Me playing poker with one of my older brothers. I’m on the left.

On the way to Phish shows at Madison Square Garden, one of my best friends and I used to make markets on the setlist.

We didn’t call them markets, obviously.

Phish has hundreds of songs they might play on any given night, which gives you plenty of room to overthink things. We’d each pick five songs. If your song got played, you won $10. Pick the opener and maybe that was worth another $10. We’d add wrinkles to the rules depending on whatever we came up with on the way to the show.

The money was almost beside the point. Nobody was getting rich.

But putting $10 on it changed the question.

Instead of saying, I bet they play Tweezer tonight, you had to decide whether you actually believed it enough to spend one of your five picks on it. Was it overdue? Had they played it recently? Did the venue matter? Was there some theme developing over the run?

Suddenly being right had a price.

Looking back, I’ve apparently been trying to turn the world into prediction markets for most of my life.

Games within games

A lot of this started with games.

I loved strategy games growing up. Chess. Poker. Warcraft III. Especially the custom games people built inside Warcraft, where someone would take the basic engine and invent an entirely different game on top of it.

I could spend an absurd amount of time figuring out how to beat a tower defense map.

Not just beating it once. Figuring out why something worked. Which upgrade was actually worth buying. Whether investing early made the late game easier. Whether some strategy everyone assumed was optimal actually wasn’t.

There was always another layer underneath the game.

You could play it.

Or you could try to understand it.

That distinction has probably shaped more of my life than I realized at the time.

Then there was poker

Poker added money to the equation.

I came up during the online poker boom, when it felt like everyone was playing. I played online, played with friends, played home games around whatever table we could find. By college it was local casinos.

Poker scratched a slightly different itch. Now you weren’t just solving a game. You were making decisions without knowing all the information.

Sometimes the correct decision lost.

Sometimes the terrible decision won.

That sounds obvious, but I think learning it viscerally matters.

You can make the right call and lose the hand. You can make the wrong call and get rewarded. The outcome isn’t the same thing as the quality of the decision.

Once you internalize that, it’s hard not to start seeing it everywhere.

The world isn’t yes or no

I sometimes wonder whether I see too much of the world probabilistically.

Most things we’re taught growing up are presented as facts.

This is how things work. This person will win. This company is good. This policy will fail. This team is better. This is going to happen.

Reality usually looks more like 60/40.

Or 85/15.

Or 51/49, with everyone yelling at each other like it’s 100/0.

Thinking probabilistically doesn’t mean having no convictions. It means attaching some humility to them.

There is a chance I’m right.

There is also a chance I’m wrong.

And if you make me put a number on those two possibilities, things get interesting very quickly.

That’s part of what I love about prediction markets. They force an opinion through one additional filter: at what price?

Saying something will happen is easy.

Buying it at 80 cents is different.

Finding prediction markets

In hindsight, prediction markets feel like a fairly natural place for me to have ended up.

They combine pieces of almost everything I’ve always found interesting: games, markets, data, poker, incentives, strategy, psychology, current events, and the satisfaction of getting to the bottom of some weirdly specific question.

And then they add an order book.

That’s the part the setlist game and the poker table never had. Every opinion has a price posted by someone who disagrees with you, and you can see it, trade against it, or update because of it.

I’ve spent a lot of time criticizing parts of this industry, including some of the ways prediction markets are marketed. I’ll keep doing that. If a market’s rules are bad, its resolution source is fragile, traders are treated unfairly or someone is making claims the data doesn’t support, I want to call balls and strikes.

That criticism doesn’t come from disliking prediction markets.

It’s probably the opposite.

I find this stuff endlessly fascinating.

I love finding some obscure market and realizing there is an actual answer buried somewhere. I love watching a price move and trying to understand whether someone knows something, thinks they know something, or is simply willing to pay more than everyone else. I love the strange little economies that emerge around liquidity incentives and market structure.

There are very few places where being curious about the world and being obsessive about tiny details are rewarded quite so directly.

For the love of prediction

I’m not entirely sure where all of this goes.

I hope prediction markets keep growing. I hope the markets themselves get better. I hope we get more liquidity, better data, better rules and more serious questions worth trading.

And personally, I hope I get to spend a long time in this world.

Writing about it. Trading it. Building things around it. Meeting other people whose brains seem to have been accidentally wired the same way.

Resolution Pending didn’t start because markets are always right, or because every question needs money attached. Sometimes it’s just fun to care whether Phish plays Tweezer.

But I’ve always loved the process: take something uncertain, figure out what you know, figure out what you don’t, and make your best guess anyway.

Sometimes putting $10 on it just makes you think a little harder.

Thanks for reading.

