Evaluation summary

Five held-out exports (1,775 rows) were selected by a frozen rule after three development exports. They are not a probability sample. The post-repair coverage comparison is development evidence, not an untouched test of the new rules.

Export	Rows	Eligible 0.1.4	Unresolved
illinois	256	181	0
indiana	357	55	0
minnesota	735	630	0
wisconsin	368	265	103
mississippi	59	52	0

An independently implemented, same-specification baseline produced identical eligible sets, observations, censoring and grouping on eight exports. This is a reproducibility check, not independent scientific validation. No human expert labels exist.

Full methods, tables and limitations are in the manuscript and supplement. The demo does not produce estimator results.