Skip to content
ParamergeParamerge
Stephen Lieberman
Through Paramerge
Contact

PAN Lab example

Santa Clara County Homelessness Prevention System

A measured lever on an unmeasured target: a homelessness-prevention screen

Santa Clara's questionnaire scores households seeking emergency money to keep their homes. A trial shows the money works. Does it go to the right households?

See more

The screen is a points questionnaire, not machine learning, used at intake in Santa Clara County's Homelessness Prevention System, run by Destination: Home. It scores a household's risk of becoming homeless from self-reported answers, such as domestic-violence history, prior homelessness, and disability. The program uses that score to decide who is eligible and who is offered emergency financial help first.

How the help works

Caseworkers at Sacred Heart Community Service and partner community organizations decide case by case what help a household gets, and how much. The help has covered back rent, deposits, utilities, car repairs, medical bills, and other expenses that destabilize a household.

In the trial, the help averaged about 2,000 dollars, typically one to two months of back rent paid directly to landlords. Today the average, repeat help included, is about 6,500 dollars by CalMatters' count and about 7,000 dollars by Destination: Home's.

Non-financial help comes with the money, such as credit counselling and landlord dispute resolution. Households can come back for more help. The case file calls the help a benefit, never a denial.

Who runs it

Destination: Home, a nonprofit, launched the system in 2017 with about 1 million dollars in donations. It served roughly 200 households in its first year.

CalMatters reports that Sacred Heart Community Service coordinates the system through partner community organizations. It reports a budget of roughly 30 million dollars a year, mostly public funds. The County of Santa Clara fully integrated the system into its safety net in 2024. It now assists about 2,500 households a year.

Destination: Home reports assisting more than 31,000 households, nearly 44,000 people, since 2017. It reports that more than 90 percent remain housed two years later. These are its own figures, not independently audited. Its own page does not name Sacred Heart or the budget.

How the trial worked

The University of Notre Dame's Wilson Sheehan Lab for Economic Opportunities ran a registered randomized controlled trial, led by David C. Phillips and James X. Sullivan. In a randomized controlled trial, a lottery decides who gets the help, so the two groups can be compared fairly. It enrolled 1,263 people from July 2019 to December 2020. Of them, 514 were in the group given money and 749 in a comparison group.

Applicants had to be at imminent risk, score 8 to 13 on the questionnaire, and be ineligible for other prevention programs. Because demand exceeded the funds, a lottery decided who got the money. Both groups got non-financial help.

Homelessness was measured through the county Homeless Management Information System, called HMIS. It records whether a household later appears in shelter or homeless services.

What the figures say

The effect is large, and its figures come from different documents, so they are not combined here. Releases from Notre Dame and its evaluation lab say assisted households were 81 percent less likely to become homeless within six months. They were 73 percent less likely within twelve.

CalMatters reports that 0.9 percent of assisted households became homeless, against 4.1 percent of the others, and pairs that with a 78 percent reduction. The published article's abstract says the help cut homelessness by 3.8 percentage points from a 4.1 percent base rate. Effects were larger for people with a history of homelessness and no children.

The researchers conservatively estimate 2.47 dollars in community benefits per net dollar spent. The sources do not define the net dollar. The article appeared in the Review of Economics and Statistics in September 2025, and the findings were first publicized in mid-2023. Its abstract claims the first trial evidence isolating the effect of financial help on homelessness. Destination: Home's framing of a first randomized controlled trial of its kind is its own, and broader.

The question the trial leaves open

Becoming homeless is rare, even among applicants at risk. About 96 percent of the trial's comparison group never became homeless without the money. So the program's own evaluators name accurate targeting, not the strength of the help, as the main limit.

Co-author James Sullivan cautions on the record that "precious resources" could go to people who would have stayed housed anyway. That would divert funds from shelter and permanent housing. Janey Rountree of the UCLA California Policy Lab has made a related point, as CalMatters paraphrases it. Most people at risk stay housed with help from family and friends, which makes accurate targeting hard.

Help spent on households that would have kept their homes stays hidden. Those households do stay housed, so the spending looks exactly like success. The trial shows the help works. It does not show whether the questionnaire picks the households that would otherwise lose their homes.

