billion, across four of the five companies, each measured as of a slightly different date between February and March 2026. That excludes Amazon entirely, since it does not separately disclose an uncommenced-lease subtotal directly comparable with the other four, and excludes Meta’s additional $237.7B in non-lease purchase commitments to keep the four companies on a like-for-like lease-only basis. Worth being precise about what this number is and isn’t: it’s a real subtotal of one category — uncommenced leases — not a reconstruction of Nikkei’s $1.65 trillion, which appears to also include purchase commitments, guarantees, and structured financing exposure that this subtotal deliberately leaves out. Treat $716B as a floor under the lease category specifically, not under Nikkei’s broader figure.

• Nikkei’s $1.65 trillion is the working press estimate. It held up almost exactly where we could check it against a primary filing (Meta) and diverged modestly where we checked another (Oracle). Cite it as Nikkei’s estimate — not as an audited total.

• Morgan Stanley’s research points to an even larger number, but it’s answering a different question. Its report, “Bridging a $1.5tr Data Center Financing Gap,” projects roughly $2.9 trillion in global data center capital expenditure through 2028, of which hyperscalers can cover about $1.4 trillion from their own cash flow — leaving a $1.5 trillion gap the bank expects credit markets, chiefly private credit, to fill. That’s a forward-looking, industry-wide financing need, not a present-day estimate of hidden debt at five companies, and it shouldn’t be cited as a “ceiling” on the same scale as Nikkei’s number. It’s a related but genuinely different measurement, and worth including here only because it shows independent Wall Street research arriving at comparable magnitude by an entirely different route.

None of these figures capture the guarantee layer sitting beneath them. That’s the part worth watching, because it’s structured — not maliciously, just by the ordinary incentives of a boom — to stay invisible until a downturn makes it visible for everyone at once.

Whether any of this resolves as Bonner’s railroads did, or as something milder, comes down to one unresolved question none of these filings can answer: does the revenue that’s supposed to arrive against these obligations actually show up on schedule. That’s not an accounting question. It’s a demand question, and it’s the one none of us — Nikkei, Oracle, Moody’s, or this newsletter — can settle from a footnote.

Follow the dollar, if you can find it

Here’s the part that’s less alarming and more absurd, once you trace it all the way through.

SoftBank, Amazon, Nvidia and others — including continued Microsoft participation — funded OpenAI’s $122 billion round in March 2026. That money is financing, not revenue: it sits on OpenAI’s books as capital raised, not as sales. OpenAI then uses a portion of it, over time, to pay for compute capacity from Oracle — and when that payment happens, it becomes Oracle’s revenue, not OpenAI’s. Oracle, in turn, points to that revenue commitment — its backlog — as the commercial justification for taking on $261 billion in long-duration lease commitments with landlords and lenders who are financing the underlying construction. That’s a real distinction worth holding onto: Oracle’s backlog isn’t collateral in the ordinary sense, and Oracle didn’t literally borrow $261 billion in cash — the number represents undiscounted future lease payments, not a loan proceeds figure. It functions as the expected revenue stream used to justify long-duration capacity commitments, which is a related but different thing from a bank handing over cash. Microsoft, meanwhile, one of several backers of OpenAI, is separately building $196.6 billion of its own competing capacity — partly so it needs Oracle less.

It’s tempting to describe this as the same dollar being counted four times — as OpenAI’s financing, as Oracle’s backlog, as the justification behind Oracle’s lease commitments, and as Microsoft’s strategic position — but that’s not quite right, and the real story is more interesting than double-counting. This is ordinary double-entry accounting: every supply chain generates a matching chain of financing inflows, sales revenue, capital expenditure and investment assets across different entities. When Boeing buys steel from a supplier financed by its own bank, the same economic activity shows up as the supplier’s revenue, the bank’s loan asset, and eventually Boeing’s aircraft revenue too. Nobody calls that fraud.

What makes the OpenAI-Oracle-Microsoft loop worth watching isn’t that money is being duplicated — it’s that a steel supplier has hundreds of customers across dozens of industries, so if Boeing falters, the supplier has somewhere else to sell. Here, the customer, the landlord, the lender, and the strategic investor are all leaning on the identical, single, largely untested thesis: that AI demand grows fast enough to justify what’s being built ahead of it.

That’s correlated exposure, not duplicated accounting — a genuinely narrower and more defensible claim than “shell game,” even if “shell game” is the phrase that comes to mind first. Every step here is disclosed, audited, and legal. It’s just that “disclosed, audited, and legal” doesn’t mean the same thing as “diversified” — and diversification, not disclosure, is what actually protects a lender when a bet goes wrong.

Hype, or Profit?

This is, in the end, the question Bill and the rest of the Bonner Private Research crowd have been circling since long before Nikkei put a number on it: is this hype, or can the thing actually turn a profit? Not “is AI real” — nobody serious argues it isn’t — but whether the specific companies being asked to carry hundreds of billions in obligations can ever generate enough cash to justify them. OpenAI is the cleanest test case, because it’s the customer everyone else in this chain — Oracle, Microsoft, SoftBank — is effectively betting on.

By the numbers we have — third-party reporting, not audited public filings, since OpenAI is still private and only filed a confidential S-1 in June — the company’s revenue ran at roughly $25 billion annualized in early-to-mid 2026, up from about $13 billion at the end of 2025. That’s real, fast growth by any normal company’s standard. It’s also nowhere close to the size of the obligations stacked against it.

OpenAI’s Q1 2026 non-GAAP operating margin was reported at negative 122%, meaning it lost about $1.22 for every dollar of revenue, or roughly $7 billion for the quarter alone.

Simply annualizing that quarterly pace suggests losses approaching $28 billion for the year; one widely cited analyst estimate puts OpenAI’s own internal non-GAAP loss projection lower, near $14 billion, while adding back stock-based compensation — which non-GAAP figures typically exclude — would push the total higher still. The range is wide because the underlying methodology varies by source, not because any one figure is settled. Separately, analyst estimates put OpenAI’s compute obligations across its various partners somewhere in the tens of billions of dollars annually and climbing, on top of the roughly $300 billion, multi-year Oracle relationship discussed above — real numbers, but estimates rather than a figure OpenAI has confirmed itself.

The bull case, as OpenAI itself tells it, is that this is simply what building a generational platform looks like before it matures — Amazon lost money for years before it didn’t. OpenAI has been reported, based on internal materials described to journalists, to be targeting its first cash-flow-positive year around 2029, at revenue somewhere in the neighborhood of $125 billion — a figure that also shows up independently in third-party analyst forecasts for roughly the same year, which is worth noting since it means outside modeling and the company’s own reported plan aren’t far apart. Either way, it implies growing five-fold from today’s run-rate inside three years. Most outside analysts consider that optimistic rather than implausible, which is a real distinction: nobody credible is calling it fantasy, but almost nobody outside the company is calling it the base case either.

So: hype, or profit? The honest answer is that nobody actually knows yet, and anyone claiming certainty in either direction — doomer or true believer — is selling conviction the numbers don’t support.

What the numbers do support is this: OpenAI isn’t paying Oracle’s rent out of profit today, and won’t be for years even by its own optimistic telling. It’s paying it out of the next funding round, and the next one after that, betting that revenue arrives before an investor somewhere in the chain decides it’s tired of writing checks against a five-fold growth story.

That’s not proof of a bubble. It’s also not proof it isn’t one. It’s the actual, uncomfortable, unresolved question sitting underneath every dollar figure in this piece — and it’s the one that won’t get settled by another footnote, another filing, or another AI-generated fact-check. Only time, and whether the users and the revenue actually show up, will settle it.

That’s the useful thing to take from all this, in the end. Nikkei’s $1.65 trillion isn’t hidden because anyone’s hiding it — every figure in this piece came from a public filing. It’s hard to see clearly because so much of it sits behind footnotes, across five different disclosure formats, on five different reporting calendars, describing obligations that don’t behave like debt until, one day, they do. The real number in this story was never a single dollar amount. It’s a question: how much of this is actually the same bet, made by different parties, all counting on the same demand showing up on the same schedule?

