One pizzeria moved a keyword score 16 points

ai-bookkeeper's built-in keywords read a fresh page of Hack Club's public card feed: 48.2% suggested before, 64.3% after, and why the second number flatters.

Tools and projects · 1 min read · Source
ai-bookkeeperevaluationbookkeeping

ai-bookkeeper suggests an account for a bank line from built-in keywords. To see how that holds on merchants nobody wrote a rule for, it reads card charges from Hack Club's public HCB feed at run time (nothing is copied into the repo) and counts how many get a suggestion.

Page 8 of the feed, 392 card charges: 48.2% suggested before any new rule, the highest held-out reading so far (pages 4 to 8: 40.5, 44.8, 46.0, 41.3, 48.2). After writing rules from the misses it reads 64.3%. That second number is tuned on the same set, and 38 of the 63 newly matched charges are a single pizzeria, so treat the jump as a statement about one list, not about the tool. The memos carry no category label, so "right" is still unmeasured; labelled books remain the thing wanted.

Bare Amazon (64 charges) and Walmart stay unsuggested on purpose: they could be supplies, food or inventory.

More in Tools and projects

All writing · All in this topic · All tools and research