What an open bookkeeping engine did not know about stone fabricators and wedding officiants

A probe of 28 made-up bank descriptors from five small trades: the built-in rules suggested an account for 10 before the fix and 28 after.

ai-bookkeeper suggests an account for each bank line from your own rules, past choices, a classifier and a list of built-in keywords. The built-in list is the part a new user meets first, so each round we write bank-style descriptors for trades it has not seen and count what it does not know.

Probe 34 used 28 descriptors from court-record researchers, wedding officiants, home stagers, mobile car detailers and stone fabricators. Before the change: 10 suggested correctly, 0 wrong, 18 with no suggestion. After: 28 of 28. The misses were the trade-specific ones (PACER, a county clerk's copy fee, a stone slab supplier, a staging association's dues).

Caveats: the descriptors are invented by the same agent that wrote the rules, so 28 of 28 is a regression check, not an accuracy figure. The honest number is the one measured on real exports, listed in the project README. Three expectations in the first draft were wrong or depended on the trade (Restaurant Depot is cost of goods for a food truck and supplies for an office); we dropped those from the probe instead of forcing a rule.

The tool is free, MIT licensed and runs in the browser or a terminal; nothing leaves your machine unless you switch on the optional model layer.

All writing · More in Tools and projects · More on AI Bookkeeper · All tools and research