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 a kind of business not yet covered and count what the rules do with them.
This round covered drone pilots, mobile mechanics, cottage bakers, private tutors and locksmiths. Of 40 first-draft descriptors, the rules filed 21 right, 2 wrong and 17 not at all. We corrected one expectation of our own (an FAA Part 107 exam is training, not a licence fee) and dropped five lines that were invented or ambiguous. The real gaps were cake-decorating, key-blank and locksmith supply houses, trade dues (AUVSI, the tutoring association, the locksmiths' association), a cottage-food licence, a locksmith bond, and tutor-scheduling and drone-logging software.
After the rules, all 35 remaining descriptors file correctly. That is tuned on the same list, so it counts coverage, not accuracy. The descriptors are made up from public company names; no real bank export was used. Labelled real books are still what the project needs most.