Writing & build logs.
Essays on software development, web builds, and AI tooling. Plus build notes from what I'm actually working on right now.
AdSense returned 'Low value content' three times. The problem was the framing.
The third rejection landed on 1 August: same verdict, empty policy field, authorised ads.txt. Three rounds of compliance fixes — hub listings, SVG diagrams, trust-signal cleanup — and the same response each time. A 15-agent root-cause review against the Quality Rater Guidelines made the error plain: Google is issuing a quality judgement, not a policy verdict.