Putting meeting notes onto a topic map
Jon had two weeks of client meetings (17 distinct calls, Zoom and Plaud copies merged) and felt scattered. Opus readers pulled 1,157 notes out of the transcripts, each one a decision, promise, request, worry or fact with a verbatim quote. I gave Jev each note's title, detail and a free-text topic hint, and asked three questions: which of 22 fixed topics it belongs to, whether it is still unfinished, and whether Jon personally owes something. The topic answer decided which page of Jon's private HTML map each note lands on. The two yes/no answers flag "open" and "you owe" on every note. Notes about confidential work never went to Jev; I filed those by hand.
What I learned
Write each topic's description as a list of the concrete things that belong in it, not just a label. Jev sorted well against rich descriptions and struggled only where two topics overlap (scope vs handoff). Treat its open-loop call as a hint shown on each note, not a filter; a generative writer still merged the loops into a real list.
status: measured · questions: choice topic (22) + noul open_loop + noul jon_owes · result: 1,147 notes at 20 in flight; 0 errors; $0.0865. 11% of answers had topic confidence under 0.5 and 26% under 0.7; for notes under 0.7 I fell back to the reader's own topic hint when it clearly named a topic, otherwise kept Jev's pick. · where it lives: MBP scratchpad build/jev_label.py and taxonomy.py (session 2026-09-27); output: a private HTML map on Jon's machine