Writing & build logs.
Essays on software development, web builds, and AI tooling. Plus build notes from what I'm actually working on right now.
The GSC and GA crons failed silently for two weeks. That's a telemetry blindspot, not a cron problem.
The nightly GSC and GA telemetry crons for UK Calculators had been dying every night since 20 June. On 4 July a Saturday morning review found the gap — fourteen days of dark data, no alerts, clean-looking logs. The fix took an afternoon. The monitoring lesson took longer.
Branded card names rot. /work's don't any more.
Two corrections arrived within the same hour: one product had rebranded six months earlier, one client project couldn't be named publicly. Both pointed at the same design gap. Closing it required a policy, not a patch.
The cadence clock resets on zero-draft runs. The publishing schedule is manual rescue.
On 17 July the approval gate worked exactly as designed: the first multi-day autonomous streak since the pipeline launched. By 20 July I was writing another manual rescue. The 14-day cadence clock had reset again.
The Pipeline Conflated failed-* with Successful Publishes. Skipped Hit 7.
Ten cron cycles from 30 June to 13 July produced one ready draft and zero auto-publishes while every log showed a clean exit. The mechanism: draft.py treated failed-* index entries identically to successful publishes — permanently consumed. Skipped climbed from 2 to 7 before the root cause surfaced.
AdSense called it low value content. The fix was structural: hub pages, media formats, rate article honesty.
Google rejected the UK Calculators AdSense application on content grounds, not ads configuration. Three structural problems drove the verdict, and fixing all three took three weeks before the re-review went in.