Chasing Clarity is an independent publication supported by readers who’d rather understand what’s happening than be surprised by it. If this piece made you think, send it to someone who should be thinking about it too.

Luke Kandia
Chasing Clarity

This piece is for informational and educational purposes only and does not constitute investment, legal, or financial advice. Nothing here is a recommendation to buy, sell, or hold any security. Figures drawn from company filings were independently verified against primary sources at the time of writing; figures attributed to third-party reporting, analyst estimates, or private-company disclosures were not independently audited and may change as new information becomes available. Do your own research, and consult a licensed professional before making investment decisions.

Sources

Primary filings (SEC EDGAR)

• Meta Platforms, Form 10-Q, quarter ended March 31, 2026 — leases not yet commenced and contractual commitments footnote. sec.gov/Archives/edgar/data/1326801/000162828026028526/meta-20260331.htm

• Oracle Corporation, Form 10-K, fiscal year ended May 31, 2026 — Item 1A Risk Factors; lease commitments not reflected on balance sheet. sec.gov/Archives/edgar/data/1341439/000119312526277521/orcl-20260531.htm

• Microsoft Corporation, Form 10-Q, quarter ended March 31, 2026 — leases not yet commenced (as cited in Stifel research note; original filing at sec.gov, CIK 0000789019)

• Alphabet Inc., Form 10-K, fiscal year 2025, and Form 10-Q, quarter ended March 31, 2026 — leases primarily for data centers not yet commenced

• Amazon.com Inc., Form 10-Q, quarter ended March 31, 2026 — total contractual commitments table

Journalism and research reports

• Kohei Yamada, “Five US tech giants’ hidden debts soar to $1.65tn on opaque AI funding,” Nikkei Asia, July 21, 2026. asia.nikkei.com/business/technology/five-us-tech-giants-hidden-debts-soar-to-1.65tn-on-opaque-ai-funding

• Dealroom.co, “The $1.65tn hidden debt behind Big Tech’s AI buildout.” app.dealroom.co/news/note/the-1-65tn-hidden-debt-behind-big-tech-s-ai-buildout

• Analysis.org, “Hidden debt at five AI hyperscalers hits $1.65 trillion, Nikkei study finds.” analysis.org/hidden-debt-at-five-ai-hyperscalers-hits-1-65-trillion-nikkei-study-finds

• Fortune, on Moody’s Ratings February 2026 report estimating $662 billion in off-balance-sheet hyperscaler lease obligations. fortune.com/2026/02/25/hyperscaler-risk-off-balance-sheet-662-billion-data-center-commitments-meta-amazon-microsoft-oracle-alphabet

• Bill Bonner and Dan Denning, “Where There’s Smoke,” Bonner Private Research, July 22, 2026. bonnerprivateresearch.com/p/where-theres-smoke-4fd

• Bank for International Settlements, “Financing the AI infrastructure boom: on- and off-balance sheet borrowing,” BIS Quarterly Review, March 2026. bis.org/publ/qtrpdf/r_qt2603u.htm

• Morgan Stanley Research, “Bridging a $1.5tr Data Center Financing Gap,” July 2025. morganstanley.com/content/dam/msdotcom/en/assets/pdfs/Research_Bridging-Data-Center-Gap.pdf (institutional research report, also referenced in subsequent Morgan Stanley client commentary and press coverage)

• Oracle Corporation, XBRL filing exhibit, “Leases and Other Commitments — Narrative (Details),” as of February 28, 2026. sec.gov/Archives/edgar/data/1341439/000119312526101045/R48.htm

• Bank for International Settlements, Annual Report 2026, Chapter I, “Progress and peril.” bis.org/publ/arpdf/ar2026e1.htm

• TheStreet, “Oracle stock falls for a seventh session as filing risk lands,” on Oracle’s June 23, 2026 annual report risk-factor language. thestreet.com/investing/stocks/oracle-stock-falls-for-a-seventh-session-as-filing-risk-lands

• MLQ News, “S&P Downgrades Oracle to BBB-, One Notch Above Junk, Citing OpenAI Concentration Risk.” mlq.ai/news/sp-downgrades-oracle-to-bbb-one-notch-above-junk-citing-openai-concentration-risk

• Yahoo Finance / Simply Wall St, “OpenAI Shortfall Puts Oracle AI Cloud Commitments And Concentration Risk In Focus,” April 2026. finance.yahoo.com/markets/stocks/articles/openai-shortfall-puts-oracle-ai-050252163.html

• Barchart, “Why Reports of a Delayed OpenAI IPO Are Dragging Down ORCL Stock,” June 26, 2026 (via Yahoo Finance). finance.yahoo.com/markets/stocks/articles/why-reports-delayed-openai-ipo-204443007.html

• CNBC, “OpenAI reportedly missed revenue targets. Shares of Oracle and these chip stocks are falling,” April 28, 2026. cnbc.com/2026/04/28/openai-reportedly-missed-revenue-targets-shares-of-oracle-and-these-chip-stocks-are-falling.html

• BigGo Finance, “Oracle’s AI Bet Shows Cracks: Fears OpenAI Can’t Pay $9.6 Trillion Bill Trigger 35% Monthly Stock Plunge,” reporting on Oracle’s annual report risk language — the source of the “Oracle disclosed OpenAI non-payment risk directly” framing this piece treats as unconfirmed media characterization rather than fact. finance.biggo.com/news/607e4bf9-96fc-4904-93aa-248a9f22cbf3

OpenAI financials (reported, not audited public filings — OpenAI is privately held and filed a confidential S-1 on June 8, 2026)

• The Information, reporting on OpenAI Q1 2026 financials (revenue, operating margin) as aggregated by Wheresyoured.at, “OpenAI Had A Negative 122% Non-GAAP Operating Margin In Q1 2026.” wheresyoured.at/news-openai-had-a-negative-122-operating-margin-in-q1-2026-and-chatgpt-growth-has-stalled

• futuresearch.ai, “OpenAI Revenue, Losses, and IPO Valuation: Forecasts Through Late 2027.” futuresearch.ai/openai-financial-forecast

• ValueAddVC, “OpenAI Revenue 2026: $25B ARR and Still Losing $14B a Year.” valueaddvc.com/blog/openai-revenue-2026-20b-arr-4b-month-path-to-profitability

A note on sourcing: every figure attributed to a specific SEC filing above was independently pulled and read from EDGAR before inclusion. The OpenAI financials are third-party reporting on a private company’s internal figures and should be read with that caveat — they are the best publicly available numbers, not company-confirmed audited results.

I thought I understood hydration.

Lose sweat. Replace water. Replace sodium. Keep running.

Simple.

Or so I thought.

Then, after one long run, I looked down at the white streaks on my blue running shirt.

Salt.

Every runner has seen it. Every runner has tasted it.

And I asked myself a question that, embarrassingly, I had never considered before.

That simple question sent me down a rabbit hole I never expected.

It started with marathon hydration. It ended with one of the most beautiful pieces of chemistry I have ever learned.

Share

Salt doesn’t actually exist inside your body

Take a grain of table salt.

Chemically, it’s sodium chloride. NaCl.

Drop it into a glass of water.

Within seconds...

the crystal disappears.

Not because it has been destroyed. Because water has quietly pulled it apart.

Water molecules are tiny electrical dipoles.[1] The oxygen end carries a slight negative charge. The hydrogen ends carry slight positive charges.

Every water molecule behaves like a microscopic electrical dipole. The negatively charged oxygen is attracted to sodium. The positively charged hydrogens are attracted to chloride.

Millions of water molecules immediately surround the crystal. They begin pulling on every sodium ion. They begin pulling on every chloride ion.

Eventually, the attraction between water and the ions becomes stronger than the attraction holding the crystal together.

The crystal falls apart.

Not into smaller crystals. Into individual ions.