Peace, Aaron

Prices below are in cents and use one-minute volume-weighted Yes fills. County labels are the best-supported public explanations, not proven causes.

At 5:53 p.m. ET, Kalshi Politics advertised the race at Hong 95%, Crowley 5%. At 9 p.m., actual Yes trades were near 96 cents and 5 cents. Twenty-six minutes later, Crowley was the favorite.

Then the market spent 80 minutes changing its mind.

From 9:26 to 10:45 p.m., the lead flipped eight times. 3,046,177 contracts traded in that window, 18.6% of the market’s eventual 16.39 million-contract volume. Kalshi Politics counted the same eight changes independently.

The eight flips track a run of vote batches from counties with very different electorates.

The eight flips, in order

1. 9:26 p.m. Crowley leads, 50.1 to 46.5. The first meaningful Dane County batch came in weak for Hong, and Crowley briefly led her home county.

2. 9:35 p.m. Hong leads, 51.9 to 41.8. Hong-friendly numbers appeared in Outagamie and Kenosha, followed by votes from whiter parts of Milwaukee County. This is the best public explanation for the snapback, but no archived official feed ties one exact batch to the trade.

3. 9:38 p.m. Crowley leads, 54.6 to 44.8. I found no discrete catalyst. Crowley was still overperforming in Dane and building a Waukesha-area lead. This looks like traders unwinding the overreaction to the Hong batch three minutes earlier.

4. 9:42 p.m. Hong leads, 51.8 to 47.6. A near-complete Marathon County return favored Hong around 9:41 ET. Kalshi flipped one minute later.

5. 9:46 p.m. Crowley leads, 52.9 to 46.4. No single batch again. The discussion had settled on two structural problems for Hong: she was not winning Dane by enough, and Crowley was carrying the Waukesha collar. By 9:58, Kalshi Politics had Crowley at 85%, an 80-point reversal from its afternoon snapshot.

6. 10:18 p.m. Hong leads, 51.2 to 49.9. This was gradual. At 10:20, the Associated Press had the two “neck and neck.” AP update

7. 10:32 p.m. Crowley leads, 58.6 to 43.1. Crowley retook the market lead as a late Racine County return was reported to favor him. The timing matches the move, but no archived official feed proves that the batch caused it.

8. 10:45 p.m. Hong leads, 60.5 to 38.6. Hong retook the market lead as favorable returns from Wood, Portage and Brown County arrived. Again, the timing is suggestive, not proof of causation.

Volume peaked when the map was least clear

The largest five-minute burst was 337,386 contracts from 10:20 to 10:25, just after Hong’s sixth flip. A 10:23 Kalshi screenshot had Hong at 64% with 66.3% of the estimated vote counted. By 10:55, the count had reached 77.7%. In those 32 minutes, the market went to Crowley and came back to Hong.

That’s the tell. The remaining vote kept changing shape, and traders kept repricing what was left of the map.

One actual news event came later. Roughly 28,000 Milwaukee absentee ballots were delayed after an upload error, keeping the race unresolved past midnight. It happened well after these eight flips, so it cannot explain them. Associated Press

The seesaw outlasted the chart. After Hong’s eighth flip, Crowley regained the lead after midnight; Kalshi Politics posted him at 88% at 12:09 a.m.

What the oscillation showed

Josh Marshall read the same reversals as evidence that traders had lost their polling signal. “The oscillating back and forth was, again, the tell,” he wrote. His test was sensible: ignore the statewide topline and ask whether candidates were hitting the margins they needed in counties that were mostly counted.

The Kalshi tape cannot prove what traders saw. It records fills, not motives, and this analysis does not cover Polymarket. But the eight flips coincided with changing county returns, the same information Marshall says mattered. That makes his interpretation possible, not conclusive: the same oscillation can reflect traders chasing the topline, traders rapidly repricing the county map, or both.

What the tape establishes is narrower. During 80 minutes of extreme uncertainty, the favorite changed eight times and 3.05 million contracts traded. The market was unstable and sometimes wrong, but it was not static or information-free.

Kalshi co-founder Tarek Mansour made the separate calibration point that a 5% outcome should occur about once in 20 cases. One upset does not prove bad calibration, but eight flips and 3.05 million contracts show that 95–5 was only a starting estimate, not a stable description of the live race. Source

That is useful even when the price is not a precise forecast. A market is not a poll, but in moments of high uncertainty it can become a live record of trader sentiment: how participants absorb new information, weigh competing signals and revise their view.

Crowley won, 315,308 votes to 311,512. Margin: 3,796. The market found the upset early, then spent 80 minutes arguing with itself about how much each corner of Wisconsin was worth.

