Skip to content
ParamergeParamerge
Stephen Lieberman
Through Paramerge
Contact

PAN Lab example

CrimSAFE criminal-record tenant screening

Criminal-record tenant screening: the flag the leasing desk was configured not to see

A rental application is screened against a criminal-records store aggregated from more than 800 jurisdictions. Modeled on the CrimSAFE tenant-screening litigation: its shape, not the real system. Nothing here computes a risk score. A filter matches an applicant's identity against held records and applies criteria the housing provider configured — which offenses disqualify, which dispositions count, and how far back to look, from 0 to 99 years for convictions and 0 to 7 years for charges. What comes back is a lease decision, a flag saying records were found, and a message the provider itself wrote. Watch what the provider then did with the settings it owns. It suppressed full record detail from its own on-site staff so that record-based decisions would be made by someone in a more elevated position, out of a concern that leasing commissions should not drive them. The vendor's default would have given every user full access with no cap. So the person facing the applicant knew a disqualifying record existed and nothing about what it was, and told the applicant the application was denied. The elevated tier that did hold the full report later told a state commission it did not know the facts behind the criminal-background findings, because it had trust in the vendor's reports. The record behind the flag was a pending out-of-state shoplifting charge, later withdrawn at the court that filed it. Correction happened there, by petition, because the consumer file-disclosure channel into the store ran four months against a documentation demand the trial court called an impossible condition and was never completed. An appellate court read all of this in 2026 and drew the authority map: the vendor supplies the store, the classification and the mechanism, and the provider owns configuration, record relevance, staff access, letter content and the final call. On that map the housing denial sits beyond the first step from the vendor. The question this board asks is what governs a decision whose formal owner and whose practical owner are different people. Before you pick a target level: this board cannot be won under Service and Safety Targets or All Governance Targets. The benefit reading is not what blocks it. Inside the budget the best legal settings clear the benefit margins comfortably on both tiers, with the service measure well above its floor. Lift the pathway requirement on its own and the board wins at both tiers, from a stack costing 6 of the 13 you have. The pathway gate is the only gate that fails. Under Service and Safety Targets, three pathways stay open at every affordable price: the housing provider configuring the filter criteria, the applicant's identity data entering the match, and the desk applying the configured policy. Under All Governance Targets, path dependence holds a fourth open, the flag delivered at the desk. Each is the product working as sold, and no lever this deployment could actually pull closes them. That is a measurement of the deployment this network is derived from, not a puzzle waiting to be cracked. Explore and Service Targets Only can be won.

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 Record-filter-class tenant screening deployment network: 12 components and 25 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: 4 assumed · 6 published baseline. In the Lab, the shaded evidence band behind each headline readout draws its width from the least-established class below.

  • assumed

    Binding framing, and it is posture discipline. The widely quoted 'hand-in-glove' and 'integral participant' language belongs to the August 2020 summary-judgment opinion, where allegations are construed for the non-movant, and it is deliberately absent from this map: the July 2023 trial ruling and the February 2026 affirmance reached the opposite conclusion on vendor causation. The sentence about no housing provider reasonably believing the vendor makes housing decisions for them is the district court's own language, adopted by the appellate panel at the proximate-cause step, and is attributed that way. No disparity figure appears anywhere on this map, because disparate impact was never proven here: the race and national-origin claim failed at the prima facie causation step, so the underlying statistics were never adjudicated, and the disability claim was defeated by a finding that the conservator request was the first and only one the company had ever received. The fair-housing organization's own claim was dismissed for lack of standing under a doctrine about organizational injury that concerns advocacy groups generally rather than tenant screening, which vacated rather than affirmed the merits disposition of that claim.

  • baseline

    This map draws both layers the appellate authority map certifies, because the case is evidence of the gap between them rather than evidence that the gap does not exist. The formal layer is drawn live: the provider sets the criteria at full strength, holds the approve-or-deny call and the letter at a substantial level, and may approve despite any flag. The practical layer is drawn where the record puts it: the record detail reaching the leasing desk is empty because the provider affirmatively reconfigured it that way, the individualized review of a flag is empty because the denial was communicated without one and the elevated tier told the state commission it did not know the facts behind the findings, and the flag itself reaches the desk at full strength as the operative fact of the decision. A reader who takes either layer alone will misread this deployment.