NaCl → Na⁺ + Cl⁻

One grain of salt has become billions upon billions of separate sodium and chloride ions.

And from that moment on...

Your blood contains no dissolved grains of salt

This realization stopped me in my tracks.

My bloodstream isn’t carrying microscopic crystals of salt.

It is carrying sodium ions. It is carrying chloride ions.

Moving independently. Doing completely different jobs.

Sodium helps regulate fluid balance. It allows nerves to fire. It helps muscles contract.

Chloride balances electrical charge. It helps regulate acid-base balance. It helps produce stomach acid.

Together they maintain the chemistry that keeps us alive.

If they stayed locked together as crystals...

nothing would work.

Life depends on them being separated.

The heat problem

Before the salt, there’s a simpler problem: heat.

Every muscle contraction that pushes you forward also generates heat as a byproduct. Muscle is only about 20 to 25 percent efficient at turning fuel into motion. The rest — the majority of the energy you’re burning on a run — becomes heat, generated right there inside the muscle, with nowhere to go but into your bloodstream.[2]

Your core temperature runs in a narrow, non-negotiable range. The enzymes that run essentially every process in your body work properly only within a few degrees of 37°C. Push core temperature much past 40°C and those proteins start to lose their shape — the same unfolding that happens when you cook an egg, just slower, and happening inside you.

So somewhere in your brain, a small structure called the hypothalamus runs a constant thermostat check, reading signals from your blood and your skin, and deciding, moment to moment, how urgently that heat needs to leave.

Heat leaves the body four ways.[2]

Conduction — direct contact, heat moving into whatever’s touching your skin. Convection — heat carried away by moving air or water passing over your skin. Radiation — heat radiating outward as infrared energy, the same way a wood stove warms a room without touching it.
Evaporation — heat absorbed by water as it turns from liquid to vapor, and carried away with it.

At rest, in a comfortable room, radiation often does most of the work. You are simply glowing much of your excess heat away as infrared energy, most of the time without sweating at all.

Running changes the heat equation completely. Your working muscles can produce ten to twenty times more heat than your body produces at rest.

Normally, some of that heat escapes through radiation and convection because your skin is warmer than the surrounding air and surfaces. But as the environment approaches skin temperature, that difference becomes smaller, so radiation and convection remove less and less heat.

Once the air becomes warmer than your skin, convection reverses direction. Instead of carrying heat away, the air begins transferring heat into you. Radiation can reverse as well when the sun, pavement, buildings or other surrounding surfaces are radiating more energy toward your body than your body can radiate back.

The simplest summary is this: running creates far more heat, while hot weather disables two of the body’s main escape routes. At that point, evaporation of sweat becomes the body’s primary effective method of cooling.

Evaporation doesn’t care about that gradient. It runs on humidity, not temperature. Water still evaporates off your skin even when the air is hotter than you are, as long as that air isn’t already saturated with moisture.

Which is exactly why a marathon runner in 20°C weather and someone standing still in a festival crowd in 35°C heat can both end up drenched in sweat — for almost opposite reasons.

The runner’s problem is mostly internal: a huge, self-generated heat load from working muscle, met by air still cool enough to let radiation and convection help rather than hurt, and dry enough to keep the evaporation gradient strong. Sweat is doing real work, but it isn’t working alone.

The festival-goer’s problem is mostly external: almost no internal heat load to speak of, but standing in air hot enough that radiation and convection have flipped sides — and, if the crowd is humid, which outdoor festivals often are, an evaporation gradient too weak to carry much away. The one exit still technically open is the one now least able to do anything with it.

A runner in cool, dry air might have three of four working in their favor. Someone standing still in a hot, humid crowd might have exactly one, and it’s barely functioning.

Which is why, once the thermal math stops working in your favor, your hypothalamus reaches for the one mechanism that still functions no matter how hot it gets outside. It triggers your eccrine sweat glands.

Your sweat gland performs a remarkable trick

Then the story became even better.

Quick detour, because the word matters: eccrine just means “secreting outward.” It’s the ordinary, all-over sweat gland — millions of them, covering nearly your entire skin — doing the actual thermoregulatory work. The kind responsible for the salt streaks on your shirt.

There’s a second type, apocrine, and it’s a different animal entirely. Apocrine glands cluster in the armpits and groin, attach to hair follicles, and produce a thicker, protein-and-lipid-rich secretion that bacteria break down into body odor. They’re not really in the cooling business, and they have nothing to do with this story. When people talk about “sweating it out” on a run, they mean eccrine glands. So does this article, from here on.

Inside an eccrine sweat gland, chloride and sodium don’t even take the same road.

A transporter on the outside of the gland’s secretory cells — called NKCC1 — actively pumps chloride out of the surrounding tissue fluid and into the cell. The energy for that ultimately comes from the same sodium-potassium pump every cell in your body runs. Once inside, chloride flows onward through a channel called CFTR into the fluid-filled core of the coil — that second step isn’t pumped, it just flows downhill along its own gradient.

Sodium never goes through the cell at all. The sodium that gets dragged in alongside chloride gets pumped straight back out again, to keep the cycle running. The sodium that actually ends up in your sweat slips in through the narrow gaps between cells, pulled along by the electrical charge chloride leaves behind as it moves into the coil.

Water follows both of them through its own dedicated channels, drawn along osmotically.

Sweat is born.

As the sweat travels toward the skin, the body tries to reclaim much of the sodium and chloride. Some is recovered. Some isn't — and the two aren't reclaimed at exactly the same rate, or in the same proportion sweat-to-sweat. Potassium, lactate, and bicarbonate are riding along too, competing for the same electrical bookkeeping. The harder and longer you sweat, the less time the duct has to reclaim anything, so sweat gets saltier the more you produce.[4]

The remaining water and ions emerge onto the surface.

Not as salt.

As dissolved sodium ions... and dissolved chloride ions... and a supporting cast of potassium, lactate, and a few other stowaways.

Sweat doesn’t cool you. Evaporation does.

This may be the biggest misconception in running.

Liquid sweat sitting on your skin isn’t doing very much.

The cooling happens when the water molecules escape into the air. Every molecule that evaporates steals heat from your body.

No evaporation. No cooling.

Which explains why running in humid weather feels so miserable.

You may be producing enormous amounts of sweat...

while very little of it is actually evaporating.

You’re losing water. You’re losing electrolytes. But you aren’t getting the cooling you paid for.

Then something extraordinary happens

As the water disappears...

the sodium ions remain. The chloride ions remain.

The water molecules that had been keeping them apart gradually disappear into the atmosphere.

The shielding is gone.

Now the sodium and chloride begin finding each other again.

Not because they recognize one another. Because opposite electrical charges attract.

One by one...

they recrystallize — mostly as sodium chloride, mixed in with whatever potassium and other minerals were riding along in the same drop of sweat.

The white streak on your running shirt... The crust around the brim of your cap... The salt on your forearm...

is newly formed sodium chloride, alongside its supporting cast.

Not the same crystal you dissolved yesterday. A brand-new crystal — smaller and messier than a lab specimen, but real.

Built from atoms that travelled through your bloodstream... helped cool your body... and then found each other again as the water disappeared.

I find that breathtaking.

And then I turned over the labels...

This is where my original question returned.

If sweat leaves my body as water carrying sodium and chloride...

why do hydration products almost always tell me only about the sodium?

Before I get to the products, it’s worth pinning down what “the ratio sweat actually loses” even means, since the rest of this piece measures everything against it. Sweat sodium concentration varies enormously between people and conditions — roughly 20 to 80 mmol/L is the typical range reported in the exercise physiology literature, with chloride tracking close behind sodium, usually at a slightly lower molar concentration rather than an exact 1:1 match.[4] That’s a wide range, not a fixed target, which is part of the point: there’s no single “correct” sodium-chloride number a product should hit, but there is a relationship — chloride roughly tracking sodium — that almost none of the ten products below are built around.

