What the Treatment First Toolkit Gets Right — and What It Cannot Prove
Common Ladder — Learn / Evidence. An analysis of the federal Best Practices Toolkit: Addressing Homelessness and Addiction through “Treatment First” (ONDCP, HHS and HUD, August 12, 2026), written for funders, CoC leadership, federal and state policy staff, and the engaged public. Source status verified as of August 17, 2026.
The toolkit gets real things right: warm handoffs, peer specialists in positions of authority, keeping families together, and measuring long-term outcomes instead of service volume. The field should say so plainly — and should also concede the toolkit’s sharpest observation, that much of what America built and called “Housing First” was housing without the services the evidence always included.
But its central claim — that Housing First failed — rests on a single unnamed study in a footnote, set against two decades of randomized trials the document never engages. Its replacement model arrives with no comparative outcome evidence at all; the toolkit’s own closing chapter concedes that evidence still has to be built.[1]
Its measurement architecture is the part that will actually operate. Twelve-month abstinence for every program, price per client per day, and self-sufficiency defined as unsubsidized housing and employment will reward programs for excluding the sickest people. That is not a side effect of those three measures. It is what they do in combination.
The toolkit arrived five days after a federal court vacated the FY2026 CoC funding notice on procedural grounds. It is the merits argument, published through a vehicle that requires no public comment. The place that argument gets settled is a comment docket that has not opened yet.
A person completes ninety days of treatment. Every group session attended, every drug screen clean. On the morning of discharge the plan lists their housing as “shelter waitlist.” Within a month they are back in an emergency department, and the note in the file records a relapse — as though the failure belonged to the person rather than to the gap they were discharged into.
That sequence — treatment completed, housing absent, relapse recorded — is the oldest failure pattern in this field, and any veteran outreach worker can describe it from memory. So when the White House Office of National Drug Control Policy, the Department of Health and Human Services, and HUD released a toolkit in August built around warm handoffs — the principle that no one should be discharged from one part of the system into a gap — the document deserved a careful read rather than a reflexive one.[1]
Read carefully, it contains a good deal worth agreeing with.
Start with what it gets right
Much of the toolkit’s operational instinct is sound.
Warm handoffs are not a new idea, and they are well evidenced. Staff or peers physically accompanying a person from crisis care to the next appointment is the intuition behind Critical Time Intervention, a model tested in randomized trials since the 1990s and among the best-evidenced practices in the field.[2] The toolkit is right to put it at the center.
Peer specialists “in influential positions” matches what programs across the country have learned the hard way: the person who has slept where you sleep can reach you when no one with a lanyard can.
Its chapter on families is right that shelters which separate parents from children compound trauma, and right that the children of the overdose crisis are a population this field has barely begun to count.
Its demand that programs measure long-term outcomes rather than service utilization — bed nights delivered, meals served — names a real discipline problem in the homelessness sector.
And beneath the policy argument sits an observation the field should stop deflecting: much of what Americans have experienced as “Housing First” is not the model the evidence tested. The research evaluated housing plus intensive services — assertive community treatment teams, real case management at real caseloads. What many communities built was the housing with a fraction of the services, because capital and services come from different budgets and the services budget never materialized.[3] A supportive housing building with one part-time case manager for sixty residents is not Housing First. It is an underfunded building wearing the name, and its failures have been teaching the public — unfairly, but understandably — that the model itself does not work.
If the toolkit stopped there, it would be a useful document. It does not stop there.
The claim the whole argument stands on
The toolkit’s foundation is the assertion that Housing First — the principle that housing comes first and treatment is offered rather than required — has failed. The evidence offered for this, in the introduction, is one sentence: “A recent large-scale study found that veterans in supported housing were consistently more likely to die from drug overdose than homeless veterans.” The citation is footnote 8. The study is not named in the text.[1]
The shape of that claim deserves attention, because everything else in the document rests on it.
Supported housing programs do not enroll people at random. They are built — deliberately, by federal design — to prioritize the most vulnerable: the sickest, the longest-homeless, the people with the most severe combinations of mental illness, addiction and medical fragility. A raw comparison between people in supported housing and homeless people in general is therefore not a comparison of treatment against no treatment. It is a comparison of the sickest people a system could find against everyone else.