  • baseline

    The two write-back paths into the vendor's store are both drawn empty, and each rests on a separate documented fact rather than on one fact counted twice. The disclosure-desk correction is empty because the file-disclosure channel ran from June 2016 to mid-November 2016 against a documentation demand the trial court called an impossible condition and was never completed. The reconciliation against the originating court is empty because the charge was withdrawn at that court in April 2017 after the family petitioned it, and the record describes no path by which that withdrawal reached the aggregated store. Together they are why the dossier describes this deployment as cleanly modelable as memory-store contamination with a broken audit edge.

  • baseline

    This map uses the staff-to-model pathway for documented operator input into the model's decision surface, not for conversational framing, and the reading is recorded here so a reviewer can check it rather than infer it. Two such pathways exist and they carry different things. The screening policy owners set the offense types, dispositions and lookback window at full strength, within vendor-supplied ranges of 0 to 99 years for convictions and 0 to 7 years for charges; the appellate court treated that configuration as the first of five provider decisions and rested its causation holding on the chain. The leasing desk enters the applicant's identifying details at a substantial level, which is what the engine matches on across more than 800 jurisdictions and therefore where a false match originates. Neither is a prompt, and no prompt surface exists in this deployment.

  • baseline

    baselineDemand 2 is derived and is deliberately not 3. The record documents a routine per-application screening stream at one residential property, with no surge, no backlog and no queue anywhere in it; the only volume figures in the evidence are industry-level journalism estimates of a roughly three billion dollar tenant-screening sector with near-universal landlord adoption, which are not measurements of this deployment. The one primary-source workload fact points the other way: the conservator file-disclosure request was the first and only one the company had ever received. manualCapacity 3 is the counterfactual floor, authored high because the screen substituted for the individualized landlord review that the report's own default message still instructs the provider to perform, and because every human channel in this record that was actually engaged reached a better answer than the screen did. The fourteen-month latency to that engagement is a governance failure rather than evidence of a weak human counterfactual, so this deployment can legitimately read net-negative.

  • baseline

    Three review steps are drawn because the record documents three oversight bodies whose behavior differs from one another, and each is wired in by an inbound read rather than left dangling. The vendor's legal department reversed its own disclosure desk once, about five months after the request arrived and within about two weeks of being reached, and no completed disclosure followed, so its outbound review is faint. The state commission held one evidentiary hearing and the provider approved the move-in ten days later, which is the only oversight action in this record documented to change an outcome, so its outbound check is faint. The federal courts produced the authoritative authority map and imposed nothing: the sole liability finding was reversed, final vendor liability was nothing, and no injunction, consent decree or ordered policy change appears in the record, so the binding requirement they place on the disclosure procedure is empty. Three reviewers at faint, faint and empty is the shape of oversight this deployment actually had.

  • assumed

    Documented actors deliberately not drawn, and why. The federal fair-housing agencies filed an amicus brief in November 2023 arguing that tenant-screening companies are not categorically outside fair-housing law, and the appellate panel accepted the doctrinal point while affirming on causation; their documented act is a one-time legal-position filing rather than a recurring pathway into this deployment, so it is narrated in the courts node rather than drawn as a reviewer that would need an invented inbound flow. The fair-housing organization that monitored the practice as a co-plaintiff is narrated in the same node, because its claim was vacated for lack of standing and no adjudicated monitoring relationship with this deployment survives; drawing it would be drawing an oversight edge the courts removed. No enforcement node is drawn because no downstream action system driven automatically by a record is documented here, and the denial is the housing provider's own decision. No worklist is drawn because the four-month disclosure impasse is a documented latency and not a documented queue, and the record affirmatively states this was the only request of its kind. No externalBoundary is drawn because no source describes applicant data leaving the governed system; the report reaching the provider is a permitted delivery to the intended recipient under a consumer-reporting regime, not egress.

  • assumed

    The vendor's product and engineering authority is carried in copy rather than as a fifth group of staff, because the dossier's operators block names only two vendor-side operator groups, the consumer-relations disclosure desk and the legal department above it. The vendor's real design authority is substantial and is stated where it belongs: it supplies the store, the three-category classification with its sub-classifications, the matching mechanism, the configurable ranges, the default report message and the default that authorizes every new user to receive full record data with no cap. Drawing a vendor product group the evidence does not describe would be a decorative component, and the appellate holding that the harm sits beyond the first step from the vendor is precisely a holding about where that authority stops.