One more thing worth being upfront about before the rankings: in the literature I reviewed through July 2026 — not a formal systematic review — I didn’t find a published study tying chloride under-replacement specifically to documented harm or impaired performance in runners or endurance athletes. This piece’s argument isn’t “you will get hurt if your electrolyte drink is chloride-light.” It’s narrower and, I think, still worth making: the products’ own labels and formulas don’t reflect the physiology of what’s actually leaving your body, and the industry-wide silence on chloride is a labeling and communication problem before it’s ever a demonstrated safety one.

So I bought ten different hydration products, across eight brands.

Liquid I.V. — Original, with sugar.
Liquid I.V. — Sugar-Free.
Electrolyte Gastro (an oral rehydration solution).
LMNT.
Tailwind.
Huma.
XACT.
Nuun.
GU Roctane Energy Gel, caffeinated.
GU Hydration Drink Tabs, no caffeine.

And I started reading labels — ingredient lists, and where available, the actual milligram or milliequivalent numbers.

What I found was more interesting, and more specific, than I expected.

One product declared its chloride outright, rather than making me calculate it. Electrolyte Gastro — a medical oral rehydration solution, not a sports brand, meant for dehydration from diarrhea, vomiting, or heavy sweating — lists 12.1 mEq of sodium and 12.0 mEq of chloride per sachet, which the label itself separately states as 288mg of sodium.[5] In plain English: mEq (milliequivalent) counts ions, not weight — for a monovalent ion like sodium or chloride, equal mEq numbers mean exactly equal numbers of charged particles, no rounding involved. 12.1 versus 12.0 is about as close to a 1:1 sodium-to-chloride count as a real product gets. The chloride figure works out to about 425mg by the same conversion. One reconciliation note worth being upfront about: standard atomic-weight math on 12.1 mEq of sodium gives 278mg, not the 288mg the label itself prints — a roughly 3.5% gap between the company’s own mEq and mg figures that this piece doesn’t have an explanation for; both numbers are reported here as they appear on the label rather than silently corrected. Chloride is the heavier ion, so even a near-perfect ionic match ends up chloride-heavy by weight — a pattern that shows up again below. Worth being precise about what actually makes Gastro stand out: it isn’t the only product that turns out close to a 1:1 molar ratio once you do the conversion — LMNT and Huma Plus, calculated below, land in roughly the same territory. What makes Gastro different is that the 1:1 relationship is declared on the label, not something a reader has to reconstruct from an ingredient list and a periodic table.

One more detail worth knowing if you’d actually use this for exercise rather than illness: the label gives two different mixing instructions depending on why you’re using it. For diarrhea or vomiting, dissolve the sachet in 200mL of water. For heat exposure or prolonged exercise sweating, the label specifically directs dissolving the same sachet in 300mL instead.[5] That’s a real, if modest, acknowledgment that exercise hydration and illness rehydration aren’t quite the same dosing problem — diluted the exercise way, that 288mg of sodium works out to roughly 960mg per litre, which is on the higher end of published exercise sodium targets but not outside them for heavy sweaters in heat, and meaningfully lower than the roughly 1,440mg per litre you’d get mixing it the illness way. The label also specifically warns against combining it with other sports or electrolyte drinks — worth knowing if you’re someone who layers products.

Liquid I.V. turned out to be the most surprising case in the whole list — and not in the direction I expected. Its label breaks sodium down by source: the sugared version gets roughly 409mg of its sodium from plain salt (sodium chloride) and another 128mg from sodium citrate, for 537mg of sodium total.[6]

Salt is only about 39% sodium by weight — the rest, about 61%, is chloride.

Run that math and the sugared version is actually delivering somewhere around 630mg of chloride, more than its sodium content, just never printed as its own number. The sugar-free version follows the same pattern — roughly 580mg of chloride against 504mg of sodium.

Liquid I.V. isn’t hiding chloride. It’s delivering more of it than sodium and getting zero credit for it.

Because nobody — including me, until I did the arithmetic — bothers to convert “salt” on an ingredient list into a chloride figure.

The two versions also lean on different absorption science, and it’s worth being precise about both the mechanism and the current state of the evidence behind each. The sugared version uses the sodium-glucose cotransporter (SGLT1) — the actual mechanism the WHO and UNICEF built the modern oral rehydration solution standard around, with decades of peer-reviewed clinical evidence behind it.[7] The sugar-free version replaces sugar with L-glutamine and L-alanine (plus allulose, a non-metabolized rare sugar, for sweetness), leaning on sodium-amino acid cotransporters in the small intestine — a real, established transport pathway, not invented science.[7]

But the evidence specifically supporting each product’s own claim is at different stages, and it’s worth stating plainly, as of this writing:

A differently-formulated amino acid rehydration product — not Liquid I.V.’s, a separate brand using a different amino acid blend — does have one full peer-reviewed human trial behind it, showing superior fluid restoration versus a generic sports drink after induced dehydration.[17] That trial was funded in part by the company that makes the product, with a co-author serving as that company’s Chief Scientific Officer — a disclosed conflict of interest, not a hidden one, but worth knowing before treating the result as independent confirmation.

Liquid I.V.’s own specific formula has one completed, peer-reviewed, published trial behind it — but it’s a safety and tolerability study, not an efficacy one. Funded by Liquid I.V. and conducted at the University of Memphis, it found the L-alanine/L-glutamine blend well-tolerated over four weeks of daily use, with no meaningful changes to blood chemistry, thyroid function, or vital signs.[18] The efficacy claim that actually matters for this article — that the sugar-free version hydrates as well as the sugar version — currently rests on two Liquid I.V.–funded studies that, as of this writing, have only been presented as conference abstracts at ACSM and ASN annual meetings, not yet published as full, independently peer-reviewed papers.[19,20] Conference abstracts undergo lighter review than a full manuscript. That doesn’t make the findings wrong — reported results showed the sugar-free version performing statistically equivalently to the sugar version, both beating water — but it does mean the specific claim is still in progress, not yet the settled kind of evidence that backs the glucose-SGLT1 route.

Most of the other sports products contain real chloride, but don't say so. LMNT is the cleanest case: its only sodium source is salt, its only potassium source is potassium chloride, and its magnesium is bound to malate, not chloride — which means all of its chloride content is fully determined by the label's own sodium and potassium figures, no estimation required. 1000mg sodium and 200mg potassium, both entirely chloride-paired, work out to 1724mg of chloride — more than the sodium itself, delivered on a label that never mentions the word.[8] Huma Plus (the "Double Electrolytes" fruit-based gel, not Huma's standard energy gel — a distinction that cost me a correction mid-research) has only one plausible sodium source in its ingredient list — sea salt, since the rest is fruit purees and juice concentrates — providing 240mg of sodium.[9] If essentially all of that sodium is chloride-paired, straight stoichiometry gives about 370mg of chloride; allowing for sea salt's small fraction of non-chloride trace minerals brings the honest range down to roughly 355–370mg — either way, more than the sodium content, on a label that lists potassium, calcium, and magnesium by name but never chloride. XACT's ingredient list — sodium bicarbonate, trisodium citrate, and salt for sodium; potassium chloride, alone, for potassium — puts it in a similar position to LMNT on the potassium side: since potassium chloride is its only potassium source, all 100mg of labeled potassium is chloride-paired, working out to a confirmed minimum of about 91mg of chloride from that source alone.[11] Tailwind's packaging tells the same story: sodium citrate and sea salt compete for its 630mg of labeled sodium with no disclosed split, but potassium chloride is again its only potassium source, so its roughly 136mg of potassium (the label states only a %DV, so this is a best estimate within a plausible 119–153mg range) accounts for something like 100–140mg of chloride on its own.[10] On both XACT and Tailwind, the sodium side is murkier — competing sources, no disclosed split — so there's more chloride in there than the calculable minimum, just not a figure that can be pinned down. None of this is concealment, exactly. Neither the FDA (in the US) nor Health Canada (its Canadian equivalent) requires a chloride line on a Nutrition Facts panel unless the company chooses to make a chloride claim — it's genuinely optional on both sides of the border.[12,13] It's a labeling blind spot that the marketing language — "electrolytes," full stop — quietly rides on top of.