Method: This reconstruction uses every public API trade across all 11 KXGOVWINOMD-26 nominee contracts from 9:00 to 10:55 p.m. ET. Prices are one-minute volume-weighted Yes fills, carried forward when no trade occurred. A flip is a sign change in the Crowley-minus-Hong price spread. Separate contracts trade at different moments, so their minute prices need not sum to 100. County attributions rely on timestamped public reactions because no historical official batch-by-batch feed was available.

It would be easier to make this forecast a month from now. By then, preseason and Week 1 will have given us more evidence and narrowed the range of plausible outcomes. I am taking the shot now, when uncertainty is high.

My estimate: 21.9B Kalshi NFL game-winner contracts through the divisional round. Scope: Kalshi game-winner markets only, both team sides summed, Hall of Fame Game through the divisional round. Resolves when the last divisional market settles. I’ll report the sum of Kalshi’s two separately traded team markets. That is a measurement choice, not cash handle; using only one team market would produce a lower, but not necessarily exactly half, total.

Last season’s 272 regular-season games averaged 17.1M contracts. The ten in-scope playoff games generated 3.36 times as many contracts per game as the regular season. Assume every 2026 regular-season and in-scope playoff game trades at last year’s playoff pace and you get 16.2B, or 3.09x last season. I’ll use that as an aggressive starting benchmark because it comes from the same market series and ten completed games. It assumes a Week 3 game between two 0-2 teams draws playoff attention. The assumption is aggressive by design because last season’s trading levels may already be outdated in a rapidly expanding prediction-market ecosystem.

This is a convergence model, not a uniform-growth model. Last season’s 57.6M playoff average serves as a per-game modeling ceiling: regular-season games rise toward that rate, while the ten playoff games remain flat. If playoff trading grows too, the 16.2B benchmark understates the season. I prefer that asymmetry to automatically multiplying the highest-volume games by 3.36x again.

That 3.09x benchmark plus a 10% judgmental allowance produces our rounded 3.4x starting point. The opening game receives 10% weight and moves the selected forecast to 4.17x, or 21.9B contracts.

Base case: 21.9B contracts. The eight-in-ten model range is 10.2B to 47.3B.

What is excluded: spreads, game totals, player props, quarter and half markets, multi-event contracts, team and season futures, awards, the two conference championship games, the Super Bowl, and all college football. Those products have different launch dates and histories. Simply adding their volumes would create a false all-NFL estimate.

Historical data: complete within the defined scope

Kalshi’s 2025-26 archive holds 5.25B contracts across 331 games with volume through the divisional round. The schedule reconciliation matched every game inside that boundary: 49 preseason, 272 regular season, six wild-card, and four divisional. Two zero-volume records were excluded. The conference championships and Super Bowl are outside the model rather than missing from the audit.

Preseason is noise. Forty-nine games produced 0.24% of the final volume. Regular-season volume tracked games played. Weeks 1-4 were 23.53% of games and 23.47% of volume. The ten wild-card and divisional games averaged 57.6M contracts, 3.36 times the regular-season average. The Hall of Fame Game alone was 0.009% of the season.

Current data: strong evidence, but only one game

At August 10, 2026, 33 listed events had booked 13.9M contracts. One closed game accounts for 13.0M of that. The other 32 events have traded 945,783 combined. That game traded 491,855 contracts last year, so the year-over-year signal is roughly 26x. This is a one-observation dataset wearing a 33-event costume.

The comparison is like-for-like: both years sum the two team markets for one finalized Hall of Fame Game.

The structural break cuts both ways. It likely inflates the opener’s 26x growth rate, while making last season’s absolute volume too low for today’s larger platform. The separate 10% platform-growth allowance recognizes the latter; the opener still receives only 10% weight because one game cannot identify the season-wide effect.

If the opening game represented the same tiny share of the season as last year, the answer would be 138B contracts, which I’ll reject. That hands one game the authority of a 331-game season.

How the forecast works

1. Calculate the benchmark. Last season’s 272 regular-season games averaged 17.1M contracts. The ten wild-card and divisional games averaged 57.6M. Applying that playoff pace to the 282 regular-season and in-scope playoff games produces 16.2B contracts, or 3.09x last season.

2. Set the starting point. Add a 10% platform-growth allowance and round to 3.4x, or about 17.9B contracts.

3. Use the new evidence carefully. The Hall of Fame Game grew about 26x, but it represented only 0.009% of last season’s final volume and spans a structural break.

4. Blend multiplicatively. Give 3.4x 90% weight and 26x 10% weight. The result is 4.17x, or 21.9B contracts. A straight weighted average would give 5.7x.

The 10.2B to 47.3B range contains eight in ten model scenarios. In practical terms, the low case is about half the base and the high case is about twice the base. It is not a confidence interval. I chose the width judgmentally because one complete season and one current game do not justify tighter precision.

