Writing

Notes from things I built, and the mistakes worth keeping.

• The filter was fine. The adapter had been lying for months. 2026-10-08

  A scheduled job started reporting nothing. The obvious explanation was wrong, and finding that out took one measurement instead of one guess.

• Adding sources would not have helped 2026-10-08

  The funnel was capped by the thing doing the judging, not the thing doing the collecting. One model call per item was the whole problem.

• Eligibility does not belong in a prompt 2026-10-08

  A model read a hard geographic restriction as a note about timezones. Four times. The fix was not a better prompt.