Domain Atlas / Child welfare & family services
Illinois Rapid Safety Feedback
Illinois's Rapid Safety Feedback flagged thousands of children at 90-percent-or-higher risk of serious harm — beyond any caseload's capacity to act — while children who died in known-to-system cases had not been flagged; the agency ended its use in 2017.[3]
What happened
Illinois adopted the Eckerd Connects/MindShare Rapid Safety Feedback tool as a centerpiece of the reforms promised under DCFS director George Sheldon, who arrived in 2015 after a run of child deaths and brought the roughly $366,000 program in through a no-bid arrangement the state classified as a grant — a classification a July 2017 joint report by the Illinois Executive Inspector General and the DCFS Inspector General found to be mismanagement, because it sidestepped bidding-transparency rules. The tool scored every child named in a hotline maltreatment report on a 1-to-100 scale for the probability of death or serious injury within two years, and a high score was designed to trigger a supervisory "second set of eyes" review and caseworker coaching. In Illinois the scores that piled up at the top were extreme at scale: internal tracking data released under public-records law showed more than 4,100 children flagged at 90 percent or higher — including 369 children under age 9 assigned a literal 100 percent — far more than any caseload could act on, even as children who died in cases already known to the system were never flagged as top-risk, among them 17-month-old Semaj Crosby, found dead after at least ten DCFS investigations, and 22-month-old Itachi Boyle. The records feeding the nightly scores were themselves degraded: entry errors, a frequent failure to link a child's history to siblings or other adults in the home, and a state-law requirement to erase "unfounded" investigations left the model working from a thinned, fragmentary picture. New director Beverly "B.J." Walker ended the predictive scoring in December 2017, saying it "wasn't predicting any of the bad cases," while keeping a smaller Eckerd case-review training component. Reporting and a later ACLU review placed Illinois among several jurisdictions — Alaska, Louisiana, Ohio, and Oklahoma among them — that took up and then dropped the tool, typically without the scored families ever knowing.
The sociotechnical reading
RSF is the alert-saturation case. A flag that fires for thousands is operationally equivalent to no flag: it cannot change how finite attention is allocated, and it manufactures both alarm fatigue and false reassurance (the un-flagged case reads as cleared). The tool even shipped with an actionability layer — a high score was meant to route a case into a supervisory "second set of eyes" review — but that control only works when high scores are scarce; buried under thousands of them, the review layer had nothing left to triage on. The misfire ran deeper than calibration: the scores were re-derived each night from administrative records that were themselves thinned and fragmentary — unlinked siblings, erased "unfounded" investigations, entry errors — so the loop meant to learn from history instead over-flagged en masse while missing children who carried long real histories. And the decisive corrections came entirely from outside the deploying agency: investigative reporting, a public-records release that put the extreme scores (a literal 100 percent for hundreds of young children) in plain view, and a state inspector-general finding that the program had entered through a no-bid deal reclassified as a grant, quietly bypassing the procurement-transparency that is itself a governance check. The throughput-honesty and provenance patterns (pace the pipeline; risk-tiering that respects real capacity; label what a flag was computed from) are this case's direct descendants.
The concepts used in this reading are defined in the Field Guide; the governance responses live in the Practice Library. The model organization for this case can be stress-tested in the PAN Lab.