Nuun is the interesting middle case. It does declare chloride: 40mg, right there on the label. But nearly all of its 300mg of sodium comes from sodium bicarbonate and sodium carbonate, not sodium chloride — chloride only enters the formula via a small amount of potassium chloride.[14] By mass, sodium outweighs chloride roughly 8 to 1; convert both to millimoles and the gap is even more stark — about 13.0 mmol of sodium against 1.1 mmol of chloride, closer to 12 to 1 — a ratio nothing like what actually leaves your body in sweat.

GU turned out to be two different stories under one brand name. The Roctane Energy Gel gets its sodium from sodium citrate plus a small amount of sea salt — so it likely carries some real chloride via that sea salt, though the ingredient list gives no way to quantify how much.[15] The Hydration Drink Tabs, meanwhile, get their sodium entirely from sodium bicarbonate and sodium carbonate. No chloride-bearing ingredient appears anywhere in that formula.[16] Same company, same shelf, and one product has an unquantified amount of chloride while the other has no identifiable source for any.

Read across all ten, and a pattern holds even though the mechanism varies product to product: almost none of them are built to replace sodium and chloride together, in anything like the ratio sweat actually loses. One — Electrolyte Gastro — replaces both in something close to a declared 1:1 ratio. Four — LMNT, both Liquid I.V. versions, and Huma — deliver more chloride than sodium by weight, unlabeled. Four more — Tailwind, XACT, Nuun, and GU Roctane — contain some real chloride, mostly unlabeled, in amounts well under their sodium content. One — GU Hydration Tabs — contains no chloride at all. And nearly all of them, regardless of what’s actually inside, use the same word on the front of the package.

Here’s the same ten products as a reference table, so you don’t have to hold all of it in your head:

A methodological note on this table, since the normalization itself makes a claim worth stating plainly: prep volumes come from each manufacturer's own stated directions where one exists (Gastro, XACT, Nuun, GU Tabs, Liquid I.V. all give a single fixed number; LMNT and Tailwind give ranges, which is why those two rows show a range rather than a point figure). The two gels don't fit this framework at all — a gel isn't "prepared" to a target concentration, it's eaten alongside water the athlete brings separately, so forcing a per-litre figure onto them would manufacture false precision rather than remove it. That's a real limit of this comparison, not an oversight: it means the electrolyte tablets, powders, and the ORS sachet can be compared apples-to-apples on concentration, while the gels can only be compared to each other and to the rest of the table on a per-serving basis, which is why the earlier per-serving figures are preserved in the two gel rows above.

Electrolytes.

As though that answered the question.

It doesn’t.

Chloride is another. Potassium is another. Magnesium. Calcium. Bicarbonate. Phosphate.

They’re all electrolytes.

The word hides far more than it reveals — not necessarily on purpose, but the FDA’s labeling rules (and Canada’s, under Health Canada) leave room for it to happen, and marketing has no incentive to close the gap.

The discovery

I began this project trying to answer one practical question.

How much sodium do I need during a marathon?

Instead, I discovered something much more specific: outside of medical oral rehydration solutions, almost nothing on a runner’s shelf is actually formulated to replace what sweat loses in the ratio sweat loses it. Not because anyone is hiding it — because nobody standardized the requirement, and the word “electrolytes” lets every formula, no matter how lopsided, claim the same territory.

If I had to rank them

Ten products, one obvious question left unanswered: which one would I actually reach for?

Ranked purely on how faithfully they mirror what sweat loses — sodium and chloride together, in something close to the ratio your body is actually shedding — three stand out from the other seven.

1. Electrolyte Gastro. The only product on the list with a declared, not calculated, near-1:1 sodium-to-chloride match — 288mg sodium against 425mg chloride, 12.1 mEq against 12.0 mEq.[5] It’s not trying to be a sports drink. It’s a rehydration therapy, built for diarrhea and vomiting — and notably, its own label gives a separate, more dilute mixing instruction specifically for exercise sweating rather than treating every use case identically, which is more exercise-awareness than most sports products bother with.[5] It happens to be the most physiologically honest thing on the shelf.

2. LMNT. The highest absolute sodium of the ten, at 1000mg — and, unlike every other product on this list except Electrolyte Gastro, its chloride content isn't a guess. LMNT has exactly one sodium source (salt) and exactly one potassium source (potassium chloride), with no competing source pulling either away from a chloride pairing — magnesium is bound to malate, not chloride, and contributes nothing to the total. That makes the math fully determined: 1000mg sodium and 200mg potassium, both entirely chloride-paired, work out to 1724mg of chloride, more than the sodium itself.[8] The tradeoff for that clean ratio is zero sugar, so LMNT doesn't engage the SGLT1 pathway at all — water and sodium still cross the intestinal wall through other osmotic and transport routes, just without the glucose-driven boost the other two get. Worth flagging plainly: 1000mg of sodium in one serving is a lot in absolute terms, appropriate for a heavy or salty sweater in real heat, less appropriate as a daily habit for someone managing blood pressure or on a sodium-restricted diet for other reasons. This ranking is about matching sweat's chemistry, not a blanket recommendation for how much sodium any given person should be drinking.[7]

3. Liquid I.V. (sugar). A weaker sodium-to-chloride ratio than the top two — part of its sodium rides on sodium citrate, which carries no chloride at all — but it does use glucose-sodium cotransport, the same mechanism the WHO built the modern oral rehydration standard around.[7] Worth correcting an error from an earlier pass of this piece: Liquid I.V. isn’t unique among the top three in using that mechanism — Electrolyte Gastro contains 3.58g of glucose per sachet and is itself built on the same SGLT1 principle, arguably more deliberately than Liquid I.V., since that’s literally what an oral rehydration solution is designed around. LMNT is the actual outlier of the three: zero sugar, no SGLT1 pathway engaged at all. Decades of clinical evidence sit behind the glucose-sodium route specifically, which both Gastro and Liquid I.V. draw on.

One caveat worth being explicit about, because it changes the ranking depending on what you’re optimizing for. This order answers “which product’s chemistry most faithfully mirrors what sweat actually loses” — which is this article’s actual thesis, so it’s the right frame here. It is not the same question as “best product to carry through mile 18 of a marathon.” Electrolyte Gastro is a rehydration therapy, not a fueling product — low sugar by design, built for illness recovery, not for replacing the carbohydrate a runner burns through over three-plus hours. If the ranking criterion shifts to “best for actually getting through a long run,” Liquid I.V. arguably jumps LMNT, since LMNT gives you sodium and chloride and nothing else to fuel the last six miles.

Rankings aside, that was never really the question I started with. I didn’t set out to build a buying guide. I set out because of a white streak on a blue shirt, and a question I’d never thought to ask before.

Because the instant salt touches water...

it ceases to exist as salt.

It becomes sodium. It becomes chloride.

From that moment on, biology treats them as related but distinct — not locked to each other in a fixed ratio, moving and being reclaimed at their own rates through the sweat duct.

And yet, after an extraordinary journey through the human body...

after helping regulate nerves, muscles, kidneys, sweat glands and temperature...

they meet again, mostly, on the surface of your skin.

The water is gone. The heat has escaped. The shielding water molecules have vanished. The opposite charges attract.

And the crystal is reborn — messier than the one you started with, but real.

The next time you finish a long run and notice the white streaks on your shirt...

don’t just see sweat.

You’re looking at a crystal that dissolved...

travelled through a living human body as billions of independent charged particles...

helped keep you alive...

helped keep you cool...

and then quietly, imperfectly, assembled itself once again.

One last practical note, since a chemistry story shouldn’t end without one: everything in this piece measures products against a general physiological pattern, not against your sweat specifically. Sweat sodium concentration varies by a factor of four across individuals — genetics, heat acclimatization, diet, and fitness all move that number. If you’re serious about dialing in your own hydration rather than following a ranking, an actual sweat-sodium test (a few labs and some sports-medicine clinics offer this) will tell you more about your own numbers than any article, including this one, ever could.

