Writing & build logs.
Essays on software development, web builds, and AI tooling. Plus build notes from what I'm actually working on right now.
Good copy is a reader-side audit. My site failed it twice.
Two copy bugs in the same two-week window. The NAPA SEASON body prose gave Suno the authorship. The /contact/ timelines note read as a pre-rejection to anyone with urgent work. Both made complete sense when I wrote them. Neither survived a cold read.
I wrote gsc-telemetry with no pip dependencies. It found six invisible pages in four days.
Building ga-telemetry and gsc-telemetry as stdlib-only Python skills removed the friction that keeps monitoring tools from running reliably. The daily digest surfaced six pages showing “URL is unknown to Google” four days after deployment. The ultracode audit found no technical blockers. The fix was a crawl path, not a configuration error.
Ten cron cycles, one ready draft — draft.py had no concept of retry
From 30 June to 13 July, ten cron cycles ran clean and the skipped counter climbed from 2 to 7. The root cause: draft.py treated any index entry — including failed-* — as permanent completion. Shipped and failed needed opposite treatment and got the same.
Four rescues in a fortnight: the cadence clock should not own the queue
The morning cron finished with drafted=0: five failure modes, zero publishes, one gate-passing article waiting. I pulled it manually. The word 'rescued' appears four times in thirteen days — that is not a success metric.
One extra JPG, one distance calculation: how a bass-reactive portal made spring and winter share a frame
The temporal portal — a bass-reactive radial hole in the spring layer that reveals a winter panorama underneath — cost one JPEG, one per-pixel distance check, and three lines of bell-curve maths. The seasonal shift was already in the file system.