Skip to contents

Stages A-C fail on names that are spelled differently from the BMF rather than absent from it. The NO-tier audit found three recurring, mechanical causes, and this module addresses each. Together they were the single largest identified block of recoverable recall.

Details

Acronyms

NJOP, CSCOE, OMSI, PVCC. The BMF holds the spelled-out legal name, so query and reference share no token at all and blocking never surfaces the pair — 72 of 4,627 queries in the 5k benchmark generated zero candidates from any pass, mostly for this reason.

Truncation

SIGOCPYRPNAM is hard-capped at 35 characters and 6.1% of non-empty values sit exactly at the cap, cut off mid-name (MASSACHUSETTS SOCIETY FOR THE PREVENTION O). A truncated name can never clear an exact pass.

Name typed into the street field

Society of Jesus NY Province and St Ignatius School appear as street values on rows whose name field is blank. Rare, but pure garbage-in that is cheap to catch.