Two risks sit outside that range. Contract count is sensitive to price mix: cheaper longshot contracts create more units per dollar, so contract growth does not equal cash growth. A delisting or other interruption that prevents future markets from trading could push realized volume below the modeled range.

The forecast changes materially with the starting growth assumption and the weight given to one opening game. Our selected case starts at 3.4x and gives the opener 10% weight, producing 21.9B contracts.

Kalshi, competitors, and sportsbooks

Kalshi is the forecast anchor because it is the only venue in this review with a complete public prior-season game archive.

ProphetX belongs on the watchlist. It received federal exchange and clearing approvals in June 2026 and filed NFL products in July.

Polymarket U.S. belongs in the comparison set. Its public archive matched all 164 scheduled games in the available October 30, 2025-February 8, 2026 window. That partial window contains 82.2M contracts.

Novig is a launch comparator, not the center of the forecast. Its consumer app routes to Ludlow Exchange, the federally designated exchange owned by Novig Betting, Inc. The opening snapshot showed $73.5M of venue-reported volume over 5.65 days. Dollars, not contracts, so it does not compare to anything else here.

Crypto.com and OG matter because the contracts shown in those apps trade through Crypto.com Derivatives North America, the affiliated federally regulated exchange and clearinghouse. DraftKings Predictions and FanDuel Predicts can also distribute contracts from that underlying venue.

Robinhood is also a distributor, not a separate pool of exchange volume. Count the app and the exchange and you count the same trade twice.

Sportsbooks are the incumbent wallet benchmark, not a denominator. The American Gaming Association estimated $30B of legal U.S. NFL sportsbook handle for 2025, but sportsbook handle measures cash staked, while Kalshi’s figure counts contracts. The two numbers land close. Dividing one by the other would still be meaningless.

The short version

Forecasting 21.9B Kalshi NFL game-winner contracts through the divisional round. The starting point is 3.4x last season, anchored to last year’s playoff trading rate across the 282 regular-season and in-scope playoff games. The opening game grew about 26x, but it was only 0.009% of last season and spans a structural break, so it gets 10% weight. Spreads, totals, props, futures, the final three playoff games, and college football are separate.

Preseason is the first checkpoint. Last year’s 49 games traded 12.4M contracts. When the cohort closes, it will replace, not supplement, the opener signal because the opener is inside preseason. I will blend preseason growth at 25% weight with the 3.4x starting point at 75%. A 30M preseason (2.42x last year) produces 3.12x, or 16.4B; 100M (8.06x) produces 4.22x, or 22.1B. I will update again after Week 1.

Valuation date: August 10, 2026. Public API snapshots, transformations, schedule audits, model tables, and publication links are preserved in the accompanying research package. No venue reviewed or endorsed this analysis.

I’ve been all in on this space for a year and a half: building tools on the APIs, trading, making markets, writing about it. So these are my current thoughts, held the way any opinion should be held here, ready to be repriced on new information. It would be strange if my views were the one market that never moved. I’m long the sector in basically every sense, which is exactly why the embarrassing parts bother me.

The marketing makes everyone look dumber

Official accounts serve BREAKING and JUST IN engagement bait alongside markets like whether Jesus will return before 2027. That contract has attracted about $65 million in volume and will resolve on a “consensus of credible sources.”

The incentives can become just as strange as the subject matter. Traders even tried to pump a separate derivative contract tied to the market’s odds, turning the displayed probability into something participants could try to influence.

The same attention-first instinct extends to promotion. Kalshi and Polymarket launched remarkably similar free-grocery campaigns only days apart.

Same instinct, not “proof” of copying. Sources: Kalshi on X and Polymarket on X.

Source: Polymarket market page.

In August 2025, about $500K traded on whether a sex toy would hit the WNBA court, down to its color. The throws promoted a memecoin and two people were arrested. Once that market exists it’s a bounty with extra steps: anyone could buy Yes, then make Yes happen. A market that rewards changing the outcome it measures has stopped measuring. (DL News)

Source: Polymarket Sports on X.

The Wall Street Journal found creators posting fake winning bets, and two senators asked the CFTC to examine the marketing itself (Forbes). You can’t argue you’re the heir to CME in federal court in the morning and post like a memecoin account in the afternoon.

The product is quietly excellent

Both exchanges expose market data without an API key or subscription; every pull in this newsletter comes with a zero-dollar bill.

The research side is what I’m genuinely interested in. Kalshi Research turns rate contracts into Fed forward curves and now publishes compute curves for hourly B200, H200 and A100 rental prices a year out (Kalshi’s compute-forward-curve analysis).