  • assumed

    Served people are not in these dynamics and no served-person outcome is computed anywhere on this map. Rental applicants, conservatees and their households are boundary-only: a flag, a report, a letter or a hearing on this map is an institutional signal, never a person. The documented household harm sits outside these dynamics by construction, and it is the fourteen months between the April 2016 denial and the move-in in late June 2017 for a family seeking to bring home a son who could not speak, walk or care for himself. The record also describes an information latency of about eight months before the family learned which record had caused the denial, and it learned that from the housing provider rather than from the vendor. Both live in the case file and are measured outside any diagram like this one.

  • baseline

    The feedback pathway from screening outcomes back into the engine is drawn empty because the record denies it exists, in terms: there is no retraining loop and no score to update. That is an unusual thing for this atlas to draw, and it is drawn rather than omitted because a reader needs to see that the danger here is not a learning loop. It is a store that holds a stale copy of a court record, a filter that reads the store, a desk that cannot see what the filter found, and no path by which a correction at the source reaches any of them. A future autonomy pressure could open that pathway; nothing in this record did.

What this example does not show

  • Served people are not modeled here; the Lab models institutional propagation only, and the outcomes that land on rental applicants and their households are documented in the case file and measured outside any diagram like this one. A flag, a report, a letter or a hearing on this map is an institutional signal, never a person. The documented household harm sits outside these dynamics by construction: about fourteen months between the April 2016 denial and the move-in in late June 2017, for a family seeking to bring home a son who could not speak, walk or care for himself after a 2015 injury, and an information latency of about eight months before the family learned which record had caused the denial, learned from the housing provider rather than from the company holding the record.
  • Posture discipline is the case, and this board keeps it. The widely quoted 'hand-in-glove' and 'integral participant' language is from the August 2020 summary-judgment opinion, where allegations are construed for the party opposing the motion; the July 2023 trial ruling and the February 2026 affirmance reached the opposite conclusion on vendor causation, and nothing on this map presents the 2020 language as a finding that survived trial. The sentence that no housing provider using this screen could reasonably believe the vendor makes housing decisions for them is the district court's own language, quoted with approval by the appellate panel at the proximate-cause step, and is attributed that way. The appellate panel rejected the threshold reasoning that screening companies sit categorically outside fair-housing law, and still affirmed on causation, so the two halves of that decision must be read together.
  • Disparate impact was never proven and no disparity figure appears anywhere in this network. The race and national-origin claim failed at the prima facie causation step, so the courts never adjudicated the underlying statistics; the general premise that criminal-record screening burdens some applicants more than others appears in party briefs, in a federal amicus filing and in journalism, not in validated findings from this case. The disability claim was defeated by a finding that the conservator file-disclosure request was the first and only one the company had ever received, and the argument that people with disabilities were about eleven percent of the state's population but all of those harmed by the conservatorship policy was rejected. The vendor's final liability in this litigation was nothing.
  • The fair-housing organization that monitored this practice is narrated rather than drawn, and the reason matters. Its own claim was dismissed for lack of standing under a doctrine about organizational injury, decided in a case that had nothing to do with tenant screening, which vacated rather than affirmed the merits disposition of that claim. Drawing it as an oversight node would draw a monitoring relationship the courts removed for reasons external to this deployment; leaving it out entirely would lose the fact that an oversight edge here depended on a doctrinal parameter nobody in this deployment controls. It is therefore carried in the courts node and in the case file. Federal fair-housing agencies are treated the same way: their documented act is a one-time amicus filing in November 2023, not a recurring pathway.
  • One application is documented in full, and the network is drawn at that resolution. The suppression of record detail from on-site staff, the trust in the vendor's reports, the impossible documentation condition and the single reversal by a state commission are all findings about this deployment and this applicant; they are not measured rates. The industry-scale figures that exist in the record are journalism estimates from 2021 and are not used to set anything on this board. The conservator request being the first and only one the company had ever received is a primary-source finding and is what the baselineDemand of 2 rests on, alongside the absence of any documented surge, backlog or queue. The corporate line that sold this screen was rebranded in 2016 and later divested, and the successor company is the subject of a separate case about a different product with a different legal theory; the exact divestiture date is corroborated outside the sources cited here and is therefore not asserted on this map.

