I made a flowchart as a joke.

“So You Want to Get Housed?” Start at the top with a simple question. Do you have somewhere affordable to live right now? If yes, congratulations. Please don’t move.

If no, the adventure begins.

Can you afford market rent. Do you make three times the rent. Have you applied for affordable housing. Is the waitlist open. Are you on it. Did your contact information change. Do you still have the documents you submitted. Are they current. Can you prove your income. Can you prove you are homeless. Can you prove you are about to become homeless. Do you qualify. Is there actually a unit. Does the landlord accept the subsidy.

It took about twenty minutes to draw. It was funny for about ten of them.

Then I tried to finish the arrows, and that is where the joke came apart.

I did finish them, eventually. That is the part worth telling you about. Every single correction I made to that chart was the same correction. I was adding loops.

The waitlist closes, and you go back. The letter does not reach you, and you go back. The documents are too old, so you upload them again, and the new set is too old by the time somebody opens it. The unit exists and the landlord will not take the subsidy, so you go back. Nearly every box I fixed turned out to route backward into a box I had already drawn.

By the second draft there was exactly one arrow pointing off the page, and it went to a green box that said HOUSED, with a note underneath about recertification in eleven months. Which is not an exit. It is a loop I had not drawn yet.

So the problem was never that the chart was too complicated. The problem is that a flowchart is a lie about how this system works.

A flowchart says that at every point, someone asks a question, someone gives an answer, and the answer routes you somewhere. That is not what happens to most people. Most people never reach a box. They fall out between them. And the falling out gets recorded as nothing at all.

I should tell you what I am to this system, because it is not an observer.

I have written administrative plans.

If you do not work in housing, that phrase means nothing, which is itself part of what this essay is about. An administrative plan is the document that says how a housing authority actually runs its voucher program. Federal regulation requires every PHA to adopt one in writing, requires the board of commissioners to formally approve it, requires the agency to revise it whenever HUD changes what is required, and requires it to be available for public review.[13]

The regulation also lists, by name, what the plan has to cover. Item one on that list is the selection and admission of applicants from the waiting list, including any admission preferences, the procedures for closing and reopening the list, and the procedures for removing applicant names from it.[13]

The procedures for removing applicant names from it.

I have written that chapter.

These documents run to hundreds of pages, in a vernacular built for auditors and compliance monitors. They are not user friendly. I want to be exact about this, because it is the part people misread as an accident: they are not meant to be. That is not a drafting failure. It is the audience the document was written for.

And in an earlier part of my career, working as a case manager, I told people to apply everywhere.

Apply in the next city. Apply in the next county. You can move, and if a voucher comes through you may be able to take it with you, and the receiving housing authority can either bill your original agency or absorb you into its own program.

Read that back as somebody hearing it for the first time, in an office, out loud, once.

Portability. Billing. Absorption. Every one of those is a real mechanism with real rules and real consequences. Every one of them is also a word that means something precise to a few tens of thousands of people in this country and nothing at all to everyone else.

The advice was correct. I still think it was correct. I am going to come back to what it costs.

Eligibility is not access, and we keep spending the difference

The oldest mistake in American social policy is assuming that creating a program creates access to the thing the program provides.

Those are not the same thing.

A person can qualify for housing assistance without receiving housing. A person can qualify for shelter when no bed exists. A person can hold a voucher and never find a landlord who will take it. A person can qualify for four benefits and be unable to operate the administrative machinery required to collect any of them.

On paper, the answer is yes. Operationally, the answer is not yet. And if not yet lasts long enough, it becomes no without anyone ever having to say no.

That gap has a price tag. The Office of Information and Regulatory Affairs opens its 2023 burden reduction report by citing Matthew Desmond’s estimate that more than $140 billion a year in benefits Congress has already authorized goes unclaimed.[1]

Congress appropriated it. The eligibility rules admit the people. The money sits there.

We wrote the authorization and never built the delivery.

The mechanism has an author

Here is the study that changed how I read any agency’s operations.

Huiyun Kim collected the Housing Choice Voucher administrative plans from Michigan’s local public housing authorities. Michigan has 63 of them. She obtained plans for 59, a 94 percent collection rate, and the four she missed simply never answered her. Then she interviewed PHA directors and program managers about how those plans actually run.[2]

Her review found a universal purging procedure across all 59.

Once a year, the agency mails an update request to the address on file. The household has a limited window to respond. If they do not respond, they come off the list.

Not most of them. All of them. Fifty-nine out of fifty-nine, using mail as the mode of contact, with a time-limited response.

None sent a reminder. None used a second method of contact. None allowed reinstatement except when the agency itself had made the error.

Kim calls this a rationing algorithm, and the term is exact. It is a rule that decides who receives a scarce good. It decides on the basis of reachability.

Then she does the thing that makes the paper matter. She checks who is reachable.

Among renters income-eligible for vouchers, 45 percent moved at least once in a two year window. Roughly 14 percent moved twice. Households below half the poverty line moved at a meaningfully higher rate than households well above it.

So the algorithm systematically removes the poorest applicants from the list of people waiting for a program built for the poorest applicants. Not by targeting them. By targeting the one trait they disproportionately have.

More than 40 percent of those plans also carry a residency preference, which requires verifying an address when you reach the top of the list. The preference rewards whoever managed to stay put during a wait that runs, on average, past two years.

This is what I mean when I say mechanism over intention. Nobody in those 59 agencies wants this outcome. Kim’s interviews are clear that directors named underfunding as the reason they cannot afford a second attempt at contact. I believe them. I have been the person doing that math.

The outcome is produced by the instrument, not the motive. And the instrument is one page of an administrative plan that a board can amend at any regular meeting.

Longtime readers will recognize this as the primary source underneath the unreachable rate. I have been using that term for a while. This is where it comes from.

Procedures for removing applicant names

Notice what Kim’s method actually was. She did not obtain anything confidential. She did not FOIA a hidden file or catch anybody doing something they were concealing.

She read the plans. Some she downloaded off the internet. The rest she asked for.

Every rule in her paper was already public, because federal regulation requires it to be. The plan must be available for public review. It must state the procedures for removing applicant names from the waiting list. The board has to approve it in an open meeting. When HUD changes a requirement, the agency has to go back into the document and update it so that everybody knows.

So the purge is not a secret. It is a published rule, formally adopted, sitting in a document any member of the public can request.

And essentially nobody who is subject to it has read it.

That is the gap this whole essay lives in. Availability is not comprehensibility. A rule can be fully disclosed and completely unavailable at the same time, and the mechanism that makes it unavailable is not secrecy. It is length, and register, and the fact that the document was written for auditors.

I have watched this from the drafting side. An administrative plan update is a genuinely laborious piece of work. HUD issues a change, and it has to be incorporated correctly, in the right chapter, in language that will survive a compliance review. That is a real constraint and the people doing it are not being lazy. The register exists because the reader the drafter is imagining is a monitor, not an applicant.

But here is the consequence, and it is the thing I keep coming back to.

The rule that removed you from the list is in that document.

So is the rule that could put you back.

Both of them are public. Both of them are on some page in the forties, under a heading you would have to already know to look for, in a document that runs longer than most novels.

This is why I spend as much time as I do breaking these things down and explaining what they mean. Not because translation is a nice thing to offer people. Because the translation layer is the only part of this that is missing, and it is missing on purpose in the narrow sense that nobody was ever assigned to build it.

The compound denial

So let me name the thing the flowchart cannot draw.

The compound denial: a refusal produced by stacking requirements that are each defensible on their own, issued by no one, recorded nowhere, and therefore impossible to appeal.

Every box on that chart can be defended. I know, because I have defended all of them.

We verify income because subsidy is scarce and should reach eligible households. We require documentation because public dollars require accountability. We recertify because circumstances change. We keep waitlists because demand exceeds supply. Landlords screen because they carry risk. Programs carry geographic and categorical restrictions because the funding carries them first.

Every requirement survives its own audit.

A reasonable requirement plus a reasonable requirement plus a reasonable requirement does not produce a reasonable system. It produces an obstacle course. And an obstacle course produces refusals that no individual is accountable for, because no individual issued one.

A real denial has four properties. An author. A date. A reason. An appeal.

The compound denial has none of them.

It never shows up in any denial statistic, because it is not a denial. It shows up, if it shows up anywhere, in the inactive column. That is the administrative status we assign when a person stops generating transactions.

Which is a receipt, not a metric. It records that something happened. It does not tell anyone to do anything.

The filter catches the qualified

If the compound denial were mostly screening out ineligible people, this would be an efficiency argument, and a boring one.

It is not.

Matt Unrath’s work at the California Policy Lab examined who leaves CalFresh, California’s food assistance program, and when. Enrollees are six times more likely to drop out in a month when paperwork is due than in a month when it is not. And at least half of the people who leave, possibly as many as three quarters, still appear income-eligible on the way out.[3]

The paperwork is not a fraud filter. It is a sorting mechanism, and what it sorts for is administrative capacity.

The same pattern shows up in a completely different program with completely different rules.

Under the Total and Permanent Disability discharge, borrowers found permanently disabled can have federal student debt cancelled. For years the program attached a three year income monitoring period. Of roughly one million borrowers who received an initial discharge, more than half had the debt reinstated because they did not respond to a request for income documentation.

When the Department of Education later matched those files against IRS records, 92 percent of the reinstated borrowers had never exceeded the earnings threshold.[1]

Read that twice.

More than half a million permanently disabled people had cancelled debt restored to them for not answering a letter. Almost every one of them still qualified.

That is the compound denial operating at national scale, in a program with no waitlist, no landlord, and no housing supply constraint whatsoever.

Which tells you the mechanism is not about housing. Housing is simply where we stack the most of it.

The OIRA report opens with a disability beneficiary describing what recertification felt like. The comparison they reached for was two rounds of cancer.[1]

There is no line. There are fifty-five thousand lists.

Everything above describes one waitlist, at one agency, and that framing is already too generous.

There are roughly 3,300 public housing authorities in this country. Each runs its own waitlist, sets its own preferences, and writes its own administrative plan.

And the voucher is not where most affordable units come from anymore. The Low-Income Housing Tax Credit is. HUD’s database counts 55,345 LIHTC projects and 3.9 million units placed in service between 1987 and 2024.[8]

LIHTC is not a HUD program. It is Section 42 of the Internal Revenue Code. The IRS writes the rules. Sixty-one allocating agencies, mostly state housing finance agencies, administer it and monitor compliance at the project level.[9] The properties themselves are privately owned and privately managed.

Which means there is no queue.

Each property keeps its own waitlist, under its own Tenant Selection Plan, written by ownership and approved by the state. Applicants apply property by property. There is no central list to be on. There is no single place to update your address.

So the annual letter is not run once. It is run tens of thousands of times, independently, by managers who have never spoken to one another.

I have read a number of these plans. They say the same thing. Here is a typical one: the property mails a letter asking the household to confirm continued interest in remaining on the list. Applicants must respond within the stated timeframe to keep their position. Applicants who do not respond, or whose letter comes back undeliverable, are removed.[10]

Same algorithm. No coordination.

And notice what that does to the advice we give people. This is the part I said I would come back to.

I told people to apply everywhere. Every navigator, every case manager, every housing counselor I have worked alongside says some version of the same thing, and we are right to say it. When the odds at any one door are that bad, the only rational move is more doors.

But the advice has a second half nobody says out loud, because most of us have not done the arithmetic.

Each list you join is another address to maintain. Another annual letter to catch. Another independent chance to be removed by an office that will never tell the other offices what happened.

The rational strategy increases the number of ways to disappear.

And here is the part that appears in no policy document anywhere, because it is nobody’s job to notice it.

Nobody sends you a statement of the lists you are on.

There is no place to look up which ones are still open, which ones closed, which ones already dropped you, or when the next letter is due from each. You track it yourself or it does not get tracked. Which property. Which portal. Which login. Which renewal month. Thirty applications means thirty calendars, none of them aligned, none of them reminding you, each running its own independent clock.

I have used the phrase staggered calendar before in a different context. This is the purest version of it I have found, because here the deadlines are not merely uncoordinated. There is nothing that could coordinate them.

The one thing that reliably reduces this burden is worth sitting with, because it is uncomfortable.

If the same company manages several properties, you often get one application and one list across all of them. Large operators run shared waitlist software that consolidates a whole portfolio into a single record. One vendor advertises a client who moved roughly fifteen properties off separate waitlist methods onto one platform.[15]

So the burden falls when ownership concentrates.

That is the only integration in this system that works reliably at the property level, and nobody designed it as a service improvement. It arrived as a side effect of consolidation in the affordable housing industry, which most of us in this field treat as a preservation risk. Both things are true at once. The applicant’s experience improves as the owner gets bigger, and that improvement stops exactly at the edge of the portfolio and not one property further.

An applicant cannot see any of this from the outside. You cannot tell, looking at two listings, whether they share a waitlist or run two. So you apply to both, and you track both, and you find out which it was only if one of them writes to you.

A handful of communities have tried to fix the front of this. San Francisco built DAHLIA, a single portal with a single stored profile that feeds a short application across the city’s affordable listings. By 2021 it was carrying 97 percent of the city’s affordable housing applications, and it was built open source specifically so other jurisdictions could adopt it. San Mateo County did. Santa Clara County and San Jose partnered on a local version.[11] New York runs Housing Connect, with more than a million registered users.[12]

Those are real and they matter. They are also not the same fix, and I want to be careful here, because the portal is the thing everyone reaches for and then declares the problem solved.

A portal fixes discovery and intake. It does not, by itself, fix the purge. Housing Connect is built around lotteries, not standing waitlists. A single application profile is not automatically a single address of record propagating to forty separate property managers on forty separate annual cycles.

The portal is the precondition, not the remedy. Building one and stopping is a nicer front door on the same building.

The appeal is attached to the wrong event

Here is the part I keep coming back to.

Federal law does protect voucher applicants. Under 24 CFR 982.554, when a PHA denies assistance, it must give the applicant prompt notice, state the reasons, tell the applicant an informal review is available, and describe how to get one.[13]

That is a real protection. It is enforceable. It is also almost entirely beside the point.

Because a purge is not a denial. Nobody made a decision denying assistance. No decision, no notice, no review. The protection is attached to an event that happens far less often than the outcome it was written to protect against.

Some agencies extend review to waitlist removal anyway, on their own authority. Alameda’s housing authority offers an informal review when an applicant’s name is being removed from a wait list.[14] I raise that not as a gotcha but as proof of the only thing that matters here: it is discretionary. Which means it is available. Which means declining to offer it is a choice somebody made.

On the LIHTC side there is no federal analogue at all for applicants. Section 42’s enforcement mechanism is credit recapture against the owner for failures of income certification and occupancy compliance. It is a tax compliance regime, not a due process regime. Existing tenants have good cause eviction protection through Revenue Ruling 2004-82, but that governs eviction, not admission. An applicant removed from a LIHTC waitlist has whatever the Tenant Selection Plan grants them, plus fair housing law, plus whatever the state requires. In most states that is not much.

And tenant selection criteria run on exactly the same logic. Credit thresholds. Criminal history lookbacks. Rental history verification. Minimum income requirements, frequently two or two and a half times the rent, inside a program built to house people whose incomes are low. Every one of those sits in the plan. Every one is defensible standing alone. Every one produces an exclusion that the applicant has to contest personally.

Which brings me to the sentence this essay has been walking toward.

In every one of these systems, the remedy is assigned to the applicant.

You have ten days, or fourteen. You have to know you were removed. You have to write, in the required form, to the required office, before the clock runs out.

And the system just finished demonstrating, by removing you, that it could not reach you.

The window to appeal opens with a letter mailed to the address that failed.

That is not an oversight in the design. That is the design, and it is the cheapest thing on this entire page to change.

The Housing Notes goes out every Sunday: one long read on housing, homelessness, and the machinery underneath both, written by someone who has operated it. Free. Subscribe and it arrives in your inbox.

Cracks are accidental. This is not.

We say people fall through the cracks. I have said it in public testimony.

Pamela Herd and Donald Moynihan have spent more than a decade dismantling that metaphor. Their framework breaks administrative burden into learning costs, compliance costs, and psychological costs. Their central claim is the one worth arguing with: the level of burden placed on a person, and how it is distributed between the state and the individual, is usually a deliberate political choice rather than historical accident or neglect.[4]

Burden is allocated.

Someone decides whether the agency assembles the medical record or the disabled person does. Someone decides whether one letter is enough. Someone decides whether a document expires in thirty days or three hundred.

Those decisions get made by people with names, in meetings with minutes, and they can be made differently on a Tuesday.

A crack is something nobody intended. What we built is a series of allocations, and every box on that chart marks a place where an institution decided a cost belonged to the applicant rather than to itself.

This is the integration tax again, moved to the front door. The resident becomes the coordination layer between institutions that do not talk to each other. Then we call that person difficult to serve when the coordination fails.