Polymarket has an institutional research arm (Polymarket Institutional). A February Fed staff paper found Kalshi’s rate forecasts about as accurate as professional forecasters and called them a high-frequency, distributionally rich benchmark (Federal Reserve paper).

You don’t need to trade to use the data. Prices reach headlines and dashboards, or flow through APIs. CNBC can now write that Coinbase may miss “according to prediction markets,” making the price the story.

The hedging almost nobody does

A Manhattan bar owner put $5,000 into Knicks contracts at a 37% implied probability to fund free tabs if the team won Game 1 of the Finals. It won. Kalshi pitched the idea and its CEO posted it, making the hedge a marketing exercise (Fortune).

Compute curves could let AI companies lock in GPU costs as airlines do fuel. CPI contracts already trade, and I’d like a medical CPI subindex so small providers could hedge next year’s healthcare costs. Parametric cover could extend to flight delays, rainfall, freight and electricity. Sports still gets the volume and press.

Galaxy’s OTC desk facilitated a $10M CLARITY Act position for crypto fund Arca, which bought No as a hedge against the bill failing (Galaxy announcement; Galaxy explanation).

Resolution disputes are the recurring bill

The more specific the contract, the cheaper it is to move the underlying. War markets make the same point: overly specific resolution criteria don’t add information, they add attack surface.

In November 2025, an unauthorized edit to ISW’s live map showed a Russian advance that hadn’t happened just before Polymarket’s Myrnohrad deadline. Longshots got paid, the shading vanished by morning and the staffer was reportedly fired (404 Media). The market held $1.3M, and similar ones reached $5M. The contract priced exactly one thing, which is who could alter a map at the right time.

A $376M Polymarket contract asked whether Strategy sold bitcoin by May 31. It sold 32 BTC from May 26 to 31 and disclosed it June 1, but the contract resolved No. Polymarket’s “additional context” made the confirmation date controlling, sending Yes from 10 cents to 80 to under a penny (Galaxy Research). Buyers called the future and got zero.

A WSJ analysis linked at least 60% of active UMA voters to Polymarket accounts. One in five disputes involved a voter with a position, and the ten largest wallets held a majority in most disputes. More than 1,150 markets were disputed in 2026, past the full 2025 total. Voters traded markets they helped resolve (The Wall Street Journal).

Traders bet on outcomes and on who declares them. Polymarket’s token vote produced Strategy, turning resolution itself into a second market. Criteria should be locked at listing, with each market labeled event-based or confirmation-based. Voters should be recused from disputes where they hold positions.

The sportsbook comparison nobody wants to have

A Kalshi taker order can feel like a sportsbook bet. The matching engine and bid-ask spread sit behind a similarly simple slip, even though the legal and market structures are different.

On this narrow comparison, Kalshi looks cheaper. New York mobile sportsbooks held about $91 of every $1,000 wagered in the reported months so far this year. At 50 cents, Kalshi’s standard taker formula charges $35 per $1,000 put into contracts. Most resting maker orders pay no fee. I’ve called those fees high. Against a sportsbook, they aren’t.

That is only the flattering benchmark. Compared with other financial products, prediction-market monetization is still dramatically higher.

Sheel Mohnot estimates that Robinhood generates about $23 per $1,000 traded in prediction markets, versus $2.50 in crypto, $1.47 in options and $0.13 in stocks. That is roughly 9 times crypto, 16 times options and 170 times stocks.

I’d like to see prediction-market fees come down over time as volume and liquidity continue to grow.

Source: Sheel Mohnot on X.

Then comes the tax gap. New York taxes sportsbook gross gaming revenue at 51%; Kalshi pays no comparable state gaming tax because it operates under federal commodities regulation. The comparison is imperfect: sportsbook hold is realized gaming revenue, Kalshi’s $35 is an explicit transaction fee, and an exchange has no direct equivalent of handle minus payouts. Before arguing about the rate, someone has to define the tax base.

Sources: New York State Gaming Commission revenue reports and Kalshi fee schedule.

Gamblification, from someone it’s been good to

I’ve spent years in the gambling-adjacent world. It’s treated me well, and it’s genuinely destructive for a real slice of society. Both are true, and pretending otherwise is marketing.

New York’s AG calls these gambling platforms, and the sports product makes that hard to argue at the user level. Everything is drifting toward a bet, from sports and elections to the weather and your flight. I don’t think it reverses.

Who gets paid, and for what risk

A sportsbook books against its own balance sheet. Kalshi matches traders and charges a fee. Market makers, potentially including its affiliate Kalshi Trading, carry spread inventory and earn the spread. Kalshi carries the legal risk around the platform itself (member agreement).