That test answers one question: how salty is your sweat. It doesn’t answer a second, equally important one — how much of it you’re actually losing — and that one you can measure yourself, for free, with a bathroom scale. Weigh yourself naked immediately before a run and again immediately after, without eating or drinking in between, and account for any fluid you did drink during the run (add it back to your post-run weight before subtracting). Every kilogram lost is roughly a litre of sweat, and dividing that by your run duration gives you your sweat rate in litres per hour — a number that changes with temperature, humidity, effort, and clothing, so it’s worth doing a few times across different conditions rather than once and calling it settled. Between the two — a lab test for concentration, a scale for rate — you get the full picture no article can hand you: not just what your sweat contains, but how much of it you’re actually replacing.

Regards,

Luke Kandia

Luke Kandia is a writer and researcher based in Winnipeg, Manitoba, a retired IT services owner, and a recreational runner who still can’t leave a question alone once it’s been asked.

This article started with a white streak on a running shirt and a question I’d never thought to ask: if sweat carries salt, why does every hydration product only talk about sodium? What followed was a few weeks of reading labels, doing the arithmetic ingredient lists don’t do for you, and learning more sweat gland physiology than I expected to need. It is not medical advice, a nutrition recommendation, or a verdict on any product or company named here — every figure is either a manufacturer’s own stated number or my own calculation from their ingredient list, clearly marked as one or the other.

Anyone making decisions about hydration, electrolyte intake, or exercise nutrition — especially in heat, especially over long durations — should do so based on their own sweat rate, their own tolerance, and, where it matters, a conversation with a qualified professional, not a magazine-style ranking.

If you’ve ever read a supplement label closely enough to notice what it doesn’t say, or if you’ve got a favorite hydration product I should fact-check next, I’d love to hear about it in the comments.

References

1. Chemistry LibreTexts. 4.4: Hydration of Ions in Aqueous Solutions. Davis College of the Canyons, Chem 201: General Chemistry I OER. Available from: https://chem.libretexts.org/Courses/College_of_the_Canyons/Chem_201:_General_Chemistry_I_OER/04:_Introduction_to_Solutions/4.04:_Hydration_of_Ions_in_Aqueous_Solutions [Confirmed July 2026: covers water’s bent, polar molecular structure and the ion-dipole dissolution of NaCl(s) into Na⁺(aq) and Cl⁻(aq), directly supporting the chemistry described in this section.]

2. Périard JD, Eijsvogels TMH, Daanen HAM. Exercise under heat stress: thermoregulation, hydration, performance implications, and mitigation strategies. Physiol Rev. 2021;101(4):1873-1979. doi:10.1152/physrev.00038.2020

3. Cui CY, Schlessinger D. Eccrine sweat gland development and sweat secretion. Exp Dermatol. 2015;24(9):644-650. doi:10.1111/exd.12773

4. Baker LB, Wolfe AS. Physiological mechanisms determining eccrine sweat composition. Eur J Appl Physiol. 2020;120(4):719-752. doi:10.1007/s00421-020-04323-7

5. Norwell (Pharmascience Inc). Electrolyte Gastro — Oral Rehydration Solution [product label]. Available from: https://www.norwellcanada.ca/brand-products/electrolyte-gastro and https://www.electrolytegastro.com/products/discover-electrolyte-gastro [Directly fetched and confirmed, July 2026: sodium 12.1 mEq (288mg, per label’s own stated conversion), chloride 12.0 mEq, citrate 6.0 mEq, potassium 4.0 mEq, glucose 3.58g per sachet. Label gives separate reconstitution instructions by use case: 200mL water for diarrhea/vomiting, 300mL for heat exposure or prolonged exercise sweating. Label advises against combining with other sports or electrolyte drinks.]

6. Liquid I.V. Hydration Multiplier and Hydration Multiplier Sugar-Free — two separate sources, scoped to what each actually supports: (a) Product nutrition labels [not re-fetched in this session]. Sodium-source breakdown (salt/sodium chloride vs. sodium citrate, ~409mg and ~128mg respectively for the sugared version) confirmed against a Liquid I.V. Australia label disclosure via search, July 2026. The chloride mg figures used in this article are this author’s calculation from that stated NaCl content — not a number the company publishes directly. This specific label page was not re-fetched in the most recent research pass; reconfirm before publication. (b) The Science Behind Liquid I.V. Available from: https://www.liquid-iv.com/pages/science [directly fetched and confirmed, July 2026]. Describes the four-stage “LIV HYDRASCIENCE™” hydration framework and the marketing claim that the sugar-free amino acid-allulose blend is “backed by award-winning science” — this page does not itself provide nutrition-panel mg figures or cite specific studies for that claim; see references 18-20 for the actual underlying research and its peer-review status. Notably, chloride is not mentioned anywhere on this page despite sodium chloride being the primary sodium source in both formulas.

7. Ofei SY, Fuchs GJ 3rd. Principles and practice of oral rehydration. Curr Gastroenterol Rep. 2019;21(12):67. doi:10.1007/s11894-019-0734-1

8. LMNT. Ingredients. Available from: https://drinklmnt.com/pages/ingredients [Confirmed July 2026: sodium 1000mg, potassium 200mg, magnesium 60mg per serving. Chloride (1724mg) is not stated on the label but is exactly calculable, not merely inferred: the ingredient list shows salt (sodium chloride) as the sole sodium source and potassium chloride as the sole potassium source, with magnesium bound to malate rather than chloride — so every milligram of labeled sodium and potassium has one determined chloride pairing.]

9. Huma. Huma Plus, Berries & Pomegranate (”Double Electrolytes”) [product packaging]. Nutrition Facts panel and ingredient list, photographed directly from physical packaging by the author, July 2026. Confirmed: per 1 pouch (42g), calories 100, sodium 240mg (10% DV), total carbohydrate 24g, fiber 1g, total sugars 13g (10g added), protein 1g, potassium 1% DV, calcium 2% DV, iron 2%. Separately labeled electrolyte panel: sodium 240mg, potassium 65mg, calcium 25mg, magnesium 15mg. Ingredients: grape juice concentrate, brown rice syrup, strawberry puree, blueberry puree, chia seed powder, sea salt, pomegranate concentrate, lemon concentrate, blueberry concentrate. Sea salt is the only ingredient in the list capable of contributing meaningful sodium; the fruit purees and concentrates do not.

10. Tailwind Nutrition. Endurance Fuel [product packaging]. Nutrition Facts panel and ingredient list, photographed directly from physical packaging by the author, July 2026; independently re-verified against a second photo of the same label uploaded in this conversation. Confirmed: per 54g packet, calories 200, carbohydrate 50g (all as sugar), sodium 630mg (27% DV), potassium 4% DV, calcium 4% DV, magnesium 4% DV. Ingredients: sugars (dextrose, cane sugar), citric acid, sodium citrate, sea salt, potassium chloride, organic flavor, calcium carbonate, magnesium oxide. [Primary source, supersedes prior retailer-copy figures used in earlier research.]

11. XACT Nutrition. XACT Electrolytes Tabs [product packaging]. Supplemented Food Facts panel and ingredient list, photographed directly from physical packaging by the author, July 2026. Confirmed: per tablet (4.5g) dissolved in 500mL, sodium 500mg, potassium 100mg. Ingredients: sugars (glucose syrup, natural flavour), citric acid, sodium bicarbonate, trisodium citrate, salt, potassium chloride, polyethylene glycol, stevia extract. [Primary source — physical packaging supersedes the xactnutrition.com product page, which as of this writing displays a Nutritional Information panel but no ingredient list.]

12. US Food and Drug Administration. 21 CFR § 101.9 — Nutrition labeling of food. Code of Federal Regulations. Available from: https://www.ecfr.gov/current/title-21/chapter-I/subchapter-B/part-101/subpart-A/section-101.9 [Chloride is listed among voluntary, not mandatory, minerals under 21 CFR 101.9(c)(8)(iv); confirmed July 2026.]