The end of the chart is worse than the middle

Suppose you survive it. You answer every question, hold every document, keep an address for two and a half years, and reach the top of the list.

In 2022, 57 percent of new voucher holders leased a home within a year. In 2018 it was 66 percent. The median successful search stretched from 59 days to 78.[5]

The other 43 percent had already been found eligible. Every one of them had cleared every box on the page.

What stops them is the part of the system we do not administer.

Urban Institute testers screened more than 341,000 rental listings across five metros to find 8,735 units that were both available and inside voucher rent limits. That is roughly 39 advertisements per usable unit.

Then they called. Outright landlord denial ran 78 percent in Fort Worth, 76 in Los Angeles, 67 in Philadelphia, 31 in Newark, and 15 in Washington, DC.[6]

The two low numbers are the two jurisdictions with source of income protections. That is a policy variable, not a market fact.

And the distribution inside a single city is the part that has to be named directly. In Philadelphia, denial ran 82.5 percent in census tracts with poverty below 10 percent, and 55.3 percent in tracts with poverty above 30 percent.[6]

The rejection rate climbs as you move toward opportunity.

The Furman Center’s analysis of lease-up finds Black and Hispanic voucher recipients less likely to lease successfully than other recipients in the same markets.[5]

The voucher is race-neutral on its face. The sorting is not.

This is the apartheid architecture doing what it does, which is to produce a racially organized outcome through an instrument that never mentions race. I am not reaching for that phrase as a flourish. I mean it as a description of how a facially neutral rationing instrument distributes a scarce good along the same lines the explicit instruments used to.

What this does not fix

One honest caveat, because this series does not get to skip the hard part.

Fixing the compound denial does not create units.

The National Alliance to End Homelessness has been direct that first-time entries into homelessness continue to outpace exits into permanent housing.[7] If shelter is full, outreach cannot manufacture a bed. If permanent supportive housing is full, a case manager cannot manufacture a unit. If production trails demand, an application cannot manufacture an apartment. If someone exits homelessness while another household loses housing the same week, a system can work extraordinarily hard and watch the visible number barely move.

So this is not the argument that better administration ends homelessness. It does not.

It is the argument that scarcity is hard enough without manufacturing more of it procedurally.

We are currently running two rationing systems stacked on top of each other. One is made of concrete. One is made of paper. Only one of them requires a construction timeline to fix.

What to actually do, with proof it works

The restoration section of these essays has to carry working precedent, not aspiration. Here it does. Every item below has already been implemented somewhere, at scale.

1. Determine eligibility from data the government already holds. Education, SSA and the VA built a disability data match and automatically discharged the debt of more than 450,000 permanently disabled borrowers with no application at all.[1] The determination that used to require a physician’s certification and a form now requires nothing from the person. The precedent exists. It is federal, it is recent, and it is large.

2. Move the collection burden from the applicant to the agency. SSA redesigned the Continuing Disability Review so the agency gathers records from medical providers instead of asking beneficiaries to reassemble their own medical history. It cut the paper form by 20 percent and removed two essay questions that public comment had identified as burdensome and unreliable.[1] Same eligibility standard. Different party carrying the work.

3. Stop asking for the same signature every year. HUD’s 2023 HOTMA final rule eliminated the annual consent form. Sign once. HUD’s own estimate is ten minutes per form across 4.5 million households, roughly 750,000 hours returned every year.[1] That is not a pilot. It is in force.

4. Accept self-attestation and fact-specific proxies where the risk is low. Treasury’s Emergency Rental Assistance guidance allowed grantees to use residence in a high-poverty area, or receipt of another means-tested benefit, as evidence of income eligibility.[1] It moved billions of dollars with documented practices. It is sitting there as precedent the next time someone in a room says self-attestation is impossible.

5. Build the central application, then make it propagate.

Start with the fact that makes this the least defensible item on the list.

Congress already ordered it. The Economic Growth, Regulatory Relief, and Consumer Protection Act, signed in May 2018, directed HUD to make available, within one year, one or more software programs to facilitate the use of a shared waiting list by multiple public housing agencies or assisted owners, and to publish procedural guidance for implementing them.[16]

The word carrying the weight in that provision is voluntary.

Eight years on, this is not a question of whether shared waitlists are permitted, technically possible, or federally sanctioned. They are all three. It is a question of whether anybody has to.

The precedents are past the pilot stage. Delaware built a statewide common application covering the state housing authority and the Dover, New Castle County, Newark, and Wilmington authorities, so a resident applies once across all affordable housing programs in the state, including Section 8.[17] San Francisco’s DAHLIA is open source, carries 97 percent of the city’s affordable housing applications, was adopted by San Mateo County, and was localized by San Jose and Santa Clara.[11]

The build cost is paid. Somebody already wrote the code and gave it away.

But adopt it with the second half attached. A stored profile is only worth something if a single address change reaches every list the applicant is on, including the privately managed properties that sit in nobody’s queue. That is an agreement between a city, its housing authority, and its regulated properties, and it is a term you can write into a regulatory agreement at the next allocation round.

A portal that only handles new applications is a nicer front door on the same building.

6. Flip the reinstatement default. Right now the burden of undoing a purge sits on the person the agency has already proven it cannot reach.

Reverse it. Reinstatement to the applicant’s original position, automatic on request, within a stated window, with no showing required. The agency does not need a hearing to restore a position it removed without a hearing.

And set the standard on the front end: two documented contact attempts, by two different methods, before removal. If the agency cannot document them, the removal does not stand.

Neither of those requires a statute. Both are administrative plan language, and for LIHTC, both are Tenant Selection Plan language that a state allocating agency can require as a condition of the credit.

7. Amend the waitlist maintenance section of your administrative plan. Then publish a version people can read.

This is the cheapest item on the list and the closest to home. Multiple contact methods instead of one. A reminder before the purge. Reinstatement available for applicant hardship, not only for agency error. A standing arrangement with the outreach providers who are already standing in front of the same people you are mailing letters to.

Kim’s paper is effectively a list of parameters. Every one of them is set locally, by a board, in public, with no federal permission required.

Then do the second thing, which costs a staff week and no money at all.

The plan is already required to be public. Publish a one-page companion that says, in plain language, the four things an applicant actually needs: how you contact people, how long they have to respond, what happens if they do not, and how to get back on. Put it on the website next to the plan and hand it out at intake.

You are not creating new policy. You are stating the policy you already adopted, in the language of the people it governs rather than the language of the people who audit it. And the same requirement belongs in tenant selection plans, which a state allocating agency approves and can condition.

Nobody has to lose an argument for this one to happen.

8. Publish the purge.

This is the ask.

No housing authority in the country routinely reports these two numbers side by side:

• How many households we removed from the waitlist this year.

• How many of them were still eligible when we removed them.

The second number is knowable. CalFresh researchers estimated it. The Department of Education estimated it against IRS data. It is a match, not a study.

If you produce those two numbers and the second one is large, you have converted a compound denial into an actual denial, with an author and a date and a reason attached to it.

At that point somebody has to defend it.

Defending it is how it changes.

What I am trying

Fair warning on this last part. It is not evidence. It is a thing I am building, and it has not been tested on anybody.

The propagation problem in plank five has an obvious institutional answer, which is that every agency and every property integrates with every other one. I have watched that answer fail for most of my career. It fails because it asks tens of thousands of private owners to pay for something that buys them nothing.

So I have been prototyping the other direction. A benefits passport, held by the participant rather than shared between agencies. Their verifications travel with them. They present what has already been confirmed instead of each office deriving the same facts from scratch.

I want to be precise about what that does and does not do, because I have just spent an entire essay arguing that we should stop assigning coordination work to the person with the least capacity to carry it, and this could be read as the opposite.

The burden was never the holding. The burden is the reassembling. It is proving your income to the fortieth office because the first thirty-nine will not say what they already verified. A passport does not move work onto the applicant. It stops the work from being done forty times.

That is the theory. It is a proof of concept, not a finding. I have written about the idea separately. When there is something real to report about whether it works, including if it does not, that will be its own piece.

The last box

So yes. I made a ridiculous flowchart. The uncomfortable part is that I had to simplify it, and that the simplification is not where the dishonesty lives.

The dishonesty lives in the boxes.

Boxes imply decisions. This system’s most common outcome is not a decision. It is a silence we later describe as a status change.

A functioning housing system should not require extraordinary persistence from people already living in extraordinary instability. The target state is deliberately boring.

A person needs help. They enter through one door. The system determines what they qualify for using information it already has. Their record follows them, with their consent. Somebody can see where they are stuck. They get connected to a real resource. They move through.

That will not build a single unit. It will not replace shelter capacity, supportive services, prevention funding, or production.

It will stop us from spending scarcity we already have on people who already qualified.

Housing is complicated.

Getting help should not have to be.

Notes

[1] Office of Information and Regulatory Affairs, Office of Management and Budget, Tackling the Time Tax: How the Federal Government Is Reducing Burdens to Accessing Critical Benefits and Services, July 2023. Verified against the primary PDF. Source for the TPD discharge figures, the HOTMA consent form estimate, the SSA Continuing Disability Review redesign, the Treasury ERA practices, the 450,000 automatic discharges, and the beneficiary comparison. The $140 billion figure is OIRA citing Matthew Desmond, Poverty, by America (Crown, 2023), p. 241, and is attributed that way above.

[2] Huiyun Kim, “Failing the Least Advantaged: An Unintended Consequence of Local Implementation of the Housing Choice Voucher Program,” Housing Policy Debate 32(2): 369-385, 2022. doi:10.1080/10511482.2020.1834429. Author manuscript at PMC8993032.

VERIFIED AGAINST PRIMARY. The manuscript states directly that Michigan has 63 local PHAs, that Kim collected HCV administrative plans for 59 of them at a 94 percent collection rate, that the remaining four did not respond, and that her review revealed a universal purging procedure across the 59 using mail as the mode of contact with a time-limited response.

One live discrepancy on the residency preference share. The 2019 APPAM abstract says 40 percent of Michigan PHAs use a geographically bounded waitlist preference. NLIHC’s summary of the published article says 44 percent. The published paper is authoritative but the figure sits behind the Taylor and Francis paywall. The essay now says “more than 40 percent,” which is true under both. Change it only if you pull the article and confirm.

The mobility figures, 45 percent moving at least once and roughly 14 percent moving twice, remain. They come from NLIHC’s summary. PMC returned a captcha on attempted retrieval; try the manuscript directly from a browser.

[3] Matt Unrath, “Pushed Out by Paperwork: Why Eligible Californians Leave CalFresh,” California Policy Lab, 2021.

[4] Donald Moynihan, Pamela Herd and Hope Harvey, “Administrative Burden: Learning, Psychological, and Compliance Costs in Citizen-State Interactions,” Journal of Public Administration Research and Theory 25(1): 43-69, 2015. And Pamela Herd and Donald Moynihan, Administrative Burden: Policymaking by Other Means, Russell Sage Foundation, 2018.

[5] Ingrid Gould Ellen, Katherine O’Regan and Sarah Strochak, “Success Rates in the Housing Choice Voucher Program: 2018-2022,” NYU Furman Center, 2025.Direct PDF: furmancenter.org/files/Success_Rates_in_the_Housing_Choice_Voucher_Program_508.pdf. The 57/66 percent and 59/78 day figures come from NLIHC’s April 2025 summary; the Furman server rate-limited during drafting. The racial disparity finding is from the companion peer-reviewed article, Ellen, O’Regan and Strochak, “Race, Space, and Take Up: Explaining Housing Voucher Lease-Up Rates,” Journal of Housing Economics, 2023, which reports a 61 percent overall lease-up rate for 2015-2019 and an average wait of roughly two and a half years. Decide which of the two carries the disparity claim.

[6] Mary K. Cunningham et al., A Pilot Study of Landlord Acceptance of Housing Choice Vouchers, Urban Institute for HUD, 2018. Denial rates: Fort Worth 78, Los Angeles 76, Philadelphia 67, Newark 31, Washington DC 15. The 341,000 listings / 8,735 units screening ratio and the Philadelphia 82.5 versus 55.3 poverty-tract split are from the study and its Philadelphia site brief.

[7] National Alliance to End Homelessness, “7 Takeaways from 2024 Point-in-Time Count Data on Homelessness,” 2025.

[8] HUD USER, LIHTC Database Access, Property Data. 55,345 projects and 3.9 million units placed in service between 1987 and 2024. huduser.gov/portal/datasets/lihtc/property.html

[9] HUD PD&R, statutorily mandated LIHTC tenant data collection under 42 U.S.C. 1437z-8. The program is administered by 61 predominantly state-level allocating agencies; the IRS issues LIHTC regulations and ensures compliance with the code for individual taxpayers, while states administer the program and monitor compliance at the project level.

[10] Tenant Selection Plan language quoted in paraphrase from a representative LIHTC property plan.

[11] SF.gov, “Creating an online affordable housing portal,” and Urban Institute Housing Matters, “San Francisco’s One-Stop Affordable Housing Application Portal Could Be a Model for Other Cities,” 2022. DAHLIA carried more than 650,000 applications and 97 percent of the city’s affordable housing applications as of May 2021, was built open source, and was adopted by San Mateo County with Santa Clara County and San Jose partnering on a local version.

[12] NYC Housing Preservation and Development, NYC Housing Connect. More than one million active users as of the 2020 relaunch.

[13] 24 CFR 982.54, Administrative Plan, and 24 CFR 982.554, Informal review for applicant. Both verified against eCFR and Cornell LII.

982.54(a) requires the PHA to adopt a written administrative plan establishing local policies, formally adopted by the board of commissioners or other authorized officials. 982.54(b) makes it a supporting document to the PHA Plan under 24 CFR part 903, requires it to be available for public review, and requires revision as needed to comply with HUD requirements. 982.54(c) requires the PHA to administer the program in accordance with it. 982.54(d)(1) lists as the first mandatory subject the selection and admission of applicants from the waiting list, including any admission preferences, procedures for removing applicant names from the waiting list, and procedures for closing and reopening it.

982.554(a) requires prompt notice of a decision denying assistance, with a brief statement of reasons, a statement that the applicant may request an informal review, and a description of how to obtain one. 982.554(b) requires the review to be conducted by someone other than the person who made or approved the decision, or that person’s subordinate.

The essay’s central legal point turns on the interaction of these two provisions: the purge procedure is mandatory published content under 982.54(d)(1), while the review right under 982.554 attaches only to a decision denying assistance.

[14] Housing Authority of the City of Alameda, “Informal Hearings FAQs.” Under its property management policy, an informal review is offered when an applicant’s name is being removed from a wait list.

[15] Haven Connect, property manager marketing materials, describing a client that consolidated roughly fifteen properties from separate waitlist methods onto one platform. Cited as an industry example of portfolio-level consolidation, not as an endorsement of a vendor. Similar shared-waitlist functionality is marketed by RevoList and built into AppFolio’s affordable housing module.

[16] Economic Growth, Regulatory Relief, and Consumer Protection Act, Pub. L. 115-174, signed 24 May 2018, shared waiting lists provision.

[17] AffordableHousing.com, announcement of Delaware’s statewide centralized application, November 2024, covering the Delaware State Housing Authority and the Dover, New Castle County, Newark, and Wilmington housing authorities. The vendor describes it as among the first in the nation to integrate all affordable housing programs within a single state. That is a vendor claim about its own product and should be softened or independently confirmed.

Cupid Alexander writes The Housing Notes, a weekly series on housing, homelessness, community development, and the systems we build around them.

cupidalexander.com LinkedIn: Cupid Alexander

There is always another one.

I read one this week, and the argument never varies much. A homelessness industrial complex has a structural reason to under deliver. The sector needs the problem in order to keep drawing the check. That is why the numbers have not moved.

My first instinct was defensive, and I want to say that out loud, because the defensiveness is worth examining.

So there is a version of this criticism I will take, and I will help you write it.

This is not that version.

This version blames a forty year old apparatus for a century old wound, and it wants you to land somewhere very specific: that the people doing this work are protecting their jobs. That is sloppy, it is lazy, and it is wrong.

Now, I am not going to tell you what the people writing these pieces intend. I do not know. Intent cannot be audited, and guessing at it is the exact move I am objecting to. That is what the industrial complex argument is at its core: a claim about what people secretly want, offered in place of evidence about what institutions actually did. I am not going to answer it in kind.

But you can audit what an argument is built to do, and that is a different question with a checkable answer.

This one is built to do three things. It moves the blame to whoever has the least power in the room. It makes suspicion feel like rigor, so that no evidence is required of the person making it. And it releases everyone holding an actual lever, the zoning board, the appropriations committee, the neighbor at the podium, from having to do anything on Tuesday.

Notice what it never does. It never names a statute. Never a date. Never a rung, a unit, or a number. It ends in a mood.

And when someone asks this genre for scale, for cost, for an evaluation of the alternative, the answer comes back the same way every time. Proposing solutions is not the critic’s job. That is the government’s job. [c]

Which is a convenient arrangement, and it is worth saying out loud what it produces. Critique with no obligation to build anything. And once that is the standing rule, “industrial complex” stops being a finding. It is a mood with a citation style.