That risk is no longer theoretical. New York sued on July 31, seeking a reported potential $36 billion from a company valued at $22 billion in its latest completed round, hours after the Second Circuit denied Kalshi’s final emergency motion (New York AG; NBC News). Related state and federal cases are still testing whether federal commodities law preempts state gambling rules (CFTC).

That does not make the fees cheap, but it helps explain aggressive monetization while the legal foundation for sports event contracts remains unsettled.

Where I land

I’m still market making, trading and building on the APIs. I’m extremely bullish on this sector, but I’d like more of the focus to move toward research: probabilistic odds that are actually valuable to society, and event contracts tied to real hedging needs. Venues asking courts for institutional treatment need institutional judgment in public.

Cleveland spent most of this past spring as the favorite, trading between 40% and 55%. Golden State and Miami each had their turn. Philadelphia sat in the low single digits from January through late June.

I came in late. The last few weeks were the only stretch where I was heavily monitoring and quoting it.

Philadelphia won.

What I Traded

Roughly 688k contracts across the LeBron event, with about 94% of that volume executed as a maker.

For scale: Kalshi’s final numbers on the event show over $230M in volume against $143M in open interest. Even if every contract I traded had printed at a dollar, I would still be under 0.3% of that dollar volume. I was not moving this market. I was standing inside it, collecting spread.

Most of the execution was manual. I posted bids and offers, watched the narrative rotate between Cleveland, Miami, Golden State and Philadelphia, and adjusted by hand.

I have also been building tooling on the side: fair value estimates, cross-venue price comparison, and helpers that adjust orders as prices move. Some of that informed this trade. The final decisions were still mine, which is either a feature of the process or the reason for the final number.

I was recovering from a mountain bike injury, which meant I could actually sit in front of the book and be terminally online. An announcement could land at any moment, and I did not want resting orders sitting there while I was away from the screen. No monitoring, no quotes.

The $4k Gap

My dashboard estimated approximately $6k in paired capture, matching favorable buys against favorable sells. That is the spread I collected on contracts that fully round tripped.

The book, marked immediately before the announcement, was worth about $2k.

Roughly $4k separates those two numbers, and all of it is inventory. Paired capture only counts trades that closed. It says nothing about the contracts still sitting on my book, and those were marked below what I paid. The spread I captured was real. Part of it was handed back on the positions I never traded out of.

By the time the announcement arrived, I was holding roughly 285k NO contracts across 10 teams, with about $233k committed to the open book.

The Miami Exception

Miami was the one outcome where I allowed myself to lean directionally.

A leaked YouTube video convinced me Miami was underpriced on Kalshi. Polymarket had also been marking Miami higher than Kalshi for most of the preceding week. When there was enough liquidity, I was willing to treat the Polymarket midpoint as an outside reference and pull my inventory toward it.

In practice, that meant taking Miami YES exposure, or at least carrying less Miami NO than I would have otherwise.

Everywhere else, I was recycling inventory and capturing price differences.

If Miami had won, I’d print >$13k instead of the -$7k every other team I had size in was showing.

This is where market making and having an opinion stopped being separate jobs. I was still quoting, but my opinion determined which inventory I was comfortable holding.

The Book

Winning outcome / Projected P&L

• Golden State: -$7k

• Minnesota: -$7k

• Philadelphia: -$7k

• Cleveland: -$7k

• Miami: +$13k

• Any unheld team: ~+$52k

I was short ten teams, but not evenly. Almost all of the size sat in five names. The other five were small enough that they barely moved the book in either direction. If the winner had been a team I held nothing in, all 285k NO contracts pay $1 against $233k of cost and the book clears $52k. Every team I did carry NO exposure in cut into that number. Four of them took it negative.

Minnesota was never a serious candidate and still sat in the losing column.

The book was positive EV because the market gave more probability to the branches that paid than to those that didn’t.

It was not an arbitrage.

That was less obvious while I was focused on individual fills and asking whether each quote looked favorable in isolation. Every fill was defensible. The sum of the fills was a bet. The loss branches were a map of where my inventory was, not of the news.

Settlement

That $2k was never $2k of profit. It was the probability-weighted average of a distribution based on the current orderbook pricing.

Philadelphia was one of the four losing branches.

$226k of winning NO contracts paid out against $233k in open inventory cost, an average of about 82 cents per NO contract. Nine NO positions paid $1. Roughly 59k of the 285k contracts I held were Philadelphia NO, and those paid zero.

The book settled at roughly -$7k.

The Lesson

Watching the screen protects your quotes. It does not change what your inventory pays at settlement.

I could have flattened part of it before the announcement. I chose not to. I was willing to carry the exposure through settlement.

Plenty went right: tight monitoring, no stale orders into an announcement, defensible fair values, 94% maker volume, and a book with positive expected value. The result was still red, and none of that changed it.