Sources and evidence

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

  • CrimSAFE, a criminal-record tenant-screening product sold by CoreLogic Rental Property Solutions, computes no risk score: it is a deterministic record-matching and filtering engine that matches applicant identity data against a database of court and arrest records aggregated from more than 800 US jurisdictions, classified into three primary categories (Crimes Against Property, Crimes Against Persons, Crimes Against Society) with sub-classifications, and applies filter criteria the HOUSING PROVIDER configures: offense type, disposition, and severity across felony and non-felony convictions and charges, and a lookback period configurable from 0 to 99 years for convictions and 0 to 7 years for charges (federal consumer-reporting law permits reporting non-conviction records for seven years). The output is a report carrying a lease decision driven by the provider's criteria plus a credit score, a Record(s) Found flag, message text the provider authors, full record detail for the users the provider authorizes, and an optional provider-customizable adverse-action letter template. Every new CrimSAFE user is by default authorized to receive full record data, with no cap on how many users get full access; a provider must affirmatively change configuration settings to restrict full reports to senior managers. In April 2016 Carmen Arroyo's application to move within ArtSpace in Windham, Connecticut, so her son Mikhail could live with her after a 2015 injury, was denied on 26 April 2016 after the screen returned Record(s) Found; the only matched record was a pending Pennsylvania shoplifting charge, later withdrawn in April 2017. Mikhail's report carried the vendor's default message: 'Please verify the applicability of these records to your applicant and proceed with your community's screening policies.'

    empirical
    • Government Connecticut Fair Housing Center v. CoreLogic Rental Property Solutions LLC, Nos. 23-1118(L) and 23-1166(XAP) (2d Cir. Feb. 20, 2026) (Menashi, J.), slip opinion https://business.cch.com/BFLD/CFHC-v-CoreLogic-Rental-Prop-2ndCir-02202026.pdf
    • Government Connecticut Fair Housing Center v. CoreLogic Rental Property Solutions LLC, No. 3:18-cv-705, 2023 WL 4669482 (D. Conn. July 20, 2023) (Bryant, J.), bench trial decision https://law.justia.com/cases/federal/district-courts/connecticut/ctdce/3:2018cv00705/125021/317/
    • Advocacy National Housing Law Project, Arroyo v. CoreLogic case page (2018) https://www.nhlp.org/our-initiatives/arroyo-v-corelogic/

  • The location of decision authority in this deployment was the contested question, and the Second Circuit resolved it component by component on 20 February 2026 (Cabranes, Wesley, Menashi; opinion by Menashi; Nos. 23-1118(L), 23-1166(XAP)). WinnResidential had suppressed full reports from its own on-site staff so that leasing decisions involving criminal records would be made 'by someone in a more elevated position,' out of concern about leasing-commission incentives, so the on-site agent saw only a Record(s) Found flag and told Arroyo the application was denied without individualized review; answering the state commission's complaint, WinnResidential said it did not know 'the facts behind the criminal background findings' because it had 'trust' in CoreLogic's reports. After a ten-day bench trial Judge Vanessa L. Bryant ruled on 20 July 2023 that CrimSAFE does not disqualify applicants because the housing provider decides what records matter and whether to deny; the earlier August 2020 summary-judgment characterizations that the companies 'acted hand-in-glove' and that CoreLogic 'was an integral participant' belong to that posture and did not survive trial. The Second Circuit rejected the threshold reasoning that screening companies sit categorically outside the Fair Housing Act — a point the United States had urged as amicus on 24 November 2023 — but affirmed on proximate cause, holding the denial came after a chain of the provider's discretionary decisions (configuration, record relevance, staff access, adverse-action letters, final approval) and quoting the district court's own sentence with approval: 'No housing provider who uses CrimSAFE could reasonably believe that CoreLogic makes housing decisions for them.' It rejected the 'cat's paw' theory because the screening policies applied were the provider's own, rejected liability for failing to restrict lawfully reportable non-conviction records as extending liability beyond the first step, and dismissed the Connecticut Fair Housing Center's own claim for lack of Article III standing under a 2024 organizational-injury doctrine unrelated to tenant screening, vacating rather than deciding its merits. Disparate impact was never proven: the race and national-origin claim failed at the prima facie causation step, so the underlying statistics were never adjudicated.