13. Canadian Food Inspection Agency. Information within the Nutrition Facts table. Government of Canada. Available from: https://inspection.canada.ca/en/food-labels/labelling/industry/nutrition-labelling/nutrition-facts-table [Confirms the mandatory “core list” of nutrients — fat, saturated/trans fat, fibre, sugars, cholesterol, sodium, potassium, calcium, iron — does not include chloride; confirmed July 2026.]

14. Nuun. Nuun Sport [product packaging]. Supplemented Food Facts panel and ingredient list, photographed directly from physical packaging by the author, July 2026. Confirmed: per 1 tablet (5.5g), calories 15, carbohydrate 4g, sugars 1g, sodium 300mg (13% DV), chloride 40mg (2% DV, declared directly on label), calcium 13mg, magnesium 25mg, potassium 150mg (4% DV). Ingredients: citric acid, sugars (dextrose), sodium bicarbonate, potassium bicarbonate, sodium carbonate, natural flavours, potassium chloride, magnesium oxide, calcium carbonate, stevia leaf extract, avocado oil, beet red. [Primary source, supersedes prior website-sourced figures; sodium has no chloride-bearing source (bicarbonate and carbonate only), and potassium is split between potassium bicarbonate and potassium chloride with no disclosed ratio — but chloride content doesn’t require calculation here, since it’s stated directly on the label.]

15. GU Energy Labs. Roctane Energy Gel [product packaging]. Supplement Facts panel and ingredient list, photographed directly from physical packaging by the author, July 2026. Confirmed: per 1 packet (32g), calories 100, total carbohydrate 21g, total sugars 6g (6g added), calcium 30mg (2% DV), sodium 125mg (5% DV), Roctane Amino Acid Blend 1425mg (L-leucine, taurine, L-valine, beta-alanine, L-isoleucine), caffeine 35mg. No potassium stated. Ingredients: maltodextrin, water, fructose, L-leucine, taurine, sodium citrate, L-valine, L-isoleucine, citric acid, malic acid, calcium carbonate, green tea extract, beta-alanine, natural flavor, caffeine, gellan gum, medium chain triglycerides, sea salt, sodium benzoate (preservative), potassium sorbate (preservative). Sodium is split between sodium citrate and sea salt with no disclosed ratio; only the sea salt portion carries chloride. [Primary source, confirms prior research-based figures.]

16. GU Energy Labs. Hydration Drink Tabs, Tropical Citrus [product packaging]. Nutrition Facts panel and ingredient list, photographed directly from physical packaging by the author, July 2026. Confirmed: 12 servings per tube, per 1 tablet (4.5g), calories 10, sodium 320mg (14% DV), total carbohydrate 3g (1% DV), total sugars 1g (1g added, 2% DV), potassium 55mg (1% DV). Ingredients: citric acid, dextrose, sodium bicarbonate, natural flavor, sodium carbonate, potassium bicarbonate, stevia leaf extract, safflower oil, beet root powder (color added), cane sugar, riboflavin (as color). No chloride-bearing ingredient present — sodium comes entirely from bicarbonate and carbonate, potassium entirely from bicarbonate. [Primary source, confirms prior research-based figures.]

17. Cheuvront SN, Kenefick RW, Charkoudian N, Mitchell KM, Luippold AJ, Bradbury KE, Vidyasagar S. Efficacy of glucose or amino acid-based commercial beverages in meeting oral rehydration therapy goals after acute hypertonic and isotonic dehydration. JPEN J Parenter Enteral Nutr. 2018;42(7):1185-1193. doi:10.1002/jpen.1142 [Peer-reviewed, full paper. Tests a different amino acid product than Liquid I.V.’s (a proprietary blend of threonine, valine, serine, tyrosine, and tryptophan, marketed as enterade®), not Liquid I.V.’s L-glutamine/L-alanine formula. Funded in part by Entrinsic Health Solutions, Inc. (maker of the tested product) and the US Department of Defense; co-author S. Vidyasagar disclosed as Entrinsic’s Chief Scientific Officer — a disclosed conflict of interest.]

18. Bloomer RJ, Pence J, Hellenbrand J, Davis A, Davis S, Stockton M, Martin KR. Randomized trial to assess the safety and tolerability of daily intake of an allulose amino acid-based hydration beverage in men and women. Nutrients. 2024;16(11):1766. doi:10.3390/nu16111766 [Peer-reviewed, full paper, open access. A safety/tolerability trial, not an efficacy trial — confirms the L-alanine/L-glutamine formula was well-tolerated over 4 weeks, but does not test rehydration performance. Funded by Liquid I.V.]

19. Lelko M, Bloomer R, Pence J, Davis A, Allen E, Mohamed A, Chaluvadi S, Berk E, Ryan D, et al. Oral rehydration solutions improve measures of hydration over a 4hr period in healthy men and women [conference abstract]. Curr Dev Nutr. 2025;9. Available from: https://cdn.nutrition.org/article/S2475-2991(25)02116-X/fulltext [Conference abstract presented at American Society for Nutrition and American College of Sports Medicine annual meetings; not yet published as an independently peer-reviewed full manuscript as of this writing (July 2026). Funded by Liquid I.V., conducted at University of Memphis.]

20. Gopalakrishnan ST, Kooima PK, Kavouras IE, et al. Novel allulose amino acid-containing sugar-free beverage results in rapid ad-libitum rehydration following exercise-induced dehydration [conference abstract]. Med Sci Sports Exerc. 2025;57:789-790. [Conference abstract presented at the American College of Sports Medicine annual meeting; not yet published as a full peer-reviewed manuscript as of this writing (July 2026). Liquid I.V.–affiliated research.]

A note on sourcing standards: every product reference in this piece (5, 6a, 8, 9, 10, 11, 14, 15, 16) is now backed by a primary source — either a direct fetch of the manufacturer's own current page (5, 6b) or the author's own photographed physical packaging (8, 9, 10, 11, 14, 15, 16). Reference 6a (Liquid I.V.'s sodium-source breakdown) remains the one figure still resting on an earlier search result rather than a direct fetch or photograph — worth a final check if a physical Liquid I.V. label becomes available. The Liquid I.V., LMNT, XACT, Tailwind, and Huma Plus chloride figures in this piece are this author's calculations or estimates, not company-published numbers, and rest on different footing worth distinguishing. LMNT's figure is exact: it has exactly one chloride-paired source for each of sodium and potassium, so the chloride total is fully determined by stoichiometry. XACT's and Tailwind's figures are confirmed minimums: potassium chloride is the sole potassium source for each, so the potassium-derived chloride is calculable, with the sodium-derived portion left unquantified due to undisclosed multi-source splits. Huma Plus's figure is an estimate based on sea salt's typical 97-99% NaCl purity, not an exact calculation, since sea salt (unlike lab-grade "salt") carries trace non-chloride minerals. Liquid I.V.'s figure required the company's own disclosed mg split between its two competing sodium sources (salt vs. sodium citrate) before the same kind of calculation was possible. Nuun and both GU products required no calculation at all — their chloride content (40mg and 0mg respectively) is either declared directly on the label or absent because no chloride-bearing ingredient exists in the formula.

Kirk, one of the co-owners of Mountain Trek, said something during one of his lectures that has stayed with me.

After retirement, he said, the most important thing is freedom. Not political freedom. Not even financial freedom, although many of us spend decades trying to get that part right.

He meant physical freedom.

That landed harder than I expected.

Mountain Trek is a health reset retreat in British Columbia. For one week, guests live inside a carefully structured environment: daily hikes, strength work, yoga, recovery, prepared meals, lectures, body composition testing, massages, sauna, steam room, hot tub, cold plunge, and enough support that the usual frictions of daily life are temporarily removed.

Lydia and I went because we were ready for a reset. Not a crisis. Not a miracle. A reset. We wanted to understand what the next chapter of life could look like if we took movement, sleep, food, recovery, and habits seriously.