A hospital’s intensive care unit has the highest death rate of any floor in the building. That is not evidence that intensive care kills people. It is evidence of who gets sent there.
Supported housing is prioritized by federal design for the highest-acuity people in a community. Any comparison between that population and the general homeless population measures who was selected into the program, not what the program did.
Whatever study footnote 8 names, there is a way to answer this question that does not suffer from that problem, and it has been done. When researchers randomly assigned more than two thousand people to Housing First or to treatment as usual — the largest such trial ever conducted, with mortality followed for close to a decade — survival did not differ between the groups.[4] Not lower. Not higher. A null.
That null cuts in both directions, and the honest reading matters. For years, advocates and researchers argued that Housing First saves lives, and argued it partly on mortality grounds. The randomized evidence does not support that claim, and the materials that carried it have been corrected.[5] The toolkit takes the same weak comparison the field should have abandoned and runs it in the opposite direction. The field corrected its overclaim. The toolkit built its foundation on the mirror image of it.
What the evidence actually shows — all of it
Here is what the randomized trials show, stated plainly.
Housing First produces housing. In the At Home/Chez Soi trial, participants assigned to Housing First spent 73 percent of their time stably housed against 32 percent for treatment as usual — a gap that held across seven years of follow-up.[6] Run that difference across two years of a life and it is the difference between mostly home and mostly not. A second randomized trial, in Santa Clara County, found 86 percent of Housing First participants housed during the study period against 36 percent of controls.[7] Across the literature, housing-first approaches retain 80 to 90 percent of tenants at twelve months; treatment-first and readiness models have historically retained 30 to 50 percent.[8] On the outcome of housing, this is not a close call. It is one of the largest and most replicated effects in social policy.
| Evidence | Housing First | Comparison | Outcome measured |
|---|---|---|---|
| At Home/Chez Soi RCT (5 cities) | 73% | 32% | Share of follow-up time stably housed |
| Santa Clara Project Welcome Home RCT | 86% | 36% | Housed during study period |
| Replication literature | 80–90% | 30–50% | Tenant retention at 12 months |
| At Home/Chez Soi mortality analysis | HR 0.83 (95% CI 0.43–1.22) | Survival — no difference detected | |
| Substance use and mental health | Inconsistent, in both directions | Clinical recovery — neither arm superior | |
Housing First does not reliably produce sobriety, and it does not reliably produce mental health recovery.[9] The trials are consistent on this too, and the field should say it without flinching. But here is the detail the toolkit never engages: neither did the comparison groups. The treatment-as-usual arms — which included exactly the abstinence-contingent, readiness-based programming the toolkit now proposes — produced no better sobriety outcomes. The honest reading of the whole literature is that housing status is what housing programs change, and clinical recovery requires clinical treatment. Which is precisely why the evidence-based conclusion is housing plus treatment — not housing after treatment, with the housing held hostage to the treatment’s success.
The toolkit inverts that sequence and adds a mechanism that should concern anyone who has carried a caseload. Its housing is phased — shelter, then recovery residence, then independent living — with each phase conditioned on recovery progress, and it directs that when a person relapses they be moved out of their housing to “a different level of care” before earning their way back.[1]
Read that as a person rather than as a policy. Under this design, relapse — which the toolkit itself, in its more clinical passages, correctly calls an expected feature of a long-term, nonlinear process — costs you your bed. The model’s own premises predict its churn. And the field does not have to speculate about how this performs, because it ran this experiment for thirty years. It was called transitional housing. It exited half to a third of the people it served to somewhere other than permanent housing, at a higher cost than the alternatives, and the people it lost were disproportionately the highest-need people it was built for.[10]
There is also now a piece of modeling evidence aimed at this exact question. A 2025 microsimulation in JAMA Network Open compared housing with a treatment requirement, housing without one, and encampment sweeps, for opioid-using homeless adults. Housing without the treatment condition produced both the fewest deaths and the most time in medication treatment — more treatment engagement, not less, than the version that mandated it.[11] It is a model rather than a trial and should be held accordingly. But it points the same direction as everything else: conditioning the housing does not purchase recovery. It purchases exits.
The measurement machine
Now the layer beneath the rhetoric, which is where this document will actually operate. A toolkit does not house anyone. The metrics inside it decide which programs get funded, which people those programs go looking for, and which people quietly stop being anyone’s job to find.