So it has a chronology problem, which is useful, because chronology is not an opinion. And it is one of the oldest arguments in American life wearing a new suit, because the old argument was never about efficiency either. It was about who counts as worthy of a solution.

The apparatus is younger than the wound

HUD produced the first national estimate of homelessness in 1984, a survey asking providers to guess. [1] The first federal law aimed directly at homelessness was signed in 1987. [2] We did not standardize the point-in-time count until January 2005. [3]

That is the whole machinery. The continuums of care, the counts, the rating and ranking policies, the coordinated entry systems, the conference circuit. Roughly forty years old.

You cannot blame the fire department for the fire because they arrived with equipment.

Now look at what we did in the seventy years before anybody counted.

We had a bottom rung and we removed it

Before there was a homeless services sector, there was a de facto answer, and it was private, cheap, and legal.

Boarding houses. Lodging houses. Residential hotels. Single room occupancy (SRO). During the nineteenth and early twentieth centuries, between a quarter and a third of urban households took in boarders. [4] In the 1950s, SROs were roughly ten percent of New York City’s housing stock. [4]

Then we made it go away, deliberately, by instrument.

New York banned new SRO construction in 1955. [4] Zoning codes across the country were rewritten to discourage anything that was not a conventional dwelling unit. And then, responding to city and state incentives designed to push conversion, owners destroyed two thirds of New York City’s remaining SRO units between 1976 and 1981, a State Assembly study found at the time. [5]

Not just New York. Chicago lost 32,000 rooms, about 80 percent of its stock, between 1973 and 1984. Boston lost close to 90 percent between the 1950s and 1985. Seattle demolished 5,000 units enforcing a single fire code change, and by 1982 nearly 30 percent of the remaining downtown units sat vacant. Cincinnati lost about 42 percent of its stock in the 1970s. [5] Nationally, the estimates converge on roughly one million rooms lost to code, zoning, conversion, and demolition. [4][6]

Be precise about what that means, because it is not a story about neglect.

This is the housing a person could rent this week without a co signer, a credit report, or first and last. Working an unstable job. Drawing a small check. Recently released. Newly arrived. Simply alone. We did not lose it. We outlawed it, and then condemned what remained for failing standards written after the buildings went up.

That is the bottom rung. We sawed it off.

And then we moved the people

On top of that, we ran two federal programs through the same neighborhoods.

HUD’s own 1971 Statistical Yearbook accounts for roughly 1.3 million Americans displaced from urban renewal sites as of mid 1971, and the program ran another three years. [7] Federal highway construction displaced more than 475,000 households and over a million people. [8]

The distribution was not random. In Newark, 77 percent of those displaced by urban renewal were Black in a city that was 34.1 percent Black in 1960. [9] Baldwin’s line about negro removal was not rhetoric. It was a summary of the data.

So run the sequence. We drew lines around the neighborhoods. We denied the credit inside the lines. We declared the result blighted. We condemned it. We routed the highway through what was left. We outlawed the cheapest rooms in the city and converted the rest to market rate.

And in 1984, we started counting the people standing outside.

The late count

I am flagging this as new terminology rather than treating it as established vocabulary.

The late count. a harm that goes unmeasured until after the damage is finished, so the measuring apparatus gets mistaken for the cause.

That is the entire logical error in the industrial complex framing. It takes the youngest institution in the room and hands it the oldest bill. The count arrived four decades after the demolition, so the count looks like the origin.

And this is why I cannot read the framing as merely sloppy. Sloppiness is random. This error runs in one direction every single time, and the direction is always away from the people who made the decisions and toward the people cleaning up after them.

So let me say the thesis plainly, because I have spent enough essays being careful about it.

There is no homelessness industrial complex.

What there is, is a false alarm. Someone yelling fire in a building that burned down forty years ago, pointing at the people holding the hoses.

And this is why I called it a distraction rather than an argument. An argument names a mechanism you could go change on Tuesday. This names a motive you can never verify, and assigns it to a villain who is already doing the work.

Are we surprised that our Black and brown neighbors are simultaneously overrepresented in housing insecurity, in homelessness, and in predatory lending? In January 2025, 745,652 people were counted homeless on a single night. Black Americans are about 14 percent of the population and roughly a third of that count. [10][11]

Those are not three findings. That is one instrument, applied consistently, surfacing on three different reports.

The four claims

You do not need to have read any particular piece to recognize this argument. It arrives in city council public comment, in letters to the editor, in podcast segments, in neighborhood threads, and in national opinion columns, and it is built from the same four claims almost every time. So take them one at a time.

One. It is an industry that profits from the problem persisting.

Everything above is the answer to that. The counting apparatus is forty years old. The demolition it is blamed for ran from the 1940s through the 1980s. And the sector that supposedly cannot deliver cut veteran homelessness by more than half, which I will get to.

Two. This is not a race or equity issue.

This is the one that gets stated most confidently and defended least. It usually arrives as a preference, some version of “I hate seeing this framed as a race issue,” and then slides into a claim of fact without anything underneath it.

So: 1955. 1949. 1956. 1976 to 1981. Two of 3,229 in Mississippi. Fewer than 100 of 67,000 in New York and northern New Jersey. Seventy seven percent of the displaced in a city that was thirty four percent Black.

Those are not interpretations. They are counts, taken at the time, by the agencies doing the work, every one of them above.

You are permitted to find that framing tiresome. You are not permitted to find it absent. It is in the loan ledgers.

Three. Most homeless people are addicts.

Usually phrased as a large majority, sometimes as the vast majority, almost never with a number attached.

The best available synthesis, a systematic review and meta-regression across high income countries, puts alcohol use disorders at 36.7 percent and drug use disorders at 21.7 percent among people experiencing homelessness. [a] A national survey summarized by HUD lands closer to one in five with a diagnosable substance use disorder. [b]

Elevated. Seriously elevated. Worth building a response around. Not a large majority. The number is available in about ninety seconds and it does not get looked up before it is used to characterize several hundred thousand people.

Four. The answer is rehab, or jail for those who refuse it.

This is the claim that does the most damage, and it deserves its own name, because I encounter it constantly.

It is the collapse of addiction policy into poverty policy.

Two different problems, two different evidence bases, two different sets of tools, folded into one sentence so that a treatment question can be answered with a housing budget and a housing question can be answered with a jail. They are related. People who are poor and unhoused have higher rates of substance use disorder, and being unhoused makes recovery dramatically harder. Relatedness is not identity. You cannot treat a seven million unit deletion with a rehab bed, and you cannot treat an opioid use disorder with a certificate of occupancy.

The two get merged for the same reason the industrial complex label gets used. It converts a structural problem, which would require you to do something about supply and land use in your own neighborhood, into a behavioral one, which requires you to do nothing except disapprove.

That is the pattern underneath all four. Each claim moves the explanation one step further from any decision a reader might have to participate in.

What you are actually describing

Now the part I want to say directly, because the polite version of this essay has been written enough times.

If you believe there is a homelessness industrial complex, you are not describing an industry. You are describing a government that underserved the same people so completely, for so long, that when it finally shows up with a resource, you cannot process what you are looking at.

It reads as excess to you because you have never been required to think about it.

That is the real inheritance in this country, and it is not money. It is exemption. Exemption from having to know how the thing you have got to you.

There were decades of policy that gave opportunity to people, deliberately, at enormous public expense. That opportunity was passed down. And the people holding it were never asked to look behind them at the mechanism. Never asked to share space. Never asked to share resources. Never asked to hold a common goal with anybody who did not already have one.

So the analysis stops at the front door, and anything past the front door looks like somebody’s racket, because it looks like effort that was never demanded of you.

I want to be fair about how that lands. Admitting inherited advantage is genuinely hard. It asks you to hold two things at once, that you worked and that the field was tilted, and most people are not practiced at holding two things at once about themselves. I am not pretending that is easy.

But weigh it honestly against the alternative.

For the people living inside the outcome, the hard part is not being asked to examine a comfort. It is sitting through a confident, evidence free account of their own life, presented as the finding, with nothing attached to the end of it. No fix. No rung. No date. Just a diagnosis that happens to exonerate everyone who could act.

One of those is uncomfortable. The other one is a wasted decade, and we have had several.

Veterans, which is the clearest example we have

You want the mechanism named without ambiguity? Use veterans. The example cuts both ways and both cuts are instructive.

The GI Bill was one of the largest deliberate wealth transfers this country ever executed. By the time the original bill expired, nearly half of 16 million World War Two veterans had used an education or training benefit. Veterans were 49 percent of college admissions in 1947. Nearly 2.4 million home loans were extended, and by 1955 the VA had backed 4.3 million loans, close to a third of all home lending in the country. [12][13]

That is not a safety net. That is a launchpad. We built the American middle class on purpose, with public money, in about a decade.

And Congress routed the administration of it through states and localities. On purpose. John Rankin of Mississippi chaired the committee that drafted it and made sure the money moved locally. [14]

The result: in 1947, Ebony surveyed 13 Mississippi cities and found that of 3,229 VA guaranteed home, business, and farm loans made that year, exactly 2 went to Black veterans. [15] In New York and the northern New Jersey suburbs, fewer than 100 of 67,000 GI Bill supported mortgages went to non-white buyers. [15] The VA adopted the fha’s racial exclusion practices wholesale, so a Black veteran could hold a valid federal loan guarantee and still find no bank that would lend, no agent who would show, and no owner who would sell. [16]

Same war. Same uniform. Same statute. Different verdict about who was owed something.

Now the other half, and this is the half that ends the industrial complex argument on its own terms.

Veteran homelessness is down 56.1 percent since 2010. The January 2025 count found 32,495 veterans, a record low since measurement began in 2009. It is one of the only sustained multi year declines in any subgroup during a period when nearly everything else rose. [17] At the end of FY2024, nearly 90,000 veterans were under lease with HUD-VASH vouchers, the most in the program’s history, and of the roughly 48,000 veterans permanently housed that year, about 96 percent were still housed at year end. [18]

Who delivered that?

The same providers. The same nonprofits. The same reimbursement contracts, the same notices of funding opportunity, the same procurement rules, the same allegedly self dealing sector.

The variable was never the sector. The variable was that we agreed veterans were owed a solution, and then we paid for one, consistently, across four administrations.

And note what model produced it. HUD-VASH is a voucher paired with VA case management. That is Housing First, operating at national scale, delivering the single best sustained outcome in American homelessness policy. The model most often attacked in these essays is the model with the receipts. I have said before that Housing First is a service model and should be defended as one, separate from whatever delivery mechanism a given jurisdiction bolts onto it. The veteran data is the cleanest evidence for that distinction I know of.

I want to name the actual variable, and I am flagging it as new:

The worthiness verdict. the political determination that a population is owed a solution rather than a management strategy, which is the real predictor of whether that population’s homelessness declines.

When we render that verdict, this sector performs. When we do not, we call this sector the problem.

I will not oversell it. The 2024 to 2025 veteran decline was only 1.2 percent, and 25 states recorded increases even as the national number hit a record low. [19] Sustained does not mean finished, and progress is not evenly distributed. But a 56 percent reduction over fifteen years, achieved by the same institutions people call a complex, is not a rounding error. It is a disproof.

Getting ahead of the two rejoinders

“California spent $24 billion and it got worse.”

I have read the audit. Read it past the headline. Read it like it was required reading to graduate as part of core curriculum. Here’s what I found;

The $24 billion covered more than 30 homelessness and housing programs administered by nine state agencies across five fiscal years. [20] The auditor evaluated five of them. Two, Homekey and HHAP, were found likely cost effective, and Homekey converted hotels at an average of about $144,000 a unit against $380,000 to $570,000 for new construction. The other three could not be evaluated at all, because the Interagency Council had not consistently tracked outcomes since 2021. [21]

Read that carefully. Every program the state could actually measure came back cost effective. The finding was that the state failed to collect the data, not that the money was stolen.

That is a data failure, and it is a serious one. It is exactly what I meant when I wrote that data without a decision attached is a receipt, not a metric. But a data failure and a corruption finding are different things, and the entire industrial complex argument depends on you not noticing the difference.

“then why did the numbers go up?”

They did. And in 2025 they came down. The January 2025 count fell 3.4 percent, the first year over year decrease since 2016, driven by an 11.2 percent decline in family homelessness. [10]

Which is the same lesson as the veterans. Resource a defined population deliberately and the curve bends.

And no service program on earth was going to outrun the underlying arithmetic. Over the last decade, every bit of growth in the national rental stock was in higher rent units, while the number of units renting below $1,000 fell by 7 million. [22] You cannot staff your way past a seven million unit deletion.

Yes, there is corruption

I am not skipping this, because skipping it is what makes the sector sound defensive.

I saw the Los Angeles reporting. The chief executive living in Hawaii, $907,923 in one year including a bonus larger than her base salary, $742,181 the next, at an organization with about $15 million in annual revenue that has taken at least $25 million in grants from the local homeless services authority since 2021. [23] That is indefensible. Where this kind of thing crosses into fraud it should be prosecuted, and in that same county it has been. [24]

But read the reporting all the way down. The Los Angeles times benchmarked that compensation against 16 peer organizations in the same county doing the same work. The median was about $160,000. [23]

The median is the story. The outlier is the headline.

It is also worth noting that the organization in question is primarily a domestic violence provider that runs homelessness programs, which is a small thing but a telling one. The label gets applied loosely because the label is doing rhetorical work, not analytic work.

Corruption exists in every field, public and private. Nobody proposes abolishing defense contracting, hospital systems, university endowments, or commercial real estate the first time an executive steals. They prosecute the executive and keep the function. Only here does one outlier get read as a description of everyone.

And consider what the job actually requires. Benefits eligibility rules across a dozen programs. Fund accounting and single audit compliance. Public and private contracting. Real estate finance and capital stacks. Clinical supervision. Labor law. Crisis response at two in the morning. Years of it, to be paid a median wage, to absorb public scrutiny for a systemic failure you did not design and cannot fix alone.

That is not a racket. It is one of the harder jobs in American public life, and the people doing it are being told they are the reason the problem exists.

So what is actually blocking production?

Something is, because production is not happening. And unlike motive, this one has been measured.

Einstein, Glick and Palmer went through years of land use meeting minutes across Massachusetts municipalities. Relative to voters in the same communities, the people who show up to comment are about 25 percentage points more likely to be homeowners, nine points more likely to be white, significantly older, longer tenured, and more male. Latinos are eight percent of voters in the cities studied and one percent of commenters. In Lawrence, a city that is 80 percent Latino, one commenter out of 42 across three years had a Latino surname. [25]

That is the composition. Here is the behavior.

Fewer than one in ten commenters raised property values. They raised traffic. Drainage. Parking. Safety. Character. [25][26]

Flagging this as new as well:

The traffic argument. an objection constructed to be neutral on its face and impossible to falsify, whose operative function is delay rather than denial.

That is the machine the original essay should have been about. New and overlapping process steps that function as check offs rather than decisions. Meetings convened to determine the meaning of a word written vaguely enough to require a meeting. A study to clarify the previous study. A continuance. A scoping session. A request for further community input from a room that does not resemble the community.

Nobody has to say no. The process says later, indefinitely. Later is a no with better manners.

And then nothing gets built. Not permanent housing. Not affordable housing. Not even shelter. And we look at what remains standing and call that the industrial complex.

I heard a line a while back and my first reaction was to feel attacked by it. I have come around. “there is nothing as dangerous as a liberal voting homeowner impeding progress.” You can vote correctly at the federal level twice a year and vote against a specific building every Tuesday night at seven, and only one of those votes has an address attached to it. The data above is not describing a caricature from somewhere else. It is describing rooms I have sat in, in cities that would describe themselves, accurately, as progressive.

This is the twin of the industrial complex argument, and they should be read together. One says the problem is the people paid to address homelessness. The other, the one we use for schools, says the problem is the children who attend the segregated ones. Both take a condition produced by an identifiable chain of policy decisions and relocate the agency to whoever has the least power in the room. Both feel like rigor because they sound skeptical. Both end at the same place, which is that nothing needs to be dismantled, because there was never a structure, only a failure of the people at the bottom of it.

That is not analysis. It is an exit.

The ladder

The ladder broke. It broke in specific places, on specific dates, by specific instruments, and we have the records. 1944. 1949. 1955. 1956. 1976 to 1981. The maps, the underwriting manual, the fire code, the conversion tax break, the local administration clause.

The work is three things. Find where it broke. Repair those rungs. And be honest about how high it was supposed to reach and for whom, because it was never one ladder. It was several, at different heights, assigned by geography and race, and pretending otherwise is its own dishonesty. You cannot repair a structure you refuse to describe.

And the reason to do this now, rather than argue about who is eating, is that the gap is widening mechanically.