Limits of the trial

The trial studied only a middle band of scores, so effects at other risk levels are unmeasured. Its enrollment overlapped COVID-era eviction moratoria and emergency rental assistance. Those may have lowered how often people became homeless.

The national copy

On February 24, 2026, Destination: Home launched Right at Home, a national initiative to copy the program in ten communities from 2026 to 2031. It is backed by 77 million dollars from The Audacious Project at TED, Cisco, Sobrato Philanthropies, and the Valhalla Foundation. CalMatters reports nearly 80 million dollars raised.

Eight communities are named and two are pending. Each site is to receive at least 5 million dollars over three years and begin by January 2027. The goal is to keep more than 10,000 households housed. Notre Dame's lab is the common national evidence partner, evaluating every site. In 2024, Results for America released a toolkit for copying the program.

The sites differ in structure: urban counties, a state coalition, and tribal communities. So Right at Home tests whether help measured once, in one county, works the same elsewhere. That help was measured on a middle band of scores, during an eviction moratorium.

What the available tools can and cannot address

A failure pathway is a link between two parts of the network where a mistake made by one part can be passed on to the other. As the Lab opens, six links are failure pathways, and three of them also lower the Privacy gauge. A pathway is closed when mistakes stop passing along it, though the work on that link goes on.

Explore (No Targets) sets no targets, and the targets can be met at every other level. Under Service Targets Only, many combinations of tools work. The cheapest is Escalate checks alone, for 2 units, and Gate record entries alone works for 3.

Service and Safety Targets and All Governance Targets ask that every failure pathway be closed and the Privacy gauge refilled. Within the 11-unit budget, each can be met two ways, each for 11 units. Both use Mark AI-written records, Peer sharing rules, Escalate checks, and Keep prompts neutral. One adds Gate record entries, and the other adds Store less data.

Check copied records adds the check this case turns on, Scores checked against outcomes. It closes no failure pathway, so neither way at the two higher levels uses it. Meeting every target here does not answer whether the screen picks the right households.

Stylized model of a documented deploymentHousing & homelessness services

Open this example in PAN Lab v0.1 to apply pressures and levers and watch what the system does.

What this models

This example runs on the Santa-Clara-Prevention-class screened flexible-assistance lever network: 4 components and 11 pathways between them. Every context in the Lab is a stylized model, never a reconstruction of any actual deployment, and each assumption behind it carries a provenance label.

Evidence base: 2 assumed · 4 published baseline. In the Lab, the shaded evidence band behind each headline readout draws its width from the least-established class below.

Show all 6 assumptions
  • assumed

    This network follows the pattern documented in the Santa Clara County Homelessness Prevention System case file. It does not reconstruct the actual program. It shows a points questionnaire, not a machine-learning model, that sets priority for help whose effect a randomized trial measured. The help is flexible emergency financial assistance, given with wide caseworker discretion. The trial randomized within a middle band of scores, 8 to 13. This case differs from the Los Angeles County Homelessness Prevention Unit, which ranks residents with a supervised machine-learning model. That program's trial of whether it works is pending, and its main failure is missing a high-risk person. It also differs from the separate 2016 Silicon Valley Triage Tool, which prioritises people who are already homeless. Reading this screen as a hidden formula, or the help as a rule that turns people away, misreads the case.

  • baseline

    The help is measured, and the effect is good. The University of Notre Dame's evaluation lab ran a registered randomized controlled trial of 1,263 people from July 2019 to December 2020. Of them, 514 were in the group given money and 749 in a comparison group. Outcomes were tracked through the county Homeless Management Information System, called HMIS. Releases from Notre Dame and its evaluation lab report assisted households 81 percent less likely to become homeless within six months, and 73 percent within twelve. The peer-reviewed article's abstract states a 3.8 percentage-point reduction from a 4.1 percent base rate. CalMatters reports the underlying rates as 0.9 percent assisted against 4.1 percent not, and pairs them with a 78 percent reduction. These figures come from separate sources and are not combined here. The researchers' conservative estimate is 2.47 dollars of community benefit per net dollar. The program-scale figures are Destination: Home's own reporting. They are more than 31,000 households, about 7,000 dollars on average, and more than 90 percent housed two years later.