The toolkit directs that programs be measured on, among other things: abstinence from alcohol and drugs at twelve months; price per client per day; and self-sufficiency, defined as market-rate housing and employment.[1] Each sounds reasonable stated at a podium. Consider what they do in combination.
No housing intervention on record reliably produces twelve-month abstinence — not Housing First, and not the abstinence-contingent alternatives.[9] So a program graded on abstinence has exactly one dependable way to hit its number: enroll people likely to stay abstinent, and exit the ones who are not. Grade it simultaneously on price per day, and congregate programs serving the easiest clients will beat any program serving the hardest, every time, on paper. Add self-sufficiency — unsubsidized housing and employment — as the success condition for a population whose defining characteristic is disability, in a country where the disability income floor sits below fair-market rent for a one-bedroom in essentially every metropolitan market, and the machine is complete.[12]
| Measure as written | Sounds like | Rewards in practice |
|---|---|---|
| Abstinence at 12 months | Recovery outcomes | Enrolling people likely to stay abstinent; exiting those who are not |
| Price per client per day | Cost discipline | Congregate settings serving the lowest-acuity clients |
| Self-sufficiency — unsubsidized housing and employment | Independence | Screening out disabled and long-term-homeless applicants |
| All three together | Accountability | Systematic exclusion of the highest-need people |
Consider what a program director does on the morning those three measures arrive stapled to her funding. She does not become crueler. She becomes careful about who she lets in the door. Under these measures, the sickest, the most addicted and the least employable become a liability on a spreadsheet.
This field has built systems before that quietly taught providers to avoid the hardest people, and spent two decades dismantling the incentives that did it. This document rebuilds them and calls it accountability.
Two questions the toolkit should be made to answer on the record
Does a person taking buprenorphine or methadone count as abstinent at twelve months? These are the FDA-approved medications that constitute the standard of care for opioid use disorder — medications the toolkit elsewhere accepts.[13] The document tracks “abstinence from impairing, addictive drugs” without answering the question, and a search of the published toolkit does not resolve it.
If the answer is no, the federal government’s flagship homelessness metric will penalize programs for delivering the single best-evidenced treatment for the deadliest addiction in America. That deserves a one-word answer, in writing.
The toolkit’s outcome list quietly includes “housing stability vs. days unhoused or returns to homelessness.” That is the right metric — the one the evidence supports and the one high-performing systems already track at twelve and twenty-four months. If HUD means it, then every program the toolkit celebrates should be published against it, next to every Housing First program it condemns.
A department that has not released the nation’s annual homelessness assessment in three cycles is not well positioned to lecture the field on transparent outcomes while its own exemplar programs report none.[14] Run the comparison, publish it, and the field will follow the results. That is what taking evidence seriously means, and it binds both directions.
Why the timing is the tell
One more layer, and it is the one practitioners need to hold onto this fall.
On August 7, 2026, a federal court vacated the FY2026 Continuum of Care funding notice — the multi-billion-dollar competition that had rebuilt itself around this same treatment-first pivot — in its entirety. The ruling was procedural: HUD had skipped the public notice-and-comment process the McKinney-Vento Act requires before conditions like these attach to federal funds. The court expressly declined to reach the Housing First question, denied a permanent injunction, and wrote that HUD “may attempt to issue yet another NOFO that contains these conditions after undergoing the requisite notice-and-comment process.”[15]
The toolkit appeared five days later. A toolkit requires no notice and no comment.
That sequence should not be read as incidental. This document is the merits case the court never heard — the administration’s evidence argument, published through a channel the Administrative Procedure Act does not reach, and positioned to become the scaffolding for the next funding notice: the one that will go through comment.
Which means the comment period, whenever it opens, is where this argument gets settled. As of this writing HUD has neither reissued the notice nor opened a docket; its competition page states only that the agency is evaluating the court’s order and considering its legal options, including appeal.[15] When that docket opens, the field will need to arrive with more than objection. It will need the trials, the retention data, the returns-to-homelessness rates and the two questions above, stated for the record in language a reviewing court can quote.
What follows, by audience
If you lead a Continuum of Care: know your returns-to-homelessness numbers cold, by program type, at twelve and twenty-four months. If your system does not track returns, that is the first task — a system that counts only exits is measuring the easy half of the outcome. Those numbers are the strongest evidence you own, and the strongest audit of your own weak programs besides.
