Skip to content
ParamergeParamerge
Stephen Lieberman
Through Paramerge
Contact

PAN Lab example

Douglas County Decision Aide

The score read only at the edges: a child-welfare screening aide

A 1-to-20 risk score, built from years of administrative records and deliberately blind to race, lands in front of a consensus screening team that keeps full discretion. Modeled on Douglas County's Decision Aide. An independent trial found it sped decisions without significantly changing outcomes: workers read the extremes and left the middle alone. Watch two quiet loops — memory, where today's decisions become tomorrow's inputs, and discretion, which is doing more work here than the score is.

Stylized model of a documented deploymentChild welfare & family 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 Douglas-class independently-evaluated screening aide network: 4 components and 10 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 · 1 published baseline. In the Lab, the shaded evidence band behind each headline readout draws its width from the least-established class below.

  • assumed

    This models the independently-evaluated, high-discretion human-in-the-loop screening pattern documented in the Douglas County Decision Aide case file — not a reconstruction of the actual tool.

  • assumed

    Peer pathways are authored on both signs: the RED Team's supervisor-plus-two-caseworkers consensus is a documented, built-in second read (a real inhibiting check), while desk-to-desk anchoring inside the team runs at baseline, and one scoring model rates every referral so its blind spots are shared.

  • baseline

    The consensus team retains genuine discretion: the independent trial found the score sped decisions without significantly changing outcomes and that workers heeded mainly extreme scores, so the score anchors — rather than replaces — the decision.

  • assumed

    The feedback loop from decisions into future scores is present at baseline: scores are computed from accumulated multi-agency records, and the predicted target (out-of-home removal) is itself a product of prior human decisions, so today's screening shapes tomorrow's inputs.

  • assumed

    Race predictors were deliberately excluded and no race-disaggregated Douglas outcome disparity was measured (the county held no such data); the strong positive findings sometimes attributed to this tool come from a separate sibling deployment in another county. The Lab models institutional workflow propagation, not demographics, and estimates no differential harm to served people.

What this example does not show

  • The documented deployment excluded race predictors and no race-disaggregated outcome disparity was measured for it; the Lab models institutional workflow propagation, not demographics, and estimates no differential harm to served people.
  • The strong positive findings sometimes credited to this tool in the press — a large reduction in child-injury hospitalizations and reduced surveillance of low-risk Black children — come from a separate sibling deployment in another Colorado county, documented in the case file, not from the shape modeled here.

Sources and evidence

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

  • The Douglas County Decision Aide, deployed into the county's RED-Team call-screening process in February 2019, scores each referral from 1 to 20 for a child's likelihood of out-of-home removal within two years; an independent Cornell-led randomized controlled trial found it sped up screening decisions without significantly changing child outcomes, and a companion study found workers attended mainly to extreme scores while largely disregarding mid-range ones.

    empirical
    • Academic Vaithianathan et al. (Centre for Social Data Analytics, AUT), Implementing a Child Welfare Decision Aide in Douglas County: Methodology Report (2019) https://csda.aut.ac.nz/__data/assets/pdf_file/0009/347715/Douglas-County-Methodology_Final_3_02_2020.pdf
    • Academic Fitzpatrick, Sadowski and Wildeman, Algorithms and Decision-making: Evidence from Child Maltreatment Reports (Journal of Human Resources, 2025) https://jhr.uwpress.org/content/early/2025/08/01/jhr.0224-13437R2
    • Academic Eiermann, Fitzpatrick, Sadowski and Wildeman, How Do (Human) Child Welfare Workers Respond to Machine-Generated Risk Scores? (Sociological Science, 2026) https://sociologicalscience.com/articles-v13-1-1/

Where this connects

Institutional pressures in this domain

  • Workload surge — Demand outruns staffing; per-case attention shrinks and review becomes triage.
  • Deadline pressure — Statutory or managerial timeliness rules reward fast approval of machine output over slow disagreement.
  • Staff turnover — Experienced skepticism leaves; new staff calibrate their trust on the tool itself.
  • 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 Child welfare & family services domain page.

Levers available here and the patterns behind them

Documented case histories