A record 22.7 million renter households, 49 percent of all renters, were cost burdened in 2024, with 12.1 million severely burdened. The number of cost burdened renters grew by 2.3 million since 2019. [27] Eleven million extremely low income renters compete for 3.8 million affordable and available homes, a gap of 7.2 million. [22] And the 13 million renter households earning under $30,000 have a median of $210 a month left after housing, for everything else in their lives. [22]

Nobody is willing that into existence. That is arithmetic. We are going to need housing and homelessness assistance at scale for a long time, and the reason is not that a sector wants a client base. The reason is that we removed the bottom of the market, we are not replacing it, and incomes at the bottom did not move.

It will not go away because you hope at it. But we can end it, and we have proof, because we already did it once for one group using these exact people and these exact contracts.

The restoration

All four exist. None require anyone to become less cynical.

One. Relegalize the bottom rung. Washington’s House Bill 1998 requires cities to permit co living, the modern SRO, by right on any urban lot zoned for at least six multifamily units, and preempts the usual local workarounds including parking mandates near transit. Compliance was required by the end of 2025, and one estimate put the yield above 2,400 low cost units a year statewide. [28][29] Oregon passed a version in 2023, Hawaii followed, and Colorado removed caps on unrelated adults sharing a home, joining Iowa and Oregon. [29]

Two. Require development standards to be clear and objective. Oregon has required this for years and Washington adopted it in House Bill 1293. [30] This is the direct antidote to the traffic argument: if a standard must be written so it can be applied without a discretionary judgment call, the meeting about what the word means cannot be held. You do not defeat the objection. You delete the venue for it.

Three. Stop running funding sequentially. in California, each additional public funding source adds roughly four months and $20,460 per unit, and developers generally cannot compete for the next source without the prior award in hand. The state has stood up a consolidated housing agency and is proposing to align it with the Treasurer’s office so there is one application rather than a relay. [31][32] I called this the sequence lock in a prior piece and it still holds.

Four. Count what we remove, not just who is left standing. every January we count people. We publish no equivalent annual accounting of low cost units lost to demolition, conversion, code enforcement, and expiring affordability. That asymmetry is not neutral. If the only official ledger is a ledger of people, then people look like the problem. Put unit losses on the same page as the point-in-time count and the argument changes shape immediately, because the two lines move together and everyone can see it.

The close

There is a forty year old response apparatus that is slow, procedurally overgrown, and in real need of criticism. I will help you write that criticism. I have been part of some of what deserves it.

But it did not cause this. It arrived late to something already finished, started counting, and got handed the bill for the century that preceded it.

We removed the rooms. We moved the people. We drew the lines, denied the credit, and ran the highway through. We wrote a launchpad for one generation and administered it locally so it could exclude half of its own veterans. Then we built a filing system, and now we are debating the filing system.

Rebuild the bottom rung.

And the next time you are in a room where someone raises traffic, ask them what they would accept. Not what they object to. What they would accept, in writing, with a number and a date attached.

The answer to that question is the whole conversation. So is the silence.

If this is useful to you, subscribe. It is free, it lands Sunday, and the sourcing always comes with it. Share with a friend, and let’s continue to build together.

Notes

[1] HUD, “a A Report to the Secretary on the Homeless and Emergency Shelters” (1984). Built from provider surveys rather than a statistically reliable sample.

[2] Stewart B. McKinney Homeless Assistance Act, signed July 22, 1987; renamed McKinney-Vento in 2000.

[3] National Alliance to End Homelessness, on the first standardized point-in-time count in January 2005.

[4] Governing, “SRO housing, nearly zoned out of existence, could re-emerge,” on boarder households, the 1950s share of New York housing stock, the 1955 construction ban, and roughly one million units lost between the 1970s and 1990s.

[5] The Pew Charitable Trusts, “how states and cities decimated Americans’ lowest-cost housing option” (July 2025), citing a contemporaneous New York State assembly study on 1976 to 1981, and reporting the Chicago, Boston, Seattle, and Cincinnati figures.

[6] American Enterprise Institute, “the history of SROs” (2024), on approximately one million SRO units lost by 2000.

[7] HUD 1971 Statistical Yearbook, as compiled by researchers documenting forcible displacement: approximately 1.3 million Americans displaced from urban renewal sites as of June 30, 1971, derived from 310,908 families at 1970 census average household size plus 159,099 individuals. The derivation is the researchers’, not HUD’s. I have not used the wider three million aggregate estimate, which its own authors caveat as imprecise.

[8] U.s. Department of Transportation, as reported by AARP: more than 475,000 households and more than one million people displaced by federal roadway construction.

[9] Newark changing, “urban renewal by the numbers.” Single source.

[10] HUD, 2025 Annual Homeless Assessment Report to Congress, part 1, released May 2026: approximately 745,652 people homeless on a single night in January 2025, a 3.4 percent decrease from 2024 and the first year over year decrease since 2016, driven by an 11.2 percent decrease in family homelessness.

[11] NLIHC summary of the 2025 AHAR (Black Americans about 14 percent of population, about 33 percent of the count); USAFacts reports 32.7 percent against 13.5 percent. I have used “roughly a third” to avoid overstating a denominator difference.

[12] VA figures as reported in “many Black world war ii veterans were denied their GI Bill benefits,” war on the rocks (2020): nearly half of 16 million wwii veterans used education or training benefits; veterans were 49 percent of college admissions in 1947.

[13] Same source: by 1955 the VA had backed 4.3 million loans, including nearly a third of all home loans, face value about $33 billion. Note this is cumulative through 1955, not a single year figure.

[14] Richard Rothstein, as quoted in NBC reporting on Black veterans and the GI Bill, on Rankin’s role in ensuring local administration.

[15] Ebony’s 1947 survey of 13 Mississippi cities (2 of 3,229 VA guaranteed home, business, and farm loans to Black veterans), and the New York and northern New Jersey figure of fewer than 100 of 67,000, both via ira Katznelson, “When Affirmative Action Was White.” Independently confirmed by Snopes. Note the Mississippi figure covers home, business, and farm loans together, not home loans alone. Many secondary accounts get this wrong.

[16] Richard Rothstein, “The Color of Law,” on the VA’s adoption of fha racial exclusion practices.

[17] HUD 2025 point-in-time count as reported by VA (June 2026): 32,495 veterans on a single night in January 2025, a record low since measurement began in 2009 and a 56.1 percent reduction since 2010.

[18] USICH (November 2024): nearly 90,000 veterans under HUD-VASH lease at the end of FY2024, the most in program history; nearly 48,000 veterans permanently housed that year with roughly 96 percent still housed at year end.

[19] Stateline (August 2026): the 2024 to 2025 veteran decline was 1.2 percent, and 25 states recorded increases while 23 recorded decreases.

[20] California Legislative Analyst’s Office figure of $24 billion allocated across more than 30 programs and nine agencies over five fiscal years, as reported by CalMatters.

[21] California State Auditor (April 2024): of five programs evaluated, Homekey and HHAP were found likely cost effective; Homekey averaged about $144,000 per unit in its first round against $380,000 to $570,000 for new construction; three programs could not be evaluated because Cal ICH had not consistently tracked outcomes since 2021.

[22] JCHS, “Ten Takeaways from the 2026 The State of the Nation’s Housing,” including the 7 million decline in sub $1,000 units, the $210 median residual for the 13 million renter households earning under $30,000, and the NLIHC gap report figures.

[23] Los Angeles times reporting on 1736 family crisis center: $907,923 in 2023 including a $495,000 bonus, $742,181 in 2024, against a median of about $160,000 across 16 benchmarked Los Angeles County peer organizations; organization revenue of about $15 million annually and at least $25 million in LAHSA grants since 2021.

[24] Federal and state fraud charges filed against the chief executive of a separate Los Angeles homeless services nonprofit over roughly $23 million in public funds.

[25] Katherine levine Einstein, david m. Glick, and maxwell b. Palmer, “Neighborhood Defenders” (Cambridge, 2019): commenters 25 percentage points more likely to be homeowners and nine points more likely to be white than voters in the same communities; Latinos eight percent of voters and one percent of commenters; one Latino surname out of 42 commenters in Lawrence, 2015 to 2017.

[26] Same source. Note the summaries vary slightly: some report that fewer than ten percent raised aesthetic concerns and fewer still raised home values, others report fewer than ten percent raised home values directly. Either reading supports the sentence as written, and I have used the more conservative one.

[27] Harvard Joint Center for Housing Studies, “the State of the Nation’s Housing 2026”: 22.7 million cost burdened renter households (49 percent), 12.1 million severely burdened, an increase of 2.3 million since 2019.

[28] Washington House Bill 1998 (2024).

[29] The Pew Charitable Trusts, “new laws open doors to affordable shared housing arrangements,” on the 2,400 unit annual estimate, Oregon’s 2023 law, Hawaii’s legislation, and Colorado’s shared occupancy law joining Iowa (2017) and Oregon (2021).

[30] Washington House Bill 1293; Oregon’s clear and objective standard.

[31] Terner Center for housing innovation, UC Berkeley, “reducing the complexity in California’s affordable housing finance system.”

[32] Terner Center, on the California housing and homelessness agency and the FY2027 budget proposal.

[a] Gutwinski et al., “The prevalence of mental disorders among homeless people in high-income countries: an updated systematic review and meta-regression analysis,” PLOS Medicine (2021): alcohol use disorders 36.7 percent (95% CI 27.7 to 46.2), drug use disorders 21.7 percent (95% CI 13.1 to 31.7). The same review estimates any current mental disorder at 76.2 percent, which is a separate and much broader category and should not be substituted for the substance use figures.

[b] National survey summarized by HUD finding roughly one in five people experiencing homelessness had a substance use disorder.

[c] This is a characterization of a recurring pattern rather than a single source, but it is not hypothetical. One recent example: when a reader asked an Oakland publication for the scale, cost, and comparative evaluation missing from a piece critical of homelessness programs, the editors replied that producing solutions was not their job but the city’s. I have deliberately not built the essay around that exchange, and have not named the letter writers, who are private readers. The argument is the target.

This is Mike’s story

Nearly twenty years ago a man named Mike sat down across from me in a county office and I could not help him. He has given me permission to tell this. I have changed his name and nothing else.

Mike had been housed. The worked to get him from the street into housing. That is the part people skip when they hear a story like this. He had an apartment. The community had done the thing everyone claps for. Somebody found him, somebody qualified him, somebody handed him a lease, and for a while he slept indoors.

We did the wraparound part too, and I want credit for it before I take it away. We helped Mike open a bank account. We helped him put the electricity in his own name. Somebody sat with him and filled out both. On paper that is a success story, and if you had audited my file that quarter it would have looked like one.

Then the bill came, and his income did not cover it.

Which is where the second system starts, and the second system is where he actually died. Metaphorically.

To get help with the energy bill you apply. Low income energy assistance is an annual application with its own open window, its own income documentation, its own copy of the bill, its own proof of residence, at its own office. The local crisis fund is a separate application with different criteria and a different office. Neither one talks to the utility account we just helped him open, and neither talks to the other.

He got through it once. Somebody walked him through it, probably me.

Then the season turned and it all had to be done again. Same documents. New window. And by then the person who had walked him through it the first time had forty other cases. And he was passed to a ‘recertification specialist’. This person asked him to provide what he did before.

He slipped. Then he slipped again. Arrears built on an account that existed, in a name that was correctly his, attached to a unit he was legally entitled to occupy.

Service got shut off. It was cold. Mike got a propane tank and used it inside to keep warm, which is a fire hazard, which is a lease violation, and he was evicted for it.

I want to be careful, because the easy version is that Mike made a bad decision. He did. He also made it inside a very small box. Every account he needed was open. Every one of them was in his name. What he could not do was re-qualify, on four different calendars, for the assistance that made the accounts affordable.

So Mike went back outside. When I met him, his body hurt. I was a witness for his court case-for ‘arson’. When I saw him, he said it plainly, the way people do when they have stopped performing for professionals. His back hurt from the ground. His feet were a problem. He was tired in a way that sleep does not fix.

He had no phone. No bus ticket. No computer. He had a folder. I reached in my pocket- gave him a couple bucks I had. Yes, I wasn’t supposed to- it’s not scaleable. My meager $17 an hours was barely enough for me. But I slept on a bed. He hadn’t.

I was young. I was a benefits analyst with a case manager’s title, which I will come back to, and what I had was a fax machine. So I faxed. I faxed everything in that folder to everyone I could think of. Other housing options. Benefits. The utility. I felt useful, which should have been my first warning.

The answer came back within a few days, and it was not a denial. It was worse than a denial. They needed clearer copies. They needed the originals.

I remember explaining that to him. I remember watching him do the math while I talked. Not anger. Arithmetic. He was calculating what it would take to get original documents from three different agencies, in three different buildings, in a county with one bus line that went anywhere near any of them, with no phone to call ahead and no address to have anything mailed to.

He thanked me. He was polite about it.

He did not come back.

What actually happened

For a long time I filed that as a story about a man who gave up. That framing is comfortable and it is wrong, and I want to take it apart, because every piece of it is a decision somebody made.

The energy assistance program runs an annual application with a limited window. Somebody chose that, and it was a defensible choice, because you have to allocate a fixed appropriation across a season and you need a point in time to measure income against.

The crisis fund used different criteria than the main program. Also defensible. Different money, different rules, different board.

The bus did not go where he needed on a schedule matching when those offices were open. Nobody decided that on purpose. It is what happens when two agencies plan routes and hours without ever sitting in the same room.

And nothing in the account we opened for him knew anything about the assistance that was supposed to pay it. We connected him to a utility. We never connected the utility to the help.

And the agencies wanted originals. Not because anyone was cruel. Because a fax is a copy, and a copy can be altered, and somewhere upstream there is an audit that will ask why you accepted a copy.

Every one of those rules had an author with a good reason. Not one of those authors ever met Mike. And no single one of them did anything to him. That is the part that took me fifteen years to see clearly. There was no villain in Mike’s file. There was only a sequence.

Here is what I keep coming back to. Mike did not fail to complete a process. Mike completed a cost benefit analysis, correctly, with better information than any of us had, and concluded that the trip was not worth the odds.

He was right. Given what we were asking, he was right.

Half the battle is opening the accounts

Something I did not understand at twenty six, and that I want to say plainly now, because it explains Mike better than any of the rest of it.

There are two battles, and we only ever talk about the first one.

Battle one is getting in. Prove who you are. Prove where you slept. Prove what you earn. Prove what is wrong with you. Do it at four agencies that each ask separately and none of which will accept another one’s answer. That battle is brutal and it is the one that gets written about, including by me. It is also the one we are proudest of winning, because it produces a lease and a photograph.

Battle two is staying in. This is the one nobody warns you about, and it is the one that took Mike. Every account you open starts a clock. So does every subsidy that makes the account affordable. The housing subsidy recertifies annually. The food benefit runs its own cycle. Medicaid redetermines on a third. The utility has a payment date. The clinic wants a renewed authorization. None of those calendars know about each other.

And every one of those is a fresh application, not a renewal. You do not confirm that nothing changed. You prove it all again, from the beginning.

The mechanism for nearly all of it is mail.

Think about what a single piece of that mail actually requires. The letter has to arrive at an address that is still current. Somebody has to collect it before it is thrown out or taken. It has to get opened, and envelopes from government agencies are not neutral objects to a person who has been evicted or garnished or arrested. It has to be understood, which is not nothing, because the action item is usually on page two under a heading that does not say what will happen if you ignore it. The document it asks for has to exist and be findable. It has to be copied, which means a machine, which means money. It has to be returned by a deadline, which means postage or a trip. And then somebody on the other end has to process it in time, which is entirely outside your control.

Eight things have to hold. Every cycle. For every program.

Break one and the benefit stops. And when the benefit stops, you are not back at the beginning of battle two. You are back at the beginning of battle one, opening the whole thing again from nothing.

Now do all of that from the street. No mailbox. No place to keep a folder dry. No machine to copy anything. No phone to call and ask what the letter meant.

I have heard people call this a motivation problem. It is a logistics problem that we have been misreading as a character problem for forty years.

And here is the part that I think matters most, because it is about the person and not the paperwork.

When you have been out of the loop for a while, out of work, out of housing, out of contact, the volume itself is the barrier. Not any single step. The total. You look at what is being asked, and it does not read as hard. It reads as impossible. That is not a failure of will. It is an estimate, and for a lot of people it is an accurate one.

So they opt out. And I understand why. If I were handed that list with a folder and no bus fare, I am not confident I would do better.

Before you ask, no, the case manager does not do this

Somebody always raises a hand at this point and asks whether a case manager handles all of it.

Sometimes. Not usually the way you are picturing.

A large share of the people carrying that title are benefits analysts and financial analysts. I was one. The job is to read the program requirements, compare them against what the person has documented, determine eligibility, and write it up so it survives an audit.

The job is not how they got to your office. It is not how they will obtain the birth certificate, or cross the county to the next appointment, or copy anything.