If you fund this work: the credible response to “Housing First failed” is not denial. It is fidelity. Fund the services layer that the evidence always included and that the buildings never got, and stop capitalizing housing whose case management exists only in the grant narrative.
If you work in policy: put the buprenorphine question and the head-to-head outcomes question in front of every official who touches this document, and keep them there until one of them answers in writing.
And if you are simply someone who wants this problem solved: hold onto the thing the shouting obscures. The toolkit and its critics agree that people in addiction need treatment, that handoffs should be warm, that families belong together, and that outcomes should be measured in years. The disagreement is narrower and sharper than the press releases suggest. It is whether housing is the reward a person earns by recovering, or the ground a person stands on while they try.
The evidence answered that question years ago. Evidence was never really the obstacle. The obstacle is a story that keeps being retold — that the person on the sidewalk must become someone else before they deserve a door that locks. The programs that succeed have generally understood the reverse. The door is not the prize at the end of the change. It is where the change becomes possible.
Two items in this analysis are unresolved and will be updated if they resolve. (1) The toolkit’s reference 8 — the “large-scale study” of veteran overdose mortality — is not named in the document text, and the references section was not retrievable at the time of writing. This piece is written to be robust to that study’s identity; if it turns out to be randomized rather than observational, the section “The claim the whole argument stands on” requires revision. (2) The MOUD-as-abstinence question is posed here as open because the published toolkit does not answer it; if a subsequent federal clarification answers it, that section should be updated.
Litigation status: no reissued NOFO and no open comment docket as of this date. HUD’s competition page states the department is evaluating the August 7 order and considering its options, including appeal.
Every factual claim in this piece traces to the Common Ladder evidence base (KB IDs in the notes below) or a named primary source. The one modeling study is flagged as directional rather than causal in both the body and its note. Toolkit characterizations and quotations are from the published federal PDF.
Notes
- ↩ Primary source. Best Practices Toolkit: Addressing Homelessness and Addiction through “Treatment First,” Office of National Drug Control Policy, U.S. Department of Health and Human Services, and U.S. Department of Housing and Urban Development, released August 12, 2026. Announced as HUD No. 26-059 and carried by SAMHSA as PEP26-04-009. All toolkit characterizations and quotations in this piece are from the published PDF. ⚠ The document’s References section was not retrievable at the time of writing; footnote 8 of the toolkit — the citation supporting its central mortality claim — is therefore unnamed here. See the currency note above.
- ↩ KB SR-SBP-2 (CANONICAL). Critical Time Intervention: Susser et al., American Journal of Public Health (1997); Herman et al., Psychiatric Services 62(7) (2011); systematic review evidence consistent across populations and countries. CTI is listed on SAMHSA’s Evidence-Based Practices Resource Center. The toolkit’s “warm handoff” is, operationally, a CTI-shaped intervention.
- ↩ KB F-17 (CANONICAL), with F-4 / F-5. PSH cost-offset savings accrue to Medicaid, mental health and corrections budgets rather than to housing budgets, and capital and services are appropriated separately. This cross-agency misalignment is the primary structural reason communities build supportive housing whose service layer is never funded at the level the trials tested. A single part-time case manager for sixty residents is not the intervention the evidence evaluated. See also Housing Math.
- ↩ KB F-28 (CANONICAL). Lachaud, Nisenbaum, Mejia-Lancheros, Latimer, Aubry et al., Housing and Support Intervention and Mortality Among Homeless Adults With Mental Illnesses: A Secondary Analysis of a Randomized Clinical Trial, JAMA Network Open 2025;8(7):e2524302, doi:10.1001/jamanetworkopen.2025.24302. Secondary mortality analysis of the At Home/Chez Soi randomized trial — n = 2,255 randomized, 2,108 (93.5%) linked to administrative health data, five Canadian cities, mortality followed to March 30, 2019. Pooled adjusted hazard ratio 0.83 (95% CI 0.43–1.22); pooled log mortality rate ratio −0.07 (95% CI −0.36 to 0.22). The point estimate is directionally favorable and the interval is wide enough to be compatible with a real benefit — this is an absence of demonstrated effect, not a demonstration of absence. The honest characterization is “not shown.”