  • baseline

    The case's defining feature is that measuring the help is not measuring the target. Becoming homeless is rare even among applicants at risk: about 96 percent of the trial's comparison group never became homeless. So accurate targeting, not the strength of the help, is the main limit. The study's co-author cautions on the record that precious resources could go to people who would have stayed housed anyway. That misallocation stays hidden, because those people do stay housed, so it looks exactly like success. The network draws two checks. The first compares logged scores with outcomes in the county database, to test whether the screen picks the households that would otherwise lose their homes. The sources describe no one running it as a standing practice. The second is an evaluation at each new site of whether the help works as well there.

  • baseline

    Wide human discretion is part of the design, not a failure to follow the score. Caseworkers decide the type and amount of help case by case. Non-financial services come with the money, and households can return for more help. The lottery replaced first come, first served only during the trial, because demand exceeded funds. The score leads to an offer of help, never a denial. Keep skills sharp is the tool that protects that discretion. Protecting it keeps a benefit from hardening into a list you are either on or invisible to. The county HMIS record of outcomes is the measurement that made the trial possible.

  • baseline

    The program is being copied to new places. As of February 2026, the same help is being set up in about ten structurally different US communities under Right at Home. Each site is to receive at least 5 million dollars over three years and is to begin by January 2027. The goal is to keep more than 10,000 households housed. Notre Dame's evaluation lab, the common national evidence partner, is to assess each site from 2026 to 2031. The link from Santa Clara's caseworkers to the new sites carries the program design and the Results for America replication toolkit. The risk it stands for is the playbook spreading faster than each site's own evidence. Peer sharing rules, which governs what the sites pass along, acts on that link. Right at Home was announced at 77 million dollars, and CalMatters reports nearly 80 million dollars raised.

  • assumed

    The households the program serves are outside the network. Whether any household becomes homeless is documented in the case file and measured outside a diagram like this one, never computed here. Private data enters at two points. One is the self-reported intake, including domestic-violence history and disability, given to obtain help. The other is the county HMIS record of housing outcomes. The evaluation also drew on address-change records from consumer reference data. The trial enrolled a middle band of scores, 8 to 13, during a window overlapping COVID-era eviction moratoria and rental assistance. That may have lowered how often people became homeless. So effects at other risk levels, and outside that window, are unmeasured. The program-scale figures, and the framing of a first randomized controlled trial, are Destination: Home's own claims, not independently audited. A score, a decision on help, or an outcome here stands for an institutional signal, never a person.

What this example does not show

Show all 3 limitations
  • The households the program serves, and whether they keep or lose their homes, are outside this network. Those outcomes are documented in the case file and measured outside any diagram like this one. The questionnaire denies no one and decides no case against anyone. The harm this network shows is misallocation and misplaced confidence, never a decision about a person. It shows help spent on households that would have stayed housed, reliance on a screen, and a result assumed to hold elsewhere.
  • The trial's effect figures come from separate documents and must not be combined. CalMatters reports 0.9 percent of assisted households becoming homeless against 4.1 percent of others, a 78 percent reduction. Releases from Notre Dame and its evaluation lab report 81 percent lower within six months and 73 percent within twelve. The published article's abstract reports a 3.8 percentage-point reduction from a 4.1 percent base rate, the rate without the money. The trial randomized a middle band of scores, 8 to 13, from July 2019 to December 2020. That window overlapped COVID-era eviction moratoria and rental assistance, so effects at other risk levels and outside that window are unmeasured. The 2.47 dollar benefit per net dollar is the researchers' own conservative estimate. The program-scale figures are Destination: Home's own reporting, not independently audited. They are more than 31,000 households, about 7,000 dollars on average, more than 90 percent housed at two years, and the first randomized controlled trial framing.
  • How this network looks when the Lab opens is no safety promise for any real program. The limit the program's own evaluator names is how accurately the screen targets help when homelessness is rare. The trial's effect does not settle it. It has not yet been measured at the ten new sites, whose evaluations run from 2026 to 2031. Right at Home was announced at 77 million dollars, and CalMatters reports nearly 80 million dollars raised. Its goal is more than 10,000 households kept housed.

