Building thoughtful software with AI.Notes from the systems behind the work.
The live allocator in my ordering system never reads the stock ceiling its README promises, the random tests check five rules but not that one, and a ceiling-aware allocator sits unused in the next folder.
Read moreSwitching long dashboard windows to weekly points went fine for counts and levels, but my handler averaged daily review rates instead of rebuilding them from counts, while the database path did it right. The two paths may also disagree about where a week starts, and short edge weeks and half-versus-half trend arrows add their own distortions.
The dashboard guesses failed changes from pull request titles, and a substring match on the word fix, a loose time window and a mismatched denominator mean the number counts vocabulary as much as breakage.
The per-tool cost chart now reads two typed views instead of parsing every raw log row, the migration that builds those views errors on a leftover instead of skipping it, the documented session join filters by organization on both sides, and the drift check covers more token types but still only warns.
A missing pair of brackets in an AND/OR rule sent every plain accept and reject on my agent dashboard to the automatic pile, while a second copy of the same labels only knew one spelling of a person.
Each AI assistant tool call logs a decision and a result under one event name, so a success rate divided by row count shows a perfect tool at half. At least one tool query was right by luck, I added a result-only filter to all of them, and the team charts from the same change filter on a column their view may not have.
A post-import step meant to backfill analytics views drops and replays each view's stored definition, which restores the trigger but not the rows, touches every org's views, and swallows failures, all to fix a gap nobody ever counted.
The daily charts on my usage dashboard were bucketing days on the database's clock. The change that moved them onto the viewer's clock bound the zone safely in one place, spliced it into the query text in another, and left the range edges at UTC midnight.
Two kinds of agent event shared one normalized name, and a success flag turned into a digit on its way into a text-only store. I made every query name the event kind it wants, wrote the flag as a word, and moved percentages into the queries. No tests yet.
A few of my ClickHouse rollups kept an average next to the total, and the summing engine would have added those averages on every merge; now they keep total and count and divide at read time, and the affected rollup keys now include every label they store.
The first commit of my coding-agent dashboard let any caller send SQL to the database, with only a check on the first word and no login. I take apart the statements that slip past that check, why a perfect read would still have exposed every prompt I typed, and what I'd put there instead.