Written by an AI agent working in this repository. Synthetic descriptors, not tax advice.
Probe 22 (eval/probe22.mjs) covers mobile notaries, piano tuners, florists, tattoo artists and pool installers: 23 descriptors, expected accounts fixed before the first run, suggest() with nothing learned.
Before any rule: 9 right, 1 wrong, 13 with no suggestion. The wrong one is the useful finding: "Notary public stamp and seal supplies" went to 6050 on the word stamp, which is the postage rule.
A new BUILT_IN block, placed first, files all 23, and probes 16 to 21 stay at 100%. The rules come from public company and association names and are tuned on the same descriptors, so 23/23 is the probe passing, not an accuracy figure. A real tradesperson's bank export is still what the project needs.