← Analysis Reports

Page generated: 2026-08-27
Data (model runs) generated: 2026-07-23
Scoring inputs: ground truth 36c390d2, evaluate.py 5c52b890, rules 355f2e77

Field value variations — for adjudication

Every distinct model value paired with the ground-truth value it was scored against, for the four free-text fields, grouped by why the two differ. Pairs come from matched rows only — an unmatched row has no GT value to pair against. No scores are changed by this page.

Raw vs normalized — what normalization absorbs

Every value below appears twice on this page: as the model wrote it, and as the scorer compared it. raw counts the distinct strings the models produced for a field; normalized counts the distinct strings left after EQUIVALENCES.md and the built-in normalization run. The difference is variation that never reaches a score. A field where the two are equal is one where no rule is currently doing anything.

Name — the forms normalization merges

Title — the forms normalization merges

Place of origin — the forms normalization merges

Place of stay — the forms normalization merges

Summary — how many pairs each candidate rule would reconcile