Good process and a bad outcome can show up in the same trade. Expected value describes the average of a spread of outcomes. Settlement pays one of them.

Where This Goes Next

I used to be very particular about which markets I would quote. I wanted to fully understand the subject before putting up orders.

Increasingly, I think market selection is really table selection.

Where is retail flow going? How are those traders behaving? Is there enough volume to justify the learning curve? Can I become competent at pricing it before the edge disappears?

If the answers are good, I will make a market in something I have never traded before. That is the plan: more markets, sometimes manual, sometimes with the tooling.

LeBron was exactly the kind of market I want to be quoting. $230M of flow, enough retail on both sides to fill 688k contracts, and 94% of my volume as a maker. By every question I just listed, this one qualified.

I still lost money in it. Picking the right market tells you the edge is there. It says nothing about what you accumulate while collecting it.

The difficult part is not finding the table.

It is knowing whether you are collecting the spread or getting picked off.

I trade and make markets on Kalshi, and I work on the data infrastructure at Kalshinomics. Nothing here is investment advice.

Before President Trump began his State of the Union address in February (oh, and every time he’s spoken since), traders on Kalshi bought and sold contracts on what he’d say. One of the people who allegedly knew the answers was running his teleprompter.

If the allegations are true and the person across from you already had Trump’s prepared remarks, do you get your money back?

ABC News reported on July 16 that Gabriel Perez, Trump’s longtime teleprompter operator, is in settlement talks with federal regulators over allegations that he used advance knowledge of prepared remarks to win more than $100,000 trading on the exchange. The trades reportedly covered the State of the Union and more than a dozen other speeches. The company says it froze the account and referred the case to the CFTC. The White House later said Perez would no longer work there. The allegations haven’t been adjudicated, and the CFTC hasn’t confirmed or denied an investigation.

That alleged $100,000 didn’t come from the exchange. It came from whoever was on the other side.

The exchange’s published rulebook says sanctions may include full restitution when participant harm can be identified and reasonably calculated, plus disgorgement and damages to unoffending counterparties. I couldn’t find a public answer on whether it was used here. Were trades canceled or counterparties made whole? Reuters reported the freeze kept more than $90,000 from being withdrawn. That money is sitting somewhere. Does it go back to the traders who lost it? Fines go to the government. The losses went to traders.

Traders own some of the risk

I trade and make markets on Kalshi, so I expect some people across from me to know more. Someone models every Trump speech and buys before the quote moves? Sharp trade. Same for whoever catches a Spotify trend first or builds a better election model. Having the final script isn’t research.

Traditional enforcement can take years. The SEC charged Mark Loman in July 2019 over alleged options trades from 2015, and Matthew Panuwat in August 2021 over alleged options trades from August 2016. Flagging this in months is warp speed by comparison.

If you trade a mention market where a crowd of staffers may have the answer, some of that risk is yours. The exchange isn’t responsible for every bad trade, but it did invent this particular way for you to lose money. It chose the contract, source and payout timing, and it sees flags you can’t. It can pause a compromised market, warn users or delay settlement. You can’t price information you don’t have.

Two ways these markets break

Perez allegedly had the answer early. Spotify can change the answer after payout.

In December, I bought NO on The Weeknd’s “Party Monster” reaching one billion Spotify streams before the end of 2025. I paid about $4,500 to return roughly $6,100; by the time I wrote, it was worth about $2,100. (I wasn’t neutral then and I’m not now.)

Daily streams were stable for most of 2025. The inset shows roughly 47,000 YES contracts at 2¢ and 18,000 at 4¢ on Dec. 16.

Then they jumped from roughly 290,000 to more than 1.2 million. Suspicious, not proof: Spotify never confirmed they were artificial, and nobody has tied the move to a trader.

Caleb Davies, a music-market trader who posts as Gaeten Dugas (@GaetenD), found a cleaner example. Malcolm Todd’s “Earrings” went from a one-to-five-cent long shot to No. 1 on Spotify’s U.S. chart, an 11.24-sigma move by his calculation. Spotify confirmed artificial streaming and removed more than 500,000 plays. Davies says he warned the exchange before the results posted and asked it to delay settlement.

Spotify didn’t identify who created the streams or establish a market motive. It did change the final numbers to remove botted streams, dropping the song to fourth after the market had already settled and paid out.

THE MANIPULATION RATIO
potential profit from changing the outcome ÷ estimated cost of changing the resolution source

If the payout beats the cost of moving the chart, someone will try it. The exchange didn’t create Spotify’s bot problem. It did make Spotify’s correction speed a problem for traders.

Should Spotify really have to deal with potentially multimillion-dollar contracts riding on its data? That’s another article.

