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 pressure washers and window cleaners, handmade jewelry sellers, mobile notaries, sign and vinyl shops, and beekeepers. Of 41 descriptors, the rules filed 17 right, 4 wrong and 20 not at all. Two of the "wrong" ones were our own expectations: Etsy listing fees and Shippo labels were already filed to the fee and shipping accounts the project uses. The real wrong ones were a notary bond filed as legal fees and a craft-show booth fee filed as a platform fee. We fixed those and added rules for supply houses (Rio Grande, Oracal, Mann Lake, Dadant), trade dues, apiary and inspection fees, and signing-service income.
After the rules, all 41 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.