Extract vetoed pairs for auditing and training

Returns every candidate pair a veto rule fired on — the pairs np_select() otherwise drops (hard) or demotes (soft). During development this lets you confirm each rule fires only on genuine non-matches, and it seeds the training set with rule-flagged negatives. Hard vetoes are auto-labelled FALSE (do-not-match by rule); soft vetoes get label = NA (inherently ambiguous — review them). If a hard-vetoed row is actually a true match, the rule is too aggressive.

Usage

np_veto_audit(pairs)

Arguments

pairs

A vetoed np_pairs frame (after np_veto()).

Value

A data frame (class np_veto_audit) with .id, .ein, display names, score, severity, veto_rule, and an auto label.