    empirical
    • Government Connecticut Fair Housing Center v. CoreLogic Rental Property Solutions LLC, Nos. 23-1118(L) and 23-1166(XAP) (2d Cir. Feb. 20, 2026) (Menashi, J.), slip opinion https://business.cch.com/BFLD/CFHC-v-CoreLogic-Rental-Prop-2ndCir-02202026.pdf
    • Government Connecticut Fair Housing Center v. CoreLogic Rental Property Solutions LLC, No. 3:18-cv-705, 2023 WL 4669482 (D. Conn. July 20, 2023) (Bryant, J.), bench trial decision https://law.justia.com/cases/federal/district-courts/connecticut/ctdce/3:2018cv00705/125021/317/
    • Government Connecticut Fair Housing Center v. CoreLogic Rental Property Solutions LLC, 478 F. Supp. 3d 259 (D. Conn. 2020), summary judgment opinion https://www.quimbee.com/cases/connecticut-fair-housing-center-v-corelogic-rental-property-solutions-llc
    • Government United States Department of Justice Civil Rights Division with the Department of Housing and Urban Development, Brief of the United States as Amicus Curiae, Connecticut Fair Housing Center v. CoreLogic Rental Property Solutions, No. 23-1118 (2d Cir. Nov. 24, 2023) https://www.justice.gov/d9/2023-11/connecticut_fair_housing_center_v._corelogic_rental_property_solutions_no._23-1118_2d_cir._11.24.23.pdf

  • The correction loop in this deployment ran through records the subject could not see at the point of harm. CoreLogic's Authentication Procedure Guide listed only a notarized power of attorney as third-party authorization and escalated 'any scenarios not covered' to a supervisor; staff demanded a power of attorney that Mikhail Arroyo, a conservatee, was legally incapable of executing — a demand the trial court called an 'impossible condition' — across a blocked window running from the 24 June 2016 request with a conservatorship certificate to mid-November 2016, when a 1 November call escalated to CoreLogic's legal department and the company agreed about two weeks later that a conservatorship certificate with a visible probate seal would suffice; the resubmitted copy again lacked a visible seal and the disclosure was never completed. That escalation is the only documented change in vendor behavior in the record. The family learned which record had caused the denial in December 2016, roughly eight months after the denial, from the housing provider rather than the vendor, and the adverse-action letter that should have triggered the correction loop was sent but never received; correction ultimately happened at the original source when Arroyo petitioned the Pennsylvania court and the charge was withdrawn in April 2017. Arroyo's was the first and only conservator file-disclosure request CoreLogic had ever received (478 F. Supp. 3d at 282), a finding both courts used to defeat the disability disparate-impact claim. The Connecticut Commission on Human Rights and Opportunities held an evidentiary hearing on 13 June 2017 and WinnResidential approved the move-in ten days later — about fourteen months after the denial, and the only oversight action documented to have changed an outcome. The sole liability finding, a willful consumer-reporting violation carrying $1,000 statutory and $3,000 punitive damages awarded in July 2023, was reversed on 20 February 2026; final vendor liability was zero, and no injunction, consent decree, or policy change was ever ordered.

    empirical
    • Government Connecticut Fair Housing Center v. CoreLogic Rental Property Solutions LLC, Nos. 23-1118(L) and 23-1166(XAP) (2d Cir. Feb. 20, 2026) (Menashi, J.), slip opinion https://business.cch.com/BFLD/CFHC-v-CoreLogic-Rental-Prop-2ndCir-02202026.pdf
    • Government Connecticut Fair Housing Center v. CoreLogic Rental Property Solutions LLC, 478 F. Supp. 3d 259 (D. Conn. 2020), summary judgment opinion https://www.quimbee.com/cases/connecticut-fair-housing-center-v-corelogic-rental-property-solutions-llc
    • Government Connecticut Fair Housing Center v. CoreLogic Rental Property Solutions LLC, No. 3:18-cv-705, 2023 WL 4669482 (D. Conn. July 20, 2023) (Bryant, J.), bench trial decision https://law.justia.com/cases/federal/district-courts/connecticut/ctdce/3:2018cv00705/125021/317/
    • Government CourtListener and the Free Law Project, Docket, Connecticut Fair Housing Ctr v. CoreLogic Rental Property Solutions LLC, No. 3:18-cv-00705 (D. Conn.) (2023) https://www.courtlistener.com/docket/6691528/connecticut-fair-housing-ctr-v-corelogic-rental-property-solutions-llc/

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