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 cottage bakers, drone pilots, locksmiths, streamers and YouTubers, and locum clinicians. Of 52 descriptors, the rules filed 22 right, 3 wrong and 27 not at all. The wrong ones were a farmers-market booth fee filed as a platform fee, and Elgato and B&H Photo filed as postage. We dropped the last two because an older creator test pins them to that account, and added rules for the rest: drone software and airspace authorisations, locksmith supply houses and dues, clinician licences and recertifications, and ad and locum-agency payouts.
After the rules, all 50 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. One rule (Costco Business Center) had to be narrowed because an earlier probe expects Costco Business Delivery to stay unfiled. Labelled real books are still what the project needs most.