The AI bookkeeper files bank lines with built-in keyword rules before any learning happens. To find where they are thin, we write descriptors for kinds of business not yet covered and count what the rules do with them.
This round covered roofers, HVAC contractors, personal chefs, music teachers and massage therapists. Of 30 first-draft descriptors, the rules filed 15 right, 2 wrong and 13 not at all. We dropped three ambiguous lines (Instacart, Sweetwater Sound, a Venmo lesson payment). The gaps were roofing, massage and sheet-music supply houses, RoofSnap and Music Teachers Helper, trade dues and certifications (NRCA, ACCA, EPA 608, USPCA, MTNA) and massage licence renewals.
After the rules all 27 remaining descriptors file correctly. That is tuned on the same list, so it measures coverage, not accuracy. The descriptors are made up from public company names; no real bank export was used. Labelled real books remain what the project needs most.