Because if you are fortunate enough to retire, you may have already done a version of the planning. The mortgage, the business, the pension, the savings, the insurance, the paperwork, the spreadsheets. You may have crossed the finish line you spent half your adult life running toward.

And then, one day, you realize money was only one kind of freedom.

There is another kind: the freedom to climb stairs, carry groceries, travel, hike, swim, sleep well, recover well, and say yes to the invitation because your body is still available to you.

That is the freedom Mountain Trek helped us see.

Not Just A Retirement Lesson

At first, I heard Kirk’s comment as a retirement lesson. But the more I thought about it, the more I realized it is a life lesson.

Imagine learning this in high school.

Imagine being taught that health is not just gym class. That food is not just calories. That sleep is not laziness. That movement is not punishment. That habits are not moral failures or heroic victories, but systems. That your environment can either pull you toward the person you want to become, or quietly pull you away from them.

Imagine learning early that freedom is not only something governments grant or money buys. It is also something your body allows.

Lydia and I are both at a stage of life where the next 20 years matter in a different way. Not theoretically. Not someday. Now. The decisions we make about movement, food, strength, sleep, stress, and recovery are no longer just about looking better or getting through the week. They are about whether the years ahead remain open.

Open to travel, friends, family, adventure, and still becoming.

An Engineered Environment For Change

Mountain Trek has spent more than 25 years refining this work. That matters. You can feel the difference between a program invented last season and one shaped by thousands of real people, real bodies, real struggles, and real attempts to change.

From the outside, it might look like a wellness retreat. From the inside, it feels more like an engineered environment for change.

The days are structured. The food is prepared. The hikes are planned. The lectures are science-informed, drawing from physiology, sleep, nutrition, movement, stress, recovery, mindfulness, and habit design. The distractions are reduced. The body is moved. The mind is invited to pay attention.

And slowly, almost without noticing, you begin to feel what becomes possible when a good system carries some of the load.

Lydia reminded me that the thank you needs to include the things that made the reset possible: the laundry, sometimes twice a day; the fresh towels; the cleaning staff; the massages; the hot tub, sauna, steam room, cold plunge, yoga studio, workout room, free weights, machines, prepared meals, packed schedules, and the quiet competence of people who knew exactly what they were doing.

Those things were not just luxuries. They were part of the design.

For one week, the daily worries of life were moved out of the way. No laundry pile. No grocery list. No meal planning. No wondering what to do next. No negotiating with yourself for an hour before taking a walk.

And in that cleared space, something became obvious.

But Real Life Still Has Laundry

Of course, someone could say, “That’s easy at Mountain Trek. They do your laundry. They cook your food. They clean your room. They schedule your day. Then you go home and real life is still waiting.”

Exactly.

That is not a criticism of the program. That is the lesson of the program.

Mountain Trek does not remove real life forever. It removes it temporarily so you can see the pattern clearly. Then the question becomes: what pieces of that system can we rebuild at home?

We may not have a chef waiting in our kitchen, but we can plan food on Sunday. We may not have hiking guides at the door, but we can schedule our walks, runs, swims, and strength sessions before the week begins. We may not have fresh towels appearing by magic after the workout, but we can pack the gym bag the night before.

We may not have someone taking away every excuse, but we can design our environment so the excuses have less room to breathe.

That may have been the biggest lesson for me. We often think change is about willpower: trying harder, deciding again, promising ourselves that Monday will be different.

But Mountain Trek showed us something else.

Change becomes possible when the environment is designed properly.

The food supports the movement. The movement supports the sleep. The sleep supports the mood. The lectures support the choices. The group supports the effort. The rhythm of the day supports the person you are trying to become.

And then, before you go home, comes the harder question:

What will this look like on Tuesday morning in your actual life?

Not as a wish. Not as a vague intention. As something on the calendar.

When will we walk? When will we run? When will we lift weights? When will we swim? When will we buy groceries? When will we prepare food? When will we turn screens off? When will we review the week?

Mountain Trek helped us see that if the change does not make it onto the calendar, it is probably still only a hope.

At home, the calendar becomes part of the environment. It is how the reset gets a fighting chance to survive contact with real life.

The Data

Then, after the week, the data arrives.

Our InBody results gave us a more detailed look at body composition than a bathroom scale ever could. Not just weight, but muscle, fat, water, balance, and the kind of information that helps turn vague intentions into something more specific.

The numbers did not feel like judgment. They felt like orientation.

Here is where you are. Here is what is strong. Here is what needs attention. Here is what is possible.

For Lydia, I think the week did something important. She was already walking regularly, already paying attention, already open to the next level. But Mountain Trek gave that desire a structure: running, strength training, swimming, recovery, sauna, cold plunge, better food, better planning.

Not as punishment. Not as another burden. As a path.

For me, it changed how I think about the rest of my life. Not as decline to be managed, but as design space still available.

That is a very different feeling.

The People

And then there were the people.

That part surprised me too. The friends we made became part of the medicine. Everyone arrived with their own story, their own reasons, their own private hopes. But for one week, we were moving in the same direction.

We walked the same trails, ate at the same table, learned the same lessons, and laughed at the same human absurdities.

There is something powerful about being around people who are also trying. Not pretending. Not performing. Just trying.

Atomic Habits, Made Physical

After the retreat, Lydia and I ended up in Nelson. We stopped at Packrat Annie’s used bookstore, and I picked up Atomic Habits by James Clear. Lydia already had it on her reading list.

Atomic Habits is James Clear’s book about how small behaviours become lasting systems, especially when the environment is designed to make the right action easier.

A few chapters in, something clicked.

Mountain Trek was Atomic Habits made physical.

The book explains the principles. Mountain Trek builds the room where the principles work.

Reduce friction. Create cues. Build identity. Stack habits. Make good choices easier. Surround yourself with people moving in the same direction. Design the environment so the better version of yourself has a fighting chance to show up.

That was the realization.

Mountain Trek was not simply telling us what to do. It was letting us experience what a well-designed life could feel like.

So we came home with a sentence:

That is our keystone habit. Not flashy. Not heroic. But maybe life-changing.

Because retirement does not need only a financial plan. And life, at any age, does not need only good intentions.

It needs a physical plan. A sleep plan. A food plan. A movement plan. A friendship plan. A purpose plan. A habit plan.

It needs a design.

A Thank You

So this is a thank you.

Thank you to Mountain Trek for spending three decades refining a place where people can reset.

Thank you for taking the science seriously.

Thank you for caring about people younger than us, older than us, and exactly where we are.

Thank you to the guides, the cooks, the massage therapists, the cleaning staff, the people doing the laundry, the people folding the towels, the people maintaining the grounds, the hot tub, the sauna, the steam room, and the cold plunge, and the people who made the whole thing feel effortless when I know it was anything but.

Thank you to the other guests, who trusted us with pieces of their stories and helped remind us that everyone arrives carrying something, and everyone is trying to find their way forward.

Thank you for moving the daily worries out of the way long enough for us to see what was possible.

Kirk was right.

The real freedom after retirement is not just having enough money.

It is being able to walk to the refrigerator without it being an event.

And maybe the real lesson is this:

We should not wait until retirement to learn how freedom is built.

Regards,

Luke Kandia

Luke Kandia is a writer and researcher based in Winnipeg, Manitoba, a retired IT services owner, and a recreational runner still trying to understand what it means to build a good life on purpose.

This article is a personal account of the week Lydia and I spent at Mountain Trek, and of the thoughts it sparked about retirement, health, habits, freedom, and the years ahead. It is not medical advice, financial advice, or a universal prescription for how anyone else should live.

The reflections here are based on our own experience, our own data, and our own attempt to bring some of what we learned home with us. Anyone considering major changes to exercise, nutrition, recovery, or health routines should make those decisions in conversation with qualified professionals and with respect for their own body, history, and circumstances.

If you have had a similar reset experience, or if you are trying to design the next chapter of your life before life designs it for you, I’d love to hear your thoughts in the comments.