And you frequently cannot do it for them even when you want to. Program rules, scope of practice, conflict of interest policies and liability restrictions variously bar staff from completing applications on somebody’s behalf, from transporting people in a personal vehicle, from handling their money, from vouching for facts they have not independently verified.

So the person doing eligibility cannot navigate. The person doing outreach cannot determine eligibility. The person doing housing search cannot touch benefits. There are people in this field who will walk somebody through the whole thing, and they are extraordinary, and there are nowhere near enough of them, and most of them are doing it outside their job description.

Which leaves exactly one person permitted to touch every part of the process. The one with no car, no printer, no mailing address, and no institutional standing.

Mike had me. I had a fax machine and no authority.

In program management we have a word for this

Here is the thing that took me two decades and a director’s chair to see, and it is the part I most want other practitioners to sit with.

Mike did not fail once. He failed the same step repeatedly. Re-qualify for the energy assistance. Miss it. Re-qualify. Miss it.

In program management, that has a name. When the same error recurs across cases, it is not an incident. It is a systemic error. And we have an entire professional apparatus for it.

If my eligibility staff got the same determination wrong three times, nobody would write up the staff member three times. We would run a root cause analysis. We would ask whether the guidance was ambiguous, whether the form invited the mistake, whether the training was wrong, whether the workflow set people up to fail. We would issue a corrective action plan. We would revise the desk procedure. We would document all of it and report it upward, because a repeated error is understood, correctly, as information about the design and not about the person making it.

We are good at this. Genuinely. It is one of the things public administration does well.

Now turn it around.

When the same person fails the same step three times, we have nothing. No threshold that triggers a review. No root cause analysis. No corrective action plan. No count anywhere of how many people fail at which step. The failure just gets attributed and filed, over and over, to a different name each time.

We built quality assurance that faces one direction.

Every discipline we apply to our own errors, we suspend the moment the error belongs to the person we serve. Same repetition. Same pattern. Same evidence of a design problem. Opposite conclusion.

And I want to be exact about what that means, because it is not an oversight. It is a choice about who is assumed to be trainable. Staff error implies the system needs fixing. Client error implies the client needs fixing. We have never justified that asymmetry. We have never even stated it out loud.

Everybody was frustrated, and none of it mattered

Something else about that period that I did not appreciate at the time.

Everyone in Mike’s story was frustrated.

The landlord was frustrated. He had done the socially responsible thing, taken a tenant with a voucher and a history, and now he had a fire hazard and a vacancy and a unit to repair.

The utility clerk was frustrated. She had no authority to waive anything and she had told forty people that week that they had to come in person.

I was frustrated. I had spent two hours on a fax machine and produced nothing.

My supervisor was frustrated, because she had a program with an occupancy target and a person who had just fallen out of it.

The county official above her was frustrated, because his numbers were going to look worse and he could not point to anything anybody did wrong.

And the neighborhood was frustrated, because Mike was outside again and now he was outside near them, and they had been told this problem was being handled.

Six or seven people, all annoyed, all correct, all doing their jobs.

Not one bit of that mattered to Mike.

I think about this constantly now that I sit in the rooms where the frustration gets expressed. We spend enormous energy on our own irritation with each other. Interagency friction, funding silos, the meeting about the meeting. It feels like work. It is not nothing.

But the frustration of the system is not the same thing as the failure of the system, and we consistently mistake the first for the second. A person outside does not experience our coordination problems. They experience a closed door and a bus that does not come. Our whole professional emotional life about this is invisible to the only person it is supposed to serve.

This is not a homelessness story

I told Mike’s version because it is mine. But nothing in it is specific to homelessness.

A mother applying for medical assistance for her kid brings a birth certificate, a pay stub, a lease, and a utility bill. Then she applies for food assistance and brings the same four documents to a different office. Then the school needs proof of residency, which is the lease and the utility bill again. Three agencies. Same paper. Three trips, taken during the hours she is supposed to be at work.

A senior applying for energy assistance produces proof of income, proof of residence, and a copy of the bill. Her Medicare paperwork asks for two of the three. Her property tax relief application asks for the other. She is eighty one. The offices are not close together. She does not drive at night, and in December it is dark at four thirty.

Nobody in either story is unhoused. Nobody has a disability. Nobody has a criminal record or a substance use history or any of the things we point to when we explain why our processes have to be complicated.

They are just being asked the same three questions at every door, and being required to answer with paper, in person, during working hours.

Which means the population we are actually rationing by is not the poor, or the sick, or the unhoused. It is people who cannot take a Tuesday morning off.

Where this lands hardest

I am not going to skip past who this falls on, because the pattern is documented and it is not random.

The person with the most doors to visit is generally the person with the fewest ways to reach them. And when communities finally looked at their own sorting data, they found the sorting was racial. A 2019 analysis by C4 Innovations across four Continuums of Care found that BIPOC clients scored lower on the standard prioritization tool than white clients, pushing them toward thinner help. Austin’s ECHO found the same locally and replaced the tool. In December 2020 OrgCode, which built the original, said it would stop investing in it, citing misuse and equity.

Then the other end. In a case brought by two Black women holding federal housing vouchers, a tenant screening algorithm was challenged for weighting credit history and non tenancy debts while ignoring the single most relevant fact about a voucher holder, which is that a housing authority pays most of the rent directly. The case settled in November 2024 for $2.275 million, and the company agreed to stop attaching that score to reports for voucher applicants in Massachusetts.

One of those women had sixteen years of landlord references saying she paid on time or early. Sixteen years of evidence, overridden by a score, with no appeal available.

Mike could not produce originals. She produced sixteen years of them. Neither one got through.

The lag tax

I have a name for this now.

The lag tax is what a person loses in the space between government services.

Not the wait inside one office. We measure that, sometimes. The space between all of them. The stretch where one agency is done with you and the next one has not started, and nobody on earth is responsible for the person standing there.

It gets paid in both battles. Getting in, it is what you lose while four agencies separately decide whether you have proved what you obviously are. Staying in, it is what you lose when one letter goes to an old address and the rest comes down behind it.

It gets paid in bus fare. In hours billed against a shift. In the fourth retelling of the worst year of your life to a stranger with a form, whose system the fifth stranger cannot see. In a propane tank in a cold apartment.

Nobody votes on it. It is in no budget. Mike paid it in full.

What it costs the rest of us

I have started making this argument with money, because the fairness argument has never once moved a budget I sat in.

Look at the Medicaid unwinding, which is the cleanest count this field will ever get. More than 25 million people were disenrolled. About sixty nine percent of those were procedural. Paperwork. Nobody determined they were ineligible. Do the multiplication and roughly 17 million people lost health coverage without a single eligibility decision being made about them.

We paid staff to remove them. They absorbed the loss. Then we paid staff to put a large share of them back. The only thing produced was the gap.

The unit cost is documented too. Urban Institute’s USDA funded work found that somewhere between seventeen and twenty eight percent of SNAP households churn in a year, and processing that reapplication runs an agency about eighty to a hundred thirty dollars a case. And here is the finding that should end the efficiency debate: a USDA study found broad based categorical eligibility cuts state administrative expense per case by around seven percent, by simplifying determinations and removing redundant verification.

Simplification is cheaper to administer. We pay a premium to make it hard.

Now put Mike’s eviction next to that. The deposit is gone. The unit is off line for repairs. The landlord will not take the next voucher. The outreach hours that found him the first time are spent. And the public cost of a person living outside, depending on which study you use, runs somewhere between six and twenty one thousand dollars a year more than the cost of keeping them housed.

Against a bill he needed forty dollars and a bus ride to pay.

We are protecting a hundred dollar transaction and writing off a five figure annual asset to do it. That comparison is mine, assembled across separate studies rather than lifted from one, and I will stand behind it in any room.

One more thing, and it is the detail that bothers me most. We already track returns to homelessness. Every Continuum reports them at six, twelve and twenty four months. What we never record is why. There is no box for benefit lapsed. No box for could not get to the office. Mike shows up in that data as a return with no cause, which means the gap never gets a price, which means nobody ever funds anyone to stand in it.

Mike was the only one who could see it

Nobody who built any part of Mike’s obstacle course ever walked it.

The person who wrote the in person payment policy never rode the bus to their own office. Whoever set the originals requirement never tried to obtain three original documents with no phone and no address. The transit planner and the benefits director never met. And I never walked it either, and I was standing closer to it than anybody else in this story. I sat at one station on the line and I mistook the station for the line.

That is what a fragmented system does. It has no observer.

Each of us sees our own piece and reports it as working, because from where we sit it is working. My eligibility determinations were correct. The utility’s payment policy was defensible. The bus route was reasonable given the budget. Nobody in the chain was looking at the chain.

So the only person who could see the whole thing was the one required to walk the whole thing.

Which means Mike was not only a man who needed help. He was the sole functioning diagnostic instrument in that entire arrangement. He ran the full route, end to end, under real conditions, and he came back with a finding. The finding was that it does not work.

He performed that diagnostic for free. He paid for it with a winter outside.

And we threw the result away.

Then we blamed him for the reading

Here is the mechanism, and it is a mechanism, not a mood.

When Mike stopped coming, something got written down. Not a note about the process. A note about Mike. Not engaged. Did not follow through. Declined services. Non compliant.

Every one of those phrases performs the same operation. It takes a defect in the system and re-files it as a trait of the person.

I have written those words. I want to be clear that I am not describing something other people do.

And here is what that does to the feedback loop, which is the part that keeps me up.

Our processes are built to screen for follow through. So when somebody stops following through, that does not arrive as a defect signal. It arrives as the screen working. Mike’s disappearance was received, quietly and by nobody in particular, as evidence that the filter did its job. The system got exactly the confirmation it was designed to look for.

That is why nothing changes. It is not that we ignore the signal. The signal reaches us already inverted. Our failure comes back to us labeled as a success at exclusion.

And it never gets corrected, because of that missing box I mentioned. Mike shows up in the data as an incomplete with no cause. No reason code. Nothing that would let anyone, ever, aggregate people like him into a number and put that number in front of a decision maker.

The one person who could see the whole system produced the only real evaluation it has ever received, and we have no field to record it in.

Why we have not worked feverishly to fix it

Now the question I owe you a straight answer to, because it is the one I would ask.

We do work feverishly. That is what I want to say first, because the lazy government story is wrong and everyone who has actually done this job knows it is wrong.

I have watched teams pull all nighters for a funding application. I have watched people give up weekends for the annual count, rebuild a dashboard in seventy two hours because one councilmember asked a question in one meeting, spend six weeks preparing for a monitoring visit that lasted two days. This field has enormous capacity for urgency and no shortage of people willing to spend themselves.

The urgency is simply all pointed at things that get scored.

Nobody has ever had to explain a Mike to anybody. There is no hearing where someone asks how many people started and stopped. No audit finding. No corrective action plan. No headline, no deadline, no consequence at all.

So the effort goes where the accountability is. That is not a character flaw in public servants. It is exactly what you would predict from competent people inside a system that prices one kind of error and leaves the other free.

Which brings me to why the pricing looks the way it does, and that has a date.

In 1981 the federal government began penalizing states whose food assistance overpayment error rates ran too high. Penalties eventually reached forty nine states. The system runs in roughly the same form today.

Watch what got a price and what did not. Paying the wrong person became measured, reportable and sanctionable, with a dollar figure attached to a state’s name. Failing to reach the right person became nothing. No metric. No report. No hearing.

Everything Mike ran into descends from that. Every verification step was added by a competent person defending a real audit. Nothing was ever removed, because removal has no defender.

And then there is us

I include myself completely.

In twenty three years I have never seen anyone disciplined for adding a requirement. I have seen people ended by an audit finding. So the safe move, every time, for every individual, is one more document. Thousands of safe individual choices produced Mike’s folder, and not one of the people who made them was being reckless.

Our performance measures count who we served. They do not count who walked out. Mike left no trace in any number I reported that year, which means that on paper, my year looked fine.

Nobody is paid to sit between agencies. There is no line item called integration, so the work happens when somebody does it for free on top of their job, and it dies when that person leaves.

And we say a thing is legally required when it is only customary. This is the one that gets me. Over and over I have been told something cannot be done, gone looking for the actual statute, and found a practice that hardened into a policy that hardened into folklore. Nobody had checked since before I started.

Under all of it sits the excuse: we inherited this. None of us wrote it.

I do not accept that anymore. Every renewal of a form, every training that teaches the workaround instead of fixing the cause, every year a requirement survives because nobody moved to strike it, is a decision. We are not the first authors. We are the current ones.

And then we ask why he does not just get a job

He had one.

Not a paying one. The job of rejoining. And I want to put the two side by side, because we say get a job as though the labor market were the harder door to walk through.

To get hired at a warehouse or a restaurant: you apply on a phone in twenty minutes. No original documents. No third party verification of where you slept last winter. No letter from a provider confirming your condition. If they want you, you start. Federal law gives you three business days after your first day to prove who you are. The verification happens after. And you get paid in two weeks.

To get the benefit that would keep you alive until that first check, you prove everything first, and then you wait.

Social Security’s own reporting from December 2025 put the average initial disability decision at about 193 days. Roughly two thirds of initial claims are denied. Reconsideration averages another 208 days. A hearing averages 263 days on top of that.

The private employer takes a chance on you and checks afterward. We check first, exhaustively, and take a chance on nobody.

Then look at the hours. Benefits offices answer phones from nine to three. That is when jobs are. So the process we require competes directly, hour for hour, with the work we demand. Make our appointments and you cannot hold the shift. Hold the shift and you miss the recertification.

We attached work requirements to programs whose own compliance schedule makes working harder.

And when somebody chooses the shift and loses the benefit, or chooses the benefit and loses the shift, we say they lack motivation.

Mike was not unmotivated. Mike was employed by us. Full time, unpaid, no title, no supervisor, no training, no equipment, no benefits, and a termination clause on every task. The hours conflicted with paid work. The commute was unaffordable. The performance standard was original documents he had no way to obtain.

He quit.

People quit bad jobs. It is the most ordinary thing a person can do, and we have built an entire vocabulary to describe it as a character defect when poor people do it.

That is what we have made rejoining into. A worse job than the one we are telling him to go get.

The belief underneath it

That question is not fringe, and it does not stay in comment sections. It goes viral, in two reliable forms.

One is the tender version. A father posts that his six year old asked why homeless people do not just get a job instead of asking for money, and reports that he told the child they could, but they choose not to. Hundreds of thousands of views. The other is the contemptuous version, which circulates just as widely: being homeless in a big city is a skill issue, you are surrounded by opportunity and free education on the internet, and instead you choose the bench.

Both replies run the same operation. They take a logistics problem and convert it into a moral one. And that conversion only works if you first believe the logistics are trivial.

Devon Price named this belief. In Laziness Does Not Exist, Price calls it the Laziness Lie, a widely held and mostly unspoken conviction that hard work is morally superior to rest and that people who produce less are worth less. Price identifies three assumptions holding it up: that a person’s worth is their productivity, that people cannot trust their own read on their limits, and that no matter what you have done there is always more you could be doing.

Now put Mike in the room and read those three back.

Your worth is your productivity. Mike produced nothing measurable that quarter, so he registered as worth nothing, and the full time unpaid compliance work he was actually doing did not count, because we have never counted it.

You cannot trust your own limits. When Mike said his body hurt and he could not make the trip, that got treated as something to be overcome rather than as information about the trip.

There is always more you could be doing. This is the one that should stop every practitioner cold.

Because that third assumption is not just how strangers talk about people like Mike online. It is the operating logic of our processes. Every requirement any of us has ever added rests on the belief that there is more the applicant could be doing. That is the whole justification for the ninth document.

And it creates a trap that nothing can escape. If the person can always do more, then no amount of failure ever counts as evidence about the design. The belief is unfalsifiable. A man walking out is never a finding about the process, because there was always one more thing he could have tried.

That is the same inversion I described a few sections up, wearing different clothes.

Which is why I am not going to let us stand at a distance from the guy posting skill issue. He and the caseworker typing not engaged are running the identical operation. One does it for likes and one does it in a case note that goes in a federal data system. Only one of them gets to call it documentation.

The market solved this in 1994

Short version, because the comparison is embarrassing enough without belaboring it.

The early web forgot you between pages. Load a page, click a link, and the same machine treated you as a stranger. Fine for documents. Fatal for selling anything, because you cannot build a shopping cart on a system with no memory. Every screen was a fresh intake.

In 1994 a young engineer at Netscape wrote a short specification for a small piece of data the browser hands back on every request. He called it a cookie. It shipped that October.

Four months, roughly, from problem to fix. Not because anyone was noble. Because forgetting a customer cost money and somebody counted it.

Then they kept going. Saved payment. Autofill. One credential that opens fifty doors. One click ordering. A delivery driver who finds a tent with a dropped pin and a text message, because nobody made the customer reconcile an address that does not exist.

All of it is one idea. Do not make the person prove who they are more than once.