Kalshi needs better insider screening for mention markets and slower settlement when the source can revise its data.

Flag it before it settles

Federal regulation is central to the company’s trust pitch. On March 30, the exchange ran Washington ads declaring, “RULE #1 KALSHI BANS INSIDER TRADING” and posted: “We ban insider trading. And we enforce it.” But a bank vault doesn’t become secure because the door says ROBBERY PROHIBITED.

In the first quarter, the company says it completed more than 150 investigations, blocked more than 100 potential insider trades, made more than 20 law-enforcement referrals and imposed five disciplinary actions. Good, but those numbers aren’t reassuring. More than 150 investigations in one quarter? This shit must be happening all the time. That doesn’t mean there were 150 insiders, but it does mean the exchange is looking at a lot of activity traders can’t see. I’m not interested in quoting markets with that much hidden-answer risk, unless retail flow gets so ridiculous that it pays enough to take it.

Risk scoring and employment checks are the useful part: they aim to stop high-risk traders before they enter. That protects traders before they get picked off.

The problem is what happens once a market is already under review. Investigations stay confidential until cases close. At least let traders decide whether it’s worth staying.

Even the exchange has described the cost of that silence. In a recent CFTC comment, it noted that traders who think insiders are picking them off widen spreads or leave. It argued for enforcement instead of removing vulnerable markets. Enforcement may be the right answer later. It can’t repair a blind quote made before settlement.

After the Perez report, head of enforcement Robert DeNault said the team promptly flagged and referred the trades to the CFTC. Credit for speed. The post still doesn’t say when users were warned, whether trading paused or whether settlement was delayed. Fast enforcement can still be late for traders.

Domer, an elite prediction-market trader, asked: “At what point in the investigation did you notify the traders in these markets?” Exactly.

Robert DeNault’s response and Domer’s reply.

Kalshi’s March 2026 Washington billboard.

That’s what the billboard skips. Freezing the account may help later, but traders need a warning while the market is live.

What I’d want answered before quoting another mention market:

• When did surveillance flag the activity?

• Did the exchange pause the market or only investigate the account?

• Were traders told the market was under integrity review?

• When profits are clawed back, what happens to traders on the other side?

• Do penalties compensate harmed traders or go elsewhere?

• How long can it wait to pay after the source revises its data?

• What would make the exchange stop listing a category with repeated integrity problems?

The company doesn’t need to name the trader or publish evidence mid-case. Put a generic notice on the market: this contract is under integrity review and settlement may be delayed. Then traders can get out or widen their quotes.

When a case closes, skip the case file. Tell us whether trades were cleared or canceled, profits were recovered and counterparties were made whole. An invisible investigation doesn’t help anyone price the risk.

Some markets aren’t worth it

That’s why I agree with Dustin Gouker, who argued in The Event Horizon that mention markets have almost no hedging or price-discovery value and are ripe for manipulation and insider trading. You can be great at trading a market that still shouldn’t exist.

Sports aren’t a big part of my trading. They’re the clean part of the business: the event happens in public, and nobody has tomorrow’s final score sitting in a Google Doc. GoldenPants13 captured the joke: embarrassed by the clean 93%, defending the other 7%.

GoldenPants13’s 93/7 split.

A regulated exchange shouldn’t need a government Spotify-streams detective bureau. If policing only works after payout, don’t list the market. If the exchange wants the fees, it gets the surveillance bill too.

Test the market before listing it

Before an exchange lists a market, somebody should try to break it. Start with five questions:

1. Does a final answer exist privately before the public event? Prepared speeches, unreleased rankings, edited videos and confidential product announcements are obvious examples.

2. How many people can access it? One decision-maker is risky. A campaign staff, production crew, government agency or technology company creates a much larger screening problem.

3. Can the subject manufacture the event? Attendance, candidacy, social-media posts and personal announcements can turn the subject into both the underlying and a potential trader.

4. What is the manipulation ratio? The cheaper it is to move the source relative to the available profit, the less the market resembles forecasting.

5. Can detection and correction happen before settlement? A source that revises next-day data can’t safely support a contract that pays in thirty minutes.

The answers should decide whether a market lists. If the answer exists privately, the insider pool is broad or the source is cheap to move, don’t list it.

The exchange’s new risk-scoring framework asks similar questions and says lower-importance, high-risk markets may be rejected.

Better research should win

Traders don’t need equal information. The edge should come from research, not a draft speech or a botted chart.

If you choose a dumb contract and get picked off, that’s on you. If Kalshi knows the contract may be compromised and still lets traders quote it blind, that’s on Kalshi.

Until then, we know who eats the loss.

I trade and make markets on Kalshi, and I work on the data infrastructure at Kalshinomics. Nothing here is investment advice.