10 to the 23 AI logo

Practice Library

Governance patternstructural

Reconcile copied records

Check a downstream copy against its source before anyone acts on it — so a mistake written once isn't actioned everywhere the copy lands.

What it changes

amplifiedReplicated records reconciled against their source(downstream copies reconciled against source before actioning)

Who can pull it

Deploying organizationHarness builderOversight board

What it looks like institutionally

A record written once and replicated many times spreads its errors at the speed of copying. The documented benefit-automation failures share this shape: determinations were replicated into downstream enforcement systems with no independent reconciliation against the source, and the copies drove action on their own. Michigan's MiDAS actioned replicated fraud flags automatically — garnishment and penalties applied before any human review step — and a single uniform rule set produced tens of thousands of correlated wrongful determinations, one flaw repeating at caseload scale rather than averaging out.

Reconciliation is the missing check: hold a replicated record against its source, and hold blind replication down until the reconciliation clears. Where MiDAS determinations did get human review the error rate fell sharply, which is the same lesson in the enforcement arm — a copy checked against its origin before it drives a penalty is a copy that can be wrong without being catastrophic.

This is the institutional half of the practice; the content-aware gate on a specific replication pathway is its placeable component (see Copy checker).

Addresses: Replicate-without-recheck · Automatic action on unreconciled copies · Correlated error at caseload scale. Test a version of this lever in the PAN Lab.

Deciding whether this lever fits your deployment?

Which patterns matter, and in what order, depends on your system's actual shape. Ranking your options on evidence, with what can backfire stated, is engagement work.

Sources & Evidence

Claims made on this page and what supports them. The full registry lives in Evidence.

EmpiricalA single automated rule set applied uniformly and without human review produced tens of thousands of correlate…

A single automated rule set applied uniformly and without human review produced tens of thousands of correlated wrongful fraud determinations in the documented Michigan MiDAS case — one flaw repeating at caseload scale rather than averaging out.

EmpiricalDocumented benefit-automation failures replicated determinations into downstream systems with no independent r…

Documented benefit-automation failures replicated determinations into downstream systems with no independent reconciliation against the source records — Michigan MiDAS actioned replicated flags and Robodebt reversed the onus onto recipients.

EmpiricalDocumented enforcement systems actioned replicated flags automatically — garnishment and penalties applied bef…

Documented enforcement systems actioned replicated flags automatically — garnishment and penalties applied before any human review step in the recorded MiDAS deployment.

EmpiricalIn the documented MiDAS case, error among no-review auto-adjudications ran roughly 93%, and determinations err…

In the documented MiDAS case, error among no-review auto-adjudications ran roughly 93%, and determinations erred at about 85% without human review versus 44% with it.

aiincidentdatabaseGroundingInvestigativeSave

AI Incident Database, Incident 373 (MiDAS false fraud claims) https://incidentdatabase.ai/cite/373/

https://incidentdatabase.ai/cite/373/

Grounds: model org: michigan_midas