And they measure the ones they lose. Roughly seventy percent of online carts get abandoned, and the industry knows the reasons to the percentage point, because there are people paid full time to find the step where someone gave up and delete it.

I could not tell you what percentage of people who start an intake in my system never finish. Neither could you. The data is sitting in HMIS. Nobody ever asked, because nobody’s revenue falls when a person disappears.

Front Door

I got tired of describing this in the abstract, so I built a rough version of the answer to argue with. It is called Front Door. It runs on demo data. I made it in Replit and it is not a product, not a procurement, and not a pitch.

It is a benefits passport the person holds. Share once, reuse where you choose, keep every next step yours. Two decisions in it I will defend.

It is person held rather than agency shared. That is the whole trick. Agency to agency data sharing runs straight into health privacy law, survivor confidentiality, and a stack of agreements nobody will sign. A consent the person grants and can revoke is a different legal animal. She can authorize a disclosure her caseworker is prohibited from making. We have been trying to solve this at the wrong layer for twenty years.

And the readiness indicator is not a score. It says sixty four percent ready to have a first conversation, and right beside it, not a decision, just a clearer starting line. That wording is doing real work. We already know what happens when a number attached to a poor person becomes a gate. It faces the participant, it tells her what is missing, and it gates nothing. If it ever starts deciding who gets served, it has become the thing it was built against.

Front Door, the proof of concept: attached-assets--staytackles14.replit.app

Break it. And tell me which rule you think it violates, because I want to find out whether the barrier is actually law or whether it is habit we have been calling law. The beautiful thing about working in the Bay Area? I mentioned it to a friend- and we’re ready to mobilize, but we need help, funding and support from those who want to help other human beings. More on that in the future.

Three things you can do Monday

Take originals off the list. Go find out whether your program is legally required to see an original document or whether somebody decided that in 1994. My bet is most of your list is habit. Accept copies, photos, and attestations wherever the rule allows, and put your name on the change so there is an answer when someone asks who authorized it.

Stop running renewals through the mail alone. Nothing terminates on a single unanswered letter. Text, call, email, and ask a partner agency before anything lapses, and give people more than ten days to respond. While you are in there, look at whether your renewal cycles can be aligned with each other instead of firing at random intervals. And before any appointment is required, check whether the information already exists in your own building. If the trip is genuinely necessary, pay the fare, because we should fund what we require.

Run your own QA discipline in the other direction. You already treat a repeated staff error as a systemic error requiring root cause analysis and a corrective action plan. Apply the identical standard to applicants. Pick any step in your process and count how many people fail it. If it is more than a handful, that is a systemic error by your own definition, and it gets a corrective action plan with a name attached, not a note in somebody’s file.

Start by adding a cause field to your returns and your incompletes, with benefit lapsed and could not get here as named options. Within a year you will have a number no jurisdiction in this country currently has, and that number is what gets you budget. It also stops the inversion. Right now every person who quits reads as proof the screen is working. A reason code turns them back into what they actually are, which is the only evaluation your system has ever received.

We can do better

I have thought about Mike for nearly near twenty years. Not constantly. He surfaces.

He surfaces when somebody in a meeting says a client was not engaged, which is the sentence I now hear as an alarm. He surfaces every time I hear somebody ask why people do not just get a job. He surfaces when I catch myself about to require one more thing because it would make the file cleaner. He surfaced the entire time I was building Front Door.

Here is what I would say to him if he walked into my office tomorrow, and what I could not say twenty years ago.

You were not the problem. The process was not designed for you and nobody ever checked whether it worked for anyone standing where you were standing. You did not fail to follow through. You correctly concluded that we were asking more than we were offering, and you spent your remaining energy on something with better odds.

And nobody told you the truth up front, which is that getting in was only half of it. We celebrate the lease. We hand people keys and take a photograph. Then we walk away and leave them holding eight fragile links they have to keep intact, forever, by mail, with no mailbox.

We can do better than a fax machine and a request for originals. We have been able to do better for thirty years. We have simply never been charged for not doing it.

I am not asking anyone to fix the federal government. I am asking you to find one requirement in your own program that exists because somebody once got nervous, and remove it this month.

Somebody is standing in your lobby right now doing Mike’s arithmetic.

If you have ever had to tell somebody you were not allowed to help them with the next step, tell me about it in the comments. I want that on the record. And if you are the one who removed a requirement and nothing bad happened, I especially want to hear from you.

Companion reads

• “The Virtue Exam” on conditionality, and what we require before the door opens.

• “The Economy of Poverty” on the compliance economy, and how poverty is maintained rather than failed at.

Sources

• KFF, Medicaid Enrollment and Unwinding Tracker: over 25 million people disenrolled; roughly 69 percent of disenrollments were procedural rather than a determination of ineligibility; approximately 41 percent later re-enrolled. The figure of about 17 million is the author’s multiplication of those two published numbers.

• Urban Institute (USDA funded), SNAP churn study: 17 to 28 percent of households experienced churn; roughly $80 to $130 in administrative cost per churned case. Summarized by CBPP and the Food Research and Action Center.

• USDA study finding broad based categorical eligibility reduces state administrative expense per case by about 7 percent, cited in FRAC analysis.

• Financial penalties for excessive food assistance overpayment error rates beginning in 1981, eventually levied against 49 states. SNAP quality control framework, 7 U.S.C. 2025(c).

• GAO, Payment Integrity: Agencies’ Estimated Improper Payments Increased to $186 Billion in Fiscal Year 2025, April 2026.

• Cost of homelessness comparisons: Urban Institute evaluation of the Denver Supportive Housing Social Impact Bond (approximately $15,680 difference in public spending, roughly 67 percent offset of program costs); Basu, Kee, Buchanan and Sadowski, comparative cost analysis for chronically ill homeless adults (approximately $6,307 per person per year, approximately $9,809 for the chronically homeless subgroup); Morrison Institute for Public Policy, Maricopa County analysis (approximately $73,000 per year homeless against approximately $52,000 in permanent supportive housing). The six to twenty one thousand dollar range is the author’s summary across these studies.

• HUD System Performance Measure 2, returns to homelessness at 6, 12, and 24 months after exit to a permanent housing destination.

• HUD chronic homelessness recordkeeping requirements, 24 CFR Part 578 and Notice CPD-14-012: order of priority for evidence, the 75 percent third party documentation threshold, and the 25 percent limit on full self certification.

• Wilkey, Donegan, Yampolskaya and Cannon (2019), Coordinated Entry Systems: Racial Equity Analysis of Assessment Data, C4 Innovations with Building Changes. ECHO (Austin/Travis County) Racial Disparities Report. OrgCode Consulting statement on the VI-SPDAT, December 2020.

• Louis, et al. v. SafeRent Solutions, LLC, et al., No. 1:22-cv-10800 (D. Mass.), final settlement approval November 20, 2024.

• HTTP cookie implemented in Netscape Navigator 0.9 beta, released October 13, 1994. Baymard Institute cart abandonment research.

• Social Security Administration, average processing time reporting (December 2025): approximately 193 days for an initial disability determination, approximately 208 days for reconsideration, approximately 263 days for a hearing. Roughly two thirds of initial claims denied at the initial level. SSA publishes this data at ssa.gov/data.

• Employment Eligibility Verification (Form I-9): employee completes Section 1 no later than the first day of employment; employer completes Section 2 within three business days of the first day of work.

• Price, D., Laziness Does Not Exist (2021), on the Laziness Lie and its three underlying assumptions. Paraphrased here.

• Herd, P. and Moynihan, D., Administrative Burden: Policymaking by Other Means.

No Rent, No Lease

Five housing decisions that live free in my mind, and after this, in yours too

There is a phrase people use online when something refuses to leave them alone. That lives rent free in my head.

I have five of those. They are not songs or bad breakups. They are housing policy decisions, most of them older than I am, all of them still collecting. No rent. No lease. No move-out date.

Twenty-three years in this work has taught me that a housing crisis is rarely an accident and almost never a surprise. It is a decision, made once, by people whose names have fallen out of the story, and then left running. What we call a crisis is usually an old choice still doing its job.

I have spent a long time arguing that the absence of a thing is the physical record of a decision. A neighborhood without sidewalks is not an oversight. It is a budget from 1961 that nobody reopened. Housing works the same way. The units we do not have were, at some point, specifically and deliberately not built.

Here are five doors that were closed, then locked, and nobody came back with the key.

1. Cannot grow. (1937, and again in 1995.)

The United States Housing Act of 1937 created public housing in this country. It also contained a provision called equivalent elimination. For every unit of public housing built, roughly one unit of existing “unsafe or unsanitary” housing had to be destroyed.

Net zero. By statute. On day one.

That was the price the real estate lobby charged for letting the program exist at all. Build all you want, so long as the total does not move.

Imagine agreeing to open a hospital only on the condition that you close a hospital. Then, thirty years later, standing in a crowded emergency room and concluding that public health does not work.

Now sit with the bookend, because this is the part almost nobody puts together.

For decades, federal law also ran the rule in reverse: a housing authority that demolished a public housing unit had to replace it one for one. In 1995, Congress suspended that requirement. By 1996 it was gone.

Put those two sentences side by side. In 1937 we required demolition in order to build. In 1995 we stopped requiring building in order to demolish.

What followed was predictable and enormous. Roughly two hundred thousand public housing units came down between the mid-1990s and 2010. About a quarter of them were replaced with new public housing. The rest became vouchers, or became nothing.

The symmetry is almost too neat to be real. We wrote the ratchet in one direction, then removed the catch in the other.

2. Cannot share. (1954 to 1974.)

There used to be an entire category of American shelter that has nearly vanished from memory. The boarding house. The rooming house. The residential hotel. The room over the shop. A single room with a bath down the hall.

It was not charming. It was cheap, and it was legal, and it was everywhere.

Then zoning codes across the country began defining who counted as a “family,” typically capping the number of unrelated adults permitted to live under one roof. In 1974, in Village of Belle Terre v. Boraas, the Supreme Court upheld exactly that. A town could tell unrelated adults they were not allowed to share a house.

Meanwhile urban renewal and downtown redevelopment took the residential hotels. Cities lost single room occupancy stock by the tens of thousands, and in the largest markets by six figures.

Understand what that housing actually did. It served the low-wage worker, the new arrival, the older man living alone, the person between things. It was affordable precisely because it did not come bundled with a kitchen, a car, a credit check, or a family.

We never replaced it. We criminalized the shape of it and waited.

Today, when someone sleeps on a cousin’s couch, splits a weekly motel room four ways, or rents a bed in a converted garage, we file them under doubled up, precariously housed, shadow inventory. We turned a category of housing into a category of failure. People did not invent the housing underground. They rebuilt it from memory after we tore it down.

3. Cannot build. (1973.)

In January 1973, the Nixon administration announced a moratorium on federal subsidized housing production. It was framed as a pause. Eighteen months, while the government studied what worked.

The study concluded that subsidizing demand was more efficient than producing supply. The Housing and Community Development Act of 1974 gave us Section 8. We stopped being a country that builds housing and became a country that rents it from somebody else.

That pause is now fifty-three years old.

But the decision inside the decision is the one that keeps me up. We did not make rental assistance an entitlement.

Qualify for SNAP, you get SNAP. Qualify for Medicaid, you get Medicaid. Qualify for federal rental assistance, and you get a waiting list. Roughly one in four eligible households actually receives help.

That is arguably the most consequential fact in American housing policy, and it was never debated as a matter of principle on any national stage. It was set as a budget line and renewed by inertia every year since.

Picture a fire department funded to respond to one fire in four, then evaluated on the condition of the houses it reached.

4. Cannot fit. (The 1970s, in a room with no reporters in it.)

This is my favorite, because almost nobody knows it exists and nobody ever voted on it.

In most of the United States, a residential building taller than three stories must have two separate exit stairways. That rule is not in a statute anyone campaigned on. It lives inside a model building code written by a private organization, adopted by reference into state and local law, and revised in technical committee meetings that draw no press and no public.

Two staircases sounds like plain common sense. Watch what it does to a building.

Two stairs require a corridor to connect them. A corridor with apartments on both sides requires depth. A building with that depth requires a wide lot. A wide lot in an already-built neighborhood requires assembling multiple parcels, which requires lawyers, holdouts, patience, and capital.

So the small infill apartment building, the six-unit walk-up that fits on a single city lot next door to a house, becomes impossible to draw. Not illegal. Undrawable.

We did not ban the small apartment building. We wrote a rule that no small apartment building could satisfy, which is a far more durable way to kill something.

And look at what survives. Apartments with windows on one side only. No cross ventilation. Dark interiors. Floor plans that fight every attempt to fit a third bedroom. We made family-sized apartments structurally expensive to build, then spent forty years wondering aloud why families with children cannot find anywhere to live.

Now the part that should stop you.

Seattle has allowed single-stairway buildings of four to six stories since the 1970s. New York has permitted versions of it far longer. Honolulu copied Seattle’s code in 2012. Much of Europe has built this way for a century.

In 2025, researchers at Pew and the Center for Building in North America went through twelve years of fire data. Fire death rates in modern single-stairway four-to-six story buildings in New York were comparable to other residential buildings. And not one of the fire fatalities they reviewed in those buildings, in New York or Seattle, would have been prevented by a second stairway.

We eliminated an entire category of American housing on a safety rationale the safety data does not support.

Not out of malice. Out of a committee decision that hardened into consensus, then into law, and finally into invisibility, which is the most powerful form a policy can take. You cannot argue with a rule nobody remembers making.

There is a coda here, and it starts hopeful. Since 2022, nineteen states and the District of Columbia have introduced bills to study or allow single-stairway construction. Seven states passed reform legislation in 2025 alone. California, Oregon, Washington, Colorado, and Tennessee have all moved. Seattle has done this for fifty years. Honolulu copied Seattle in 2012.

Now let me take the hope back, because I have sat in the rooms where this part happens.

Almost none of it has been done at scale, and the reasons are structural rather than accidental.

Passing a state law is not the same as changing a building code. In most states, jurisdictions that write their own codes still have to adopt the reform locally, one city council at a time. Colorado’s 2025 law applies only to municipalities above 100,000 residents, and gives them until December 2027 to comply. Tennessee passed its law in 2024, and it took a further round of local adoption in Nashville, Memphis, Knoxville, Chattanooga, and Jackson before a single building changed. Most reforms come with caps on units per floor and limits on floor area that shrink the envelope again.

And here is the piece that almost never gets said out loud.

Legalizing a building type does nothing if the zoning on the lot still forbids the building. Code reform without zoning reform is a permission slip to construct something you are still not allowed to put there. The staircase was one lock. It was never the only one.

So the reform lands, and then it lands softly, in the places where the small building was already allowed, and stops at the edge of the neighborhoods where it would actually change who lives nearby. That boundary is not drawn by accident. It is negotiated in planning and zoning commission meetings, on a weeknight, in front of the small and remarkably consistent group of people who show up to those meetings. They are almost always homeowners. They are almost always from the neighborhoods in question. And the thing they are protecting is rarely the staircase.

I have watched a good reform get narrowed into a gesture in real time, in a public hearing, with my own department’s name on the staff report. The vote is usually unanimous. Nobody in the room says anything about who is being kept out. They do not have to. The exemptions do it for them.

It took us half a century to ask the question out loud. We are still deciding how quietly to answer it.

5. Cannot ever. (1998.)

The Faircloth Amendment, passed as part of the 1998 public housing overhaul, caps the number of public housing units a housing authority may operate at the number it had on October 1, 1999.

Read that again slowly. There is an American public good with a statutory maximum, pinned to a date most of the country was not paying attention to.

We do not cap highway miles at 1999 levels. Not hospital beds. Not school seats. Not broadband. Not military bases. Housing alone carries a ceiling written into federal law.

Now, before anyone in the field corrects me in the comments, let me correct myself first.

There was a door in this wall. In 2021, HUD created what everyone called Faircloth-to-RAD, later renamed Restore-Rebuild. Housing authorities that had lost units over the years still held the legal authority to rebuild up to their 1999 number. That unused authority is enormous. HUD counted nearly 260,000 units nationwide that could theoretically be built. Chicago alone holds roughly 19,000. New York holds close to 11,000. Philadelphia around 7,000.

On paper, that is a quarter million deeply affordable homes sitting on the shelf, already authorized, waiting.

In practice, almost none of them got built, and the reason is the most ordinary thing in government. HUD gave permission, not money.

You still had to finance the whole development yourself, and the subsidy attached to those units was the problem. The Terner Center found that the contract rents under this program run too low to support debt on new construction, lower than a standard project-based voucher, because HUD’s formulas assume a new building costs less to operate than an old one. So the deal that the authority is allowed to do is the deal the deal cannot pay for. You have the right to build and no way to close.

Then, in May 2026, HUD sent a memo to every housing authority in the country winding the program down. No new requests accepted, effective immediately. Deals already in the pipeline got ninety days to prove they were far enough along to survive. The stated reason was to focus resources on the existing portfolio and reduce the public housing footprint.

And the FY2027 budget request goes further. It proposes resetting every authority’s Faircloth cap to whatever inventory it holds as of October 1, 2027. If that passes, the shelf does not just stay out of reach. The shelf gets taken down.

That sequence is the whole essay in miniature. A ceiling built in 1998. A narrow door opened in 2021. A door that never functioned because the money was never behind it. The door closed in 2026. And a proposal to move the ceiling down to meet the floor.

1937 said the supply cannot grow. 1998 said, and it never will. 2026 said, and we are going to make sure of it.

What they have in common

Not one of these five decisions was about homelessness.

They were about lobbies, tax appetite, property values, downtown land, liability, and who was allowed to count as a family. Homelessness was the residue. It is the byproduct of decisions made in rooms where nobody unhoused was present, and in most cases nobody unhoused had yet been born.

And every one of them was closed out. Not argued and lost in the open. Closed. In a section number. In a footnote. In an amendment attached to a larger bill nobody read past the summary. In a technical committee with an attendance sheet. In an eighteen-month pause that turned fifty-three.

Stack them and you get a ladder:

Cannot grow. Cannot share. Cannot build. Cannot fit. Cannot ever.

That is not a housing market. That is a design.

If you work the front lines, hold this carefully, because it cuts both ways.

When you cannot find a unit for the family sitting in front of you, you are not failing. You are standing at the end of a chain of decisions that began before your parents were born, holding a caseload that was structurally engineered to exceed you. That is not an excuse. It is orientation. You cannot put down what you have never been allowed to name.

Who got left holding it

Here is what happens when a federal government stops building.

The need does not stop. It relocates. It lands on cities, counties, housing authorities, nonprofits, and the private market, which is to say it lands on entities that never had the balance sheet, the land, or the mandate to carry it. Every one of the five decisions above was made in Washington. Every consequence gets managed at a folding table in a city hall.

I say that as a man sitting at the folding table. My department did not make any of these decisions and cannot undo them, and I still have to stand up at the community meeting and answer for the result. I am not complaining. This is the actual shape of the work, and the shape was designed.

Which brings me to the thing I most want you to take from this.

There is a real and welcome conversation happening right now about social housing. Publicly owned, mixed income, permanently affordable, taken off the speculative market. I am for it. But I keep noticing that we discuss it as though it were a new idea we imported from Vienna.

It is not new. We had it. It is called a housing authority.

Housing authorities are the entities that built at scale in this country. They hold land. They hold bonding capacity. Many of them still hold the unused legal authority we just spent a section talking about.

We did not lose that institution. We capped it, defunded it, demolished a third of its stock, and taught two generations to say its name like an accusation.

And on tools, let me be fair to the one I cut from this list. The Low Income Housing Tax Credit is the workhorse of American affordable housing and we should defend it. But a tax credit is an incentive, and an incentive can only ever produce as much housing as the incentive is funded to pay for. It was never sized to the gap. It was not designed to be. Nobody serious in this field believes we get out of this on tax credits alone.

So it has to be all of it, at once. Tax credits and public development and land trusts and zoning reform and code reform and preservation and vouchers that actually reach the people who already qualify for them. Not sequenced. Not pitted against each other across a budget hearing. Hand in glove.

And one more thing, because I do not want to be misread on it.

Housing is not the goal. Housing is the floor.

I am careful with that sentence, because it gets stolen constantly by people who use it to argue for services instead of homes. That is not what I mean. You cannot build dignity for a person who is sleeping outside, and anyone who tells you otherwise is asking you to fund a program instead of a place to live.

But you also cannot stop at the unit and call the work finished. What these five decisions destroyed was never only buildings. It was blocks and corners and congregations and the boarding house where somebody knew your name when you came in at night. We demolished the housing and the neighborhood in the same motion, then spent fifty years measuring only one of them.

If we restore nothing but our ability to produce units, we will have rebuilt the machine without remembering what it was for.

Build the housing. Then build the rest of it.

Now they are yours

I am not asking you to memorize five dates. I am asking you to keep the habit.

When you see an absence, ask what decision made it. The missing sidewalk. The missing bus line. The missing three hundred units. The staircase that was never there and the building that therefore never was. Somebody chose. Somebody wrote it down. There is almost always a paper trail, and it is almost always more boring and more deliberate than we expect.

Because the other side of this was decided is the only hopeful sentence in housing policy: it can be decided again.

I want to be careful here, because I just spent five sections showing you that “decided again” cuts both ways. A memo in May 2026 closed a door that a memo in 2021 opened. I take that as evidence the machinery is still live, not as a reason to look away.

None of this is settled law handed down from the founding. Faircloth is a sentence in a bill, and sentences get struck. The definition of family in a zoning code is a definition, and definitions get rewritten, as a growing number of states have already proven. A staircase requirement is a line in a code book, and seven states rewrote that line last year, even if most of them then wrote the exemptions carefully enough to keep the buildings where they already were. Even the moratorium, the one that has now run for half a century, is technically still just a pause.

None of this is weather. All of it is architecture.

So take them. These five have been living in my head free of charge for years. It is time they had roommates.

No rent. No lease. Just do not let them sit quiet.

We didn’t stop criminalizing homelessness because we grew kind. We stopped because it got expensive. And expensive is a mood. It can change. Twenty years after Million Dollar Murray, an accounting of what the bargain actually cost, with the receipts attached.

We did not stop arresting people for the crime of being poor because we finally saw them as human.

We stopped because it got too expensive.

I want that to land before I soften it, because we have spent twenty years telling ourselves the prettier version. In the pretty version, the country looked at people sleeping in doorways and in the backs of cars and something in us turned. We saw the cruelty of jailing a person for having nowhere to go, and our conscience did the rest.

That is not what happened. What happened is that somebody did the math, and the math turned out to be more persuasive than mercy had ever managed to be.

Here is the part we leave out of the story. The retreat from criminalizing poverty was not led by activists with signs. It was led, starting in 2002, by a Republican administration and a former seminarian named Philip Mangano. Mangano traced his own calling to a film about Saint Francis, but he did not sell belief. He sold return on investment. As the head of the U.S. Interagency Council on Homelessness, under the banner of compassionate conservatism, he carried one case to every mayor and governor and county executive who would sit still for it. He did not tell them these were their neighbors. He told them the numbers. A man living on the street runs up a fortune in ambulances and emergency rooms and jail beds, and a key to an apartment costs less than a cell. The business writers took it up. The management thinkers wrote it into their books.

It worked because it was true. And it worked because it asked nothing of anyone’s heart. You did not have to love the man in the doorway. You only had to want your money back.

By 2004 the federal government had made Housing First, the idea that you give a person a home first and steady everything else from behind a door that locks, its official approach to chronic homelessness. In the decade that followed, chronic homelessness fell by roughly a third.

Let me be clear, because it matters and I will say it more than once. Housing First works. The evidence that it works is as strong as anything in this field. I have spent my career in it and I will defend it to anyone. What I am about to take apart is not the method. It is the story we told to sell the method.

When you win a person’s right to exist by proving it saves money, you have not won them a right. You have won them a line item. And a line item can be cut.

Efficiency has no loyalty. It belongs to whoever is holding the calculator. The same spreadsheet that says a home is cheaper than a jail will tell you, in a colder room in a worse year, that a sweep is cheaper than a shelter. That a one-way bus ticket is cheaper than a caseworker. That a downtown loses more to the sight of a tent than it would ever spend to house the person inside it. The math does not care which direction you point it. You built the instrument. Anyone can pick it up and turn it around.

In the summer of 2024, the Supreme Court turned it around. In City of Grants Pass v. Johnson, a six to three majority ruled that a city may punish people for sleeping outside even when it has given them nowhere else to sleep. It swept away Martin v. Boise, the ruling that had briefly held the obvious, that you cannot punish a person for doing the one thing a body with nowhere to go must do. If you do this work, you already knew the case was coming, and you already know the workarounds. You knew them long before the Court did. We have always found ways to move people along. Grants Pass just took the apology out of it.

It is the oldest trick we have. A hundred years ago Anatole France named it. The law, in its majestic equality, forbids the rich and the poor alike from sleeping under a bridge. A rule that touches everyone and only ever lands on one kind of person is not neutral. It is aimed. We have simply learned to aim it with a straight face.

Here is the thing I cannot get past.

We will give a poor person one bed in this country with no waitlist, no means test, no recertification, no proof of income, no questions at all. We will hand it over tonight, no matter how full the rest of the system already is.

It is a jail cell.

We have built a country where the single most reliable way to be guaranteed a bed is to be arrested. Every other door has a line, and a form, and a caseworker carrying ninety other names. The cell opens the second the cuffs go on. And then we call it mercy. Three hots and a cot. As if a cage were the best we could do for a person whose only offense was needing somewhere to be.

There is exactly one exception. The single time we cannot hand you that bed is when the cell is already full. So look at what fills it. Not violence. Not the theft that leaves someone hurt. We pack our cells with sleeping, with sitting, with loitering, with camping, with the long list of things that were not crimes until a council voted to call them crimes. A man lying down on a sidewalk is not a criminal in any sense the word was built to carry. He is a person who needs to lie down. We took the plain fact of a body in public with nowhere to put it, and we wrote it into the penal code. These are infractions. In a working system they would be a matter of policy, a citation at most, handled the way we handle an expired meter or a lawn grown too tall. We chose to make them crimes instead. And so the only thing that ever closes our one open door is the crowd of people we arrested for needing it to be open.

And it is, by a wide margin, the most expensive housing we buy.

In Central Florida, researchers followed a group of 107 chronically homeless people and found the public was spending about thirty-one thousand dollars a year on each of them, in arrests and ambulances and jail time and emergency rooms. Housing that same person, with a caseworker to keep them steady, would have run about ten thousand. In Osceola County alone, thirty-seven chronically homeless people were arrested about twelve hundred and fifty times over ten years, cycling in and out for the same small handful of things that are not, when you say them plainly, crimes. And the cell itself is staggering. New York City now spends more than half a million dollars a year to hold a single person on Rikers Island, a figure that has nearly quadrupled in a decade. Supportive housing costs a fraction of it, and a long line of studies finds it very nearly pays for itself.

We do not means-test the cage. We means-test the key. That is the whole sickness in one sentence.

Let me tell you about Million Dollar Murray.

That was the name a magazine gave him twenty years ago, and it stuck. Murray Barr was a big man who lived on the streets of Reno, and a police officer there got tired of scraping the same man off the same pavement, so he did something nobody had done. He added Murray up. The detox stays, the ambulances, the emergency rooms, the years of patching a man together and setting him back down exactly where he started. The number came to about a million dollars. It had cost the public a million dollars, the officer said, to not solve Murray. That is how he got the name.

Murray became famous. His story became the argument a whole generation of us carried into those budget rooms. He was the proof that it was cheaper to house people than to punish them.

But read his story again and you find the thing the movement chose to forget. Murray was rare. He was the far end of the curve, the profoundly sick and profoundly expensive tail, not the body of the distribution. Most people who lose their housing are nothing like Murray. They are homeless briefly, for ordinary and fixable reasons, and while it lasts they cost the public almost nothing. Which means the argument that won, the argument I carried, only ever ran on the expensive few. The savings were the whole case. It was never built to cover the mother doubled up with her kids on a cousin’s floor, or the man working two jobs and sleeping in his car. They generate no savings. They do not pencil. We did not just win on cost. We quietly agreed that the people worth housing were the ones expensive enough to be worth it.

And Murray, the man who made the case, never got housed.

He got sober in a program, was sent back out to manage on his own with no one beside him, relapsed the way people with no support and every reason to despair so often do, and died not long after, of internal bleeding, on the wrong side of a door we never opened for him. The man who taught this country that housing was the cheaper choice was never given a home. The argument that made his name saved everyone but him. We decided we could not afford to let Murray keep costing us. We never once decided we could not afford to let Murray die.

There is a question underneath all of this that the cost fight never answered. If the way out of homelessness is a home, why has the way out become so hard to reach?

Because we removed the bottom rung of the ladder on purpose.

For most of the twentieth century, a person at the edge could rent a single room. It was cheap and spare and sometimes grim, but it had a door that locked and an address that counted on a job application. We called it the single-room-occupancy hotel, and across the postwar decades of urban renewal and the redevelopments that followed, we demolished most of it. We tore down the cheapest housing in America and built nothing a poor person could afford in its place. Then we stood back and marveled at all the people on the sidewalk, as though they had wandered in from the weather. You cannot climb out of the mud when the first foothold above it has been dug away.

And for the person still trying, the machine does more than fail to help. It seals the exit.

Survive the sweep and you still walk away with a ticket. The ticket becomes a fine. The fine, unpaid because it was written to a person who by definition has no money, becomes a warrant. The warrant becomes a record. And the record is the thing that turns away the landlord, fails the background check, and kills the job that would have ended the homelessness in the first place. In 2015, the Department of Justice went to Ferguson, Missouri, and found a city that had quietly turned its police into a collection agency. Tickets written to revenue targets. Fines stacked on fines. Warrants issued over fees people could not pay. It fell, overwhelmingly, on Black residents, it was legal, and it worked exactly as built. Ferguson was not an outlier. Ferguson was a mirror.

And then we ask the person caught in that machine to plan their way out of it. Sendhil Mullainathan and Eldar Shafir have shown what scarcity does to a mind. When you do not have enough, of money, of time, of certainty, the lack itself takes up room. It captures attention and taxes the bandwidth you would otherwise spend on next week and next year. A person surviving on the street is not failing to plan because they are careless. They are running a mind that scarcity has already half-occupied. And then we hand them a court date and call the missed appearance a character flaw.

None of this is a desert that formed on its own. Deserts are what nature makes. This was built.

Richard Rothstein spent a career documenting how it was built. The American map of who lives where was not drawn by preference and it was not an accident. It was drawn by law. Redlining. Covenants written into deeds. A Federal Housing Administration that would not insure a loan in a Black neighborhood, and would not insure one in a white neighborhood that might let a Black family in. With statute and subsidy we made a country where a white family could turn a paycheck into equity and a Black family, doing everything the same, could not, and then we let three generations of compounding finish the work. Camping bans and sit-lie ordinances and fines and fees are that same machine in newer paint. Each one written in the language of everyone. Each one landing on the same people, in the same neighborhoods, with a consistency no accident could produce. I have started calling it the apartheid architecture of American poverty, because that is what it is.

W.E.B. Du Bois gave us the reason a system like this holds even when it costs the people inside it. He called it the psychological wage. In the Jim Crow South, poor white workers were paid partly in money and partly in public deference, in the daily assurance of being above someone. The wage was the arrangement itself. I think about that phrase every time a city spends more to sweep an encampment than it would spend to house the people in it. The sweep is not irrational. It pays a wage. It buys the comfort of not having to look, and the older, uglier comfort of knowing there is a floor beneath you, and that the floor is made of somebody else.

So let me tell you what actually works, because naming a broken thing is only worth the breath if you mean to replace it.

Houston did not argue its way out. It built its way out. I witnessed part of this. Starting in 2011 it pulled more than a hundred agencies into one coordinated system, put every unhoused person on one shared list, and housed people first. It has since housed more than thirty thousand people. Homelessness across the greater Houston area is down about sixty-three percent. Veteran homelessness was, for all practical purposes, ended by 2015.

Denver ran the experiment the skeptics keep demanding, and it ran it honestly, as a pay-for-success program with a randomized controlled trial attached. It took people who had cycled through its jails and its streets for years and it housed them, and the Urban Institute kept score against a control group over three years. Arrests fell by forty percent. Police contacts by a third. Unique jail stays by thirty percent, and total days in jail by nearly as much. Nights in shelters dropped forty percent. More than three quarters of the people were still housed three years on. And more than half the cost of the housing came back to the public in the expense it prevented.

Charlotte built a single Housing First building called Moore Place, eighty-five apartments, and over two years its tenants made hundreds fewer trips to the emergency room and spent more than a thousand fewer nights in jail, saving the county about two and a half million dollars.

None of this is theory. None of it is hope. It is a solved problem we keep deciding to leave unsolved. Of course, we would need to discuss scalability, cost narrative, and resources- but it’s there.

And the restoration does not stop at construction. It means refusing the whole architecture of the trap. End the fines and fees that turn a night outdoors into a permanent record, because a person cannot climb out of a hole while we keep charging admission to the ladder. Stop sending armed officers to problems that are not crimes. In Eugene, Oregon, a program called CAHOOTS has for decades sent a medic and a crisis worker, not a squad car, to the calls that are really about someone in distress. It handles a meaningful share of the city’s call volume, almost never needs police backup, and costs a fraction of what an arrest costs. Denver built its own version and found the same thing. And rebuild the bottom rung. The deeply affordable apartments, the supportive units, the single rooms we spent forty years tearing down. There is no version of this that does not, in the end, run through a door a poor person can actually reach.

Here is where I have changed my mind about my own argument.

For years I thought the fix was to stop making the cost case and start making the moral one. To finally win the argument we have never had the nerve to have, the one that says a person does not have to be cheap enough to deserve a place to sleep.

I still believe that argument. But I am done waiting for it to arrive on its own. We have been waiting on a moral awakening for a very long time in this country, and Murray died waiting.

So we cannot leave this to conscience alone. Conscience is slow, and people are outside tonight. We keep the cost argument. We keep every argument. We use the calculator and the conscience at the same time, and we stop pretending we were ever forced to choose between them.

But we do the one thing we have never done. We stop letting a person’s right to exist hang on whether the argument wins again this year. We build the floor so it holds no matter the mood of the room. We make it a right, not a line item, so it cannot be repossessed the next time the spreadsheet turns cold.

We know how to open the door. The numbers have been in for twenty years, since a cop in Reno first sat down and added a man up. What we are missing is not the evidence, and it is not the money, because we are already spending the money, thirty-one thousand dollars a year of it, to make the problem worse.

What we are missing is the decision. That we will house people because they are people. That we will use every tool we have to do it. And that we will not wait for the whole country to feel it before we begin.

Say that, and mean it, and the rest is only engineering.

And the engineering, we finished a long time ago.

This is the long version, with sources. A shorter cut runs on LinkedIn. Every figure here is listed below, because an argument this pointed should be able to survive a fact-check. If yours turns up something I got wrong, the comments are open. That is what they are for.

Sources and Further Reading

• Malcolm Gladwell, “Million-Dollar Murray,” The New Yorker, February 13, 2006.

• City of Grants Pass v. Johnson, 603 U.S. ___ (2024), decided June 28, 2024, a 6-3 decision authored by Justice Gorsuch, overruling the Ninth Circuit’s Martin v. Boise (2018).

• U.S. Interagency Council on Homelessness under Philip Mangano, and the George W. Bush administration’s initiative to end chronic homelessness, from 2002 onward. Housing First adopted as federal policy for chronic homelessness by 2004.

• HUD Annual Homeless Assessment Reports (AHAR). Chronic homelessness declined by roughly 30 percent between 2007 and 2015.

• Central Florida Commission on Homelessness and Rethink Homelessness, “The Cost of Long-Term Homelessness in Central Florida,” 2014, conducted by Creative Housing Solutions ($31,065 per person per year on the street versus $10,051 to house; the Osceola County arrest figures).

• Office of the New York City Comptroller, Department of Correction spending analysis. The full annual cost per incarcerated person rose from $144,176 in FY 2011 to $556,539 in FY 2021.

• Urban Institute with the Evaluation Center at the University of Colorado Denver, “Breaking the Homelessness-Jail Cycle with Housing First: Results from the Denver Supportive Housing Social Impact Bond Initiative,” a randomized controlled trial, 2016 to 2020.

• Coalition for the Homeless of Houston/Harris County, The Way Home (approximately 63 percent reduction in homelessness since 2011; veteran functional zero reached in 2015).

• Lori Thomas et al., University of North Carolina at Charlotte, Moore Place Permanent Supportive Housing Evaluation Study.

• Richard Rothstein, The Color of Law: A Forgotten History of How Our Government Segregated America, 2017.

• W.E.B. Du Bois, Black Reconstruction in America, 1935 (the “psychological wage”).

• Sendhil Mullainathan and Eldar Shafir, Scarcity: Why Having Too Little Means So Much, 2013.

• U.S. Department of Justice, Investigation of the Ferguson Police Department, 2015.

• CAHOOTS (White Bird Clinic, Eugene, Oregon) and the Denver STAR program, non-police crisis response.

• Anatole France, The Red Lily (Le Lys Rouge), 1894, for the law’s “majestic equality.”

There is a test in American life.

It asks whether you drink. Whether you use. Whether you are in treatment. Whether you have proven, to a stranger with a clipboard, that you are ready to live indoors.

Almost no one takes this test.

Not the homeowner. Not the business owner. Not the developer. Not the farmer, the franchisee, or the founder.

The test is administered to one group: poor or income restricted people asking for housing.

When I applied for my own apartment, no one administered it to me. I was asked whether I could pay. I was asked to prove it. I was handed a lease to examine. That was the whole screening.

I have spent more than two decades on the other side of the counter, from case manager to running a city office of homelessness solutions, working inside the system that gives the longer exam. This essay is about the distance between those two counters. I want to walk through the federal subsidy landscape piece by piece, because once you see who gets screened and who does not, the Housing First debate stops being a debate about housing. It becomes a debate about trust, and about who was ever meant to be trusted.

First, the honest objection

Before the evidence, the rebuttal that deserves an answer.

The market is not unconditional. Landlords pull credit. They check eviction history. Many run criminal background checks. So how can I claim the market runs on Housing First?

Because of what the screens measure, and what clears them.

Every market screen measures one thing: risk to the landlord’s money. Will you pay. Will you damage the unit. A credit check is not a character reference. It has no opinion on your drinking, your marriage, or your soul. It is an actuarial question about an asset.

And every market screen has a price. Weak credit becomes a larger deposit. A thin file becomes a cosigner or six months prepaid. Enough money, and the screening ends entirely. At a closing table, no one asks for the biography of your cash. Inheritance, crypto, a lawsuit settlement, a lucky night. The money’s history is irrelevant. Only its presence matters.

Now compare. The poor person’s money is the most audited money in America. Its source is documented, verified, and recertified on a schedule. And the conditions attached to their housing are not financial. They are moral. Sobriety. Treatment compliance. Program participation. Demonstrated readiness.

Here is the distinction the whole debate turns on: the market’s conditions are financial, and financial conditions can be purchased. The poverty system’s conditions are moral, and moral conditions cannot be purchased at any price. Nobody has ever bought their way past a sobriety requirement.

There is one market screen that looks moral: the criminal background check. Notice what has been happening to it. In 2016, HUD warned landlords that blanket criminal bans produce racially disparate outcomes that risk violating the Fair Housing Act. In 2024, as we will see below, the SBA repealed its own criminal history screen for business loans because the evidence showed it predicted nothing. The one virtue test the market runs is the one the federal government has been actively dismantling. The virtue tests aimed at poor tenants, meanwhile, keep being proposed.

Unconditional housing when you are rich is called the market. Unconditional housing when you are poor is called enabling.

Hold that sentence. The rest of this essay is its documentation.

Part One: The subsidies nobody calls subsidies

Start with the largest housing assistance program in America. It is not public housing. It is not Section 8.

It is the mortgage interest deduction.

Every year, homeowners deduct billions in mortgage interest from their taxable income. It is a direct public subsidy of private housing, worth tens of billions annually, flowing disproportionately to the top of the income ladder. It arrived in the tax code without a policy debate about worthiness, survived the 1986 tax reform that killed most other interest deductions, and has never once asked a recipient about their drinking.

No application. No assessment. No case manager. The subsidy arrives inside your tax return, invisible, automatic, and unconditioned.

Now the FHA.

The Federal Housing Administration was created in 1934 to insure home loans. That insurance is the whole product. The federal government absorbs the risk of your default so a private bank will lend to you. It is public risk underwriting private life, at massive scale, for ninety years.

The FHA has never required a sobriety letter. It has never asked whether your marriage is stable or whether you have completed counseling.

But here is what the historical record shows, and this is where the story turns.

The FHA did run a screening test. For its first three decades, the test was race.

The FHA’s own Underwriting Manual instructed appraisers to grade neighborhoods by racial composition and to treat the presence of Black residents as a risk to be avoided. Richard Rothstein documents this in The Color of Law: the federal government did not passively tolerate segregation in its flagship housing subsidy, it wrote segregation into the underwriting standards. Whole suburbs were built on federally insured loans that Black families were structurally barred from receiving.

So the one time the great American housing subsidy screened applicants for something other than ability to pay, it screened them for race.

Not behavior. Not sobriety. Not virtue.

Race.

Hold that. We will come back to it.

The VA home loan tells the same story from a different angle. The GI Bill backed millions of mortgages for returning veterans with no morality screen attached. It also, in practice, failed Black veterans at scale, because the loans ran through the same discriminatory banking and appraisal infrastructure the FHA had built. I wrote about this lineage in “It Was Never About the Money.” The pattern repeats: the subsidy itself is unconditional, and the sorting happens in who is allowed to reach it.

Part Two: Business capital, where the worthiness test was tried and repealed

Now leave housing and follow the money into business.

The Small Business Administration guarantees more than forty billion dollars in capital to small businesses every year. When you get an SBA-backed loan, the public is co-signing your ambition. If your restaurant fails, taxpayers absorb the loss.

For decades, the SBA did run a worthiness test. Its regulations required borrowers to demonstrate “good character,” and the agency defined good character almost entirely through criminal history. People on parole or probation were barred outright. Applications carried the box: have you ever been convicted.

Then something instructive happened.

In 2024, the SBA repealed it. The agency removed the parole and probation bans, struck the criminal history questions from its applications, and dropped “character” from its lending criteria. The stated reason was empirical: there was no evidence that borrowers with criminal records defaulted at higher rates. The worthiness test had never predicted anything. It had only excluded people.

Sit with what that means.

When a morality screen stood between capital and business owners, the federal government examined the evidence, found the screen useless, and dismantled it.

When a morality screen stands between housing and poor or income limited people, we call it accountability and fight to expand it.

Same government. Same public dollars. Same question of trust. Opposite conclusions, sorted by who is asking.

Then there is the Paycheck Protection Program, the purest case in the modern record.

Between 2020 and 2021, the federal government loaned roughly $786 billion to 11.3 million businesses. More than 96 percent of that value has been forgiven. Loans became grants at a scale with no precedent in American history.

The screening was so light that fraud estimates run as high as $64 billion. Money went to businesses that did not need it, to franchises of the largest corporations in the country, and in documented cases to people who did not have businesses at all.

Not one PPP application asked about substance use. Not one required a treatment plan. Not one demanded that the recipient prove readiness before receiving public money.

Now place that next to the person applying for a shelter bed, who completes a vulnerability assessment, produces documentation, attends appointments, navigates coordinated entry, and waits.

$786 billion moved with almost no questions. A single cot requires an interrogation.

Part Three: The exam was administered once. Here is what it found.

Defenders of conditions will say the comparison is unfair. Poor people, the argument goes, have higher rates of addiction, so screening is prudent stewardship of public money.

We do not have to speculate. The experiment was run.

In 2011, Florida required every applicant for cash assistance to pass a drug test. The state assumed it would catch widespread drug use and save money.

In four months of testing, 2.6 percent of applicants tested positive. The federal estimate for drug use among Floridians generally was over 8 percent.

Read that again. The people applying for welfare used drugs at roughly a third the rate of the general population.

The program did not save money. Because Florida had to reimburse everyone who passed, the state lost nearly $46,000, before counting more than $300,000 in legal fees defending the law. Other states found the same thing. Utah tested and found a 0.2 percent failure rate. Oklahoma found 1.2 percent.

In 2014, the Eleventh Circuit struck Florida’s law down as an unconstitutional suspicionless search. The lead plaintiff was Luis Lebron, a single father finishing his degree who refused the test on principle.

Luis Lebron was a Navy veteran.

The virtue exam, administered at scale, found no epidemic. It found a population poorer and more sober than the state that tested them, it cost more than it saved, and it collapsed in federal court on the body of a veteran. And versions of it are still proposed in statehouses every year, because finding drug use was never the point.

Part Four: What the test is actually for

Here is where I have to hold two truths at once, because my work does not allow me to choose between them.

The first truth: the conditional system was built for the poor, and in America, systems built for the poor sort by race in operation even when the paperwork is colorblind. The FHA proved this in its founding documents. The modern versions prove it in their outputs. Research on coordinated entry assessment tools has found that race-neutral instruments produce racially disparate scores, routing Black applicants away from the housing resources their circumstances warrant. The instrument never mentions race. The results do. I have called this pattern an apartheid system of governance: policy written in neutral language that reliably produces racially sorted outcomes, because the sorting is the design surviving the language.

The second truth: the net is widening. The scrutiny machine was aimed at some Americans and is now waiting for more of them. Every year, working families of every race move one rent increase, one medical bill, one layoff closer to their first vulnerability assessment. The people most confident the virtue exam is for someone else are often a single bad year from taking it.

Both of these are true simultaneously. The specificity and the solidarity. Anyone who asks you to pick one is selling something.

And this is why the test survives its own failures. It was never an evidence-based screening instrument. Florida proved it screens for nothing. The SBA proved worthiness tests predict nothing. The test persists because it performs a different function: it marks the boundary between subsidy we call investment and subsidy we call dependency. Money moving up the ladder is infrastructure. Money moving down is charity, and charity, we have decided, must hurt a little, must be earned, must come with a stranger and a clipboard and a question about your habits.

The exam is not measuring the applicant.

It is reassuring the rest of us.

Part Five: We already know what happens when the exam is waived

This is not a theoretical argument, because the country has already run the counter-experiment.

In 2009, the federal government decided veterans deserved housing without a worthiness gauntlet, and funded that decision seriously and continuously through HUD-VASH and Supportive Services for Veteran Families. Housing first, support alongside, sustained money behind it.

Since then, veteran homelessness has fallen 56 percent. The 2025 count found 32,495 veterans experiencing homelessness, the lowest number since the data began. In a year when nearly every other population increased, veterans were the only group still declining.

Same human beings. Same addiction rates, same trauma, same mental health profile as the broader unhoused population, arguably heavier on all three. The only variable that changed was the design, and the design worked for sixteen consecutive years across administrations of both parties.

The method is not in question. The method is proven. What varies is only whom we are willing to apply it to.

Part Six: The redesign

I run one of these systems, so I do not get to end at analysis. Here is what taking this record seriously requires.

The market-parity rule. Any requirement attached to subsidized housing that does not exist in market-rate housing should carry the burden of proof. If a landlord on the open market cannot demand it, a publicly funded program should have to demonstrate, with evidence, why it can. Florida’s evidence is now in the record. So is the SBA’s. The default flips: conditions must justify themselves, not tenants.

The lease is the accountability instrument. Every other renter in America is governed by a lease. Pay rent, don’t destroy the unit, don’t endanger your neighbors. It works. It scales. It does not require the tenant to become a different person before receiving keys. Supportive housing should run on the same instrument, enforced the same way, with support wrapped around it.

Support within reach, not at the gate. Treatment, case management, behavioral health, and employment services work best when they are offered inside stability rather than demanded as its price. That is not a values statement. It is what the veteran data shows and what every practitioner watching people stabilize after housing already knows.

Audit the instruments. Every assessment tool in a coordinated entry system should be tested for disparate racial outputs, published, and revised. If the SBA can repeal a screening standard because the evidence showed it predicted nothing, a homelessness system can do the same.

The market already runs on Housing First and has for a century, subsidized the entire way, from the mortgage interest deduction to the FHA guarantee to the forgiven business loan. The principle is not radical. The principle is American. We have simply been rationing it by income and, longer than we admit, by race.

The work is extending the rule we already live by, to the people we built the exception for.

Keys first. Accountability through the lease. Help within reach. And no exam at the door that the rest of us were never asked to pass.

Related reading

• Housing First Was Never the Experiment (the short version of this argument) on linkedIn, look up Cupid Alexander

• It Was Never About the Money (the GI Bill, FHA redlining, and the proof that political will exists when we decide a group deserves help) linkedIn: Cupid Alexander

• The Identity of Poverty (how poverty became an identity instead of an economic condition, and why the net is wider than we admit) LinkedIn: Cupid Alexander

Sources

• HUD, Annual Homeless Assessment Report (AHAR) to Congress, Part 1, 2024 and 2025 Point-in-Time counts; VA Homeless Programs PIT Count summary (56% decline since 2009; 32,495 veterans in 2025)

• Lebron v. Secretary, Florida Department of Children and Families, 990 F. Supp. 2d 1280 (M.D. Fla.), aff’d, 11th Circuit, December 2014; ACLU of Florida case summary (2.6% positive rate vs. 8.13% general population estimate)

• CLASP, “Few TANF Applicants Test Positive for Drug Use; Testing is Costly and Ineffective” (Florida net loss of approximately $46,000; Utah 0.2%; Oklahoma 1.2%)

• Florida Department of Children and Families legal fee disclosures (approximately $307,000)

• SBA, Final Rule: Criminal Justice Reviews for the SBA Business Loan Programs, Disaster Loan Programs, and Surety Bond Guaranty Program, 89 FR 34094 (effective May 30, 2024); SBA press release, May 1, 2024 ($40 billion in annual guaranteed capital); Collateral Consequences Resource Center analysis (no empirical link between criminal history and default)

• Pandemic Response Accountability Committee and SBA PPP data (approximately $786 billion loaned; over 96% of loan value forgiven; fraud estimates up to $64 billion)

• Richard Rothstein, The Color of Law (2017), on the FHA Underwriting Manual and federal redlining

• Research on coordinated entry assessment disparities, including studies of the VI-SPDAT’s racially disparate scoring outputs