Stage 4 recovery passes

Appended after pf_staged_passes(). Every pass here runs only on the residual the earlier stages could not resolve, so the cost is bounded by how few queries are left rather than by the size of the reference.

Usage

pf_recovery_passes(state = TRUE, max_ref_freq = 5000, min_pair_idf = 8)

Arguments

state

Block these passes on state as well. TRUE keeps them cheap; FALSE is the recall-maximising setting for a small residual.

max_ref_freq, min_pair_idf

Token pruning for the acronym pass.

Value

A list of pass specifications.

On the YES bar

Acronym and prefix matches are structurally weaker evidence than a full-name match — an acronym is a lossy encoding and many organizations share one. They are expected to land mostly in MAYBE unless the address corroborates, which is the correct outcome: npmatch's acronym-aware overlap in npmatch::np_compare() only fires when the address is confirmed, and caps such matches below YES otherwise.