Sources and evidence

What this example rests on, claim by claim. Every entry resolves to the same ledger the Evidence Registry publishes.

  • In a registered randomized controlled trial of 1,263 imminent-risk applicants (514 treatment, 749 control) run by the University of Notre Dame's evaluation lab, households offered flexible emergency financial assistance averaging about 2,000 dollars, typically one to two months of back rent, through Santa Clara County's homelessness-prevention system were reported 81 percent less likely to become homeless within six months and 73 percent within twelve months; the peer-reviewed article's abstract states the assistance reduced homelessness by 3.8 percentage points from a 4.1 percent base rate, and the researchers conservatively estimated 2.47 dollars in community benefits per net dollar spent.

    empirical
    • Academic Phillips and Sullivan, Do Homelessness Prevention Programs Prevent Homelessness? Evidence from a Randomized Controlled Trial (The Review of Economics and Statistics 107(5): 1187 to 1196, 2025) https://doi.org/10.1162/rest_a_01344
    • Reference University of Notre Dame News, Targeted Prevention Helps Stop Homelessness Before It Starts (2023) https://news.nd.edu/news/targeted-prevention-helps-stop-homelessness-before-it-starts/
    • Academic Phillips and Sullivan, Do Homelessness Prevention Programs Prevent Homelessness? Evidence from a Randomized Controlled Trial (AEA RCT Registry, AEARCTR-0008261, 2021) https://www.socialscienceregistry.org/trials/8261

  • Because becoming homeless is statistically rare even among at-risk applicants - about 96 percent of the trial's control group never became homeless without assistance - the program's own co-author cautions that prevention resources can flow to households that would have stayed housed anyway, making screening precision on a low base rate the binding constraint; as of February 2026 the model is being replicated across about ten heterogeneous US jurisdictions under a 77-million-dollar initiative, with the same evaluation lab as the common evidence partner assessing each site.

    empirical
    • Investigative Kendall, A New Homelessness Strategy Is Sweeping California (CalMatters, 2026) https://calmatters.org/housing/homelessness/2026/03/homelessness-prevention-pilot/
    • Academic Phillips and Sullivan, Do Homelessness Prevention Programs Prevent Homelessness? Evidence from a Randomized Controlled Trial (The Review of Economics and Statistics 107(5): 1187 to 1196, 2025) https://doi.org/10.1162/rest_a_01344
    • Advocacy Destination: Home, Destination: Home Launches Right at Home, a National Initiative to Stop Homelessness Before It Starts (2026) https://destinationhomesv.org/news/2026/02/24/destination-home-launches-right-at-home-a-national-initiative-to-stop-homelessness-before-it-starts/
    • Reference University of Notre Dame News, Notre Dame's LEO Joins National Initiative to Stop Homelessness Before It Starts, Serving as the Lead Evidence Partner (2026) https://news.nd.edu/news/notre-dames-leo-joins-national-initiative-to-stop-homelessness-before-it-starts-serving-as-the-lead-evidence-partner/

Where this connects

Institutional pressures in this domain

  • Workload surge — Demand outruns staffing; per-case attention shrinks and review becomes triage.
  • Austerity & recovery incentives — Cost-cutting and overpayment-recovery targets tilt the system toward denial and enforcement errors.
  • Vendor opacity — The deploying institution cannot inspect the model, data, or update pipeline it is accountable for.
  • Data & policy drift — The world, the intake process, and the rules change under a system trained on how things used to be — two mechanisms with different remedies: the statistical properties of what the system processes move (concept drift), or the mixture of inputs arriving in deployment differs from the mixture it was trained on (covariate shift).
  • Compliance over substance — Paper controls (sign-offs, checklists) satisfy audits while the behavior they describe erodes.

All of them in context on the Housing & homelessness services domain page.

Levers available here and the patterns behind them

Documented case histories