- ↩ The correction runs through current evidence-based practice materials, including this project’s own evidence base. Housing stability, reduced emergency-system use, and cost offsets in high-utilizer populations remain well supported; a mortality-reduction claim does not, and has been struck where it appeared.
- ↩ KB F-2 (CANONICAL). Goering et al., Psychiatric Services (2014); Stergiopoulos et al., Journal of Urban Health (2021), seven-year follow-up. At Home/Chez Soi, five cities. Housing First participants spent 73% of follow-up time stably housed versus 32% for treatment as usual.
- ↩ KB F-2 (CANONICAL). Santa Clara County Project Welcome Home randomized trial, eClinicalMedicine (2022).
- ↩ KB F-2 / Intervention Library §6. Tsemberis et al., American Journal of Public Health (2004), and the replication literature: housing-first approaches retain 80–90% of tenants at twelve months against 30–50% for treatment-first models.
- ↩ KB F-3 (CANONICAL). Rees et al., BMJ Open (2019) and subsequent meta-analyses: large and consistent effects on housing stability; inconsistent effects on mental health, substance use and quality of life relative to treatment as usual — in both directions. The treatment-as-usual comparison arms did not outperform Housing First on substance use.
- ↩ Intervention Library §7 (transitional housing). Superseded in most adult contexts by the Housing First evidence base. Exits to permanent housing 50–70%, highly program-dependent; cost $15,000–$35,000 per unit per year; long-term stability lower than PSH for high-acuity households. Treatment-first entry requirements are recorded there as a documented mechanism for excluding the hardest-to-house. Transitional housing retains genuine value for specific populations — youth and young adults, DV survivors, people exiting incarceration — where structure and skill-building serve distinct needs; the objection here is to its use as the default adult pathway. See Intervention Matching.
- ↩ Zwick et al., JAMA Network Open 2025;8(6), doi:10.1001/jamanetworkopen.2025.17095. RESPOND microsimulation. Per 1,000 person-years: housing without a treatment requirement, 49.2 deaths and 5,014 person-weeks on medication treatment; status quo, 50.4 and 2,990; encampment sweeps, 53.1 and 1,694. Modeling study — held as directional, not causal, evidence.
- ↩ KB F-16 (CANONICAL). Glynn, Byrne & Culhane, Annals of Applied Statistics (2021): above a rent-to-income threshold near 30 percent, homelessness rises disproportionately; areas above that threshold hold roughly 15 percent of the U.S. population and 47 percent of people experiencing homelessness. On the income floor: full SSI benefits sit below fair-market rent for a one-bedroom apartment in essentially every U.S. metropolitan market. ⚠ This is the most checkable sentence in the piece and should be re-verified against the current-year FMR schedule before republication.
- ↩ Services Best Practices §2.2. Buprenorphine, methadone and naltrexone are FDA-approved medications with strong evidence for reducing opioid use, overdose risk and mortality in opioid use disorder. MOUD is not a substitute for behavioral treatment but is an evidence-based medical treatment in its own right; housing programs that prohibit or fail to facilitate access to it operate below the standard of care.
- ↩ KB W20-3 (documentary). No Annual Homelessness Assessment Report Part 2 has been published for the 2023, 2024 or 2025 cycles; the most recent is the 2022 edition, published August 2024. FY2024 remains the latest verified System Performance Measure vintage. This is a documented absence, not an inference.
- ↩ National Alliance to End Homelessness et al. v. HUD, No. 1:26-cv-00436, and Washington et al. v. HUD, No. 1:26-cv-00439 (D.R.I., August 7, 2026) (McElroy, J.): the FY2026 CoC NOFO set aside “in its entirety” for failure to undergo the notice-and-comment process required by the McKinney-Vento Act; permanent injunction denied; the court expressly declining to reach the substantive challenges, including the departure from Housing First, and stating that HUD “may attempt to issue yet another NOFO that contains these conditions after undergoing the requisite notice-and-comment process.” HUD’s Continuum of Care Program Competition page states that the NOFO “and its August 26, 2026 application submission deadline are no longer in force as HUD is unable to accept applications at this time,” and that the department is evaluating the order and considering its options, including appeal — verified August 17, 2026. Full treatment of both rulings is in the companion piece, A Federal Court Just Called HUD’s Housing First Reversal “Unreasoned.”