Independent portfolio / synthetic data

Case review quality lab

A clear handoff matters as much as a clear decision.

Explore 16 fictional cases to see how documentation checks surface disagreement, evidence gaps, and missing ownership. These checks support human review; they do not detect abuse or make enforcement decisions.

Whole-dataset measures

Follow-up includes urgent cases. Agreement measures initial reviewer decisions across all cases, including those later adjudicated. Agreement is not accuracy. These fixture counts are not measured workplace results.

Review queue

All descriptions, reviewers, and decisions are fictional. “Documented” means no missing items under this limited rubric; it does not certify a case as safe or correctly decided. Reporting tasks require an authorized human specialist.

Loading cases…

What this leaves to people

The code cannot assess the truth of a report, the quality of evidence, or whether reporting is legally required. It checks structured fields supplied by reviewers. Before operational use, this design would need approved policy, specialist validation, access controls, audit history, and evaluation against independently adjudicated cases.

Created for Clare Jordan with AI assistance in implementation, drafting, and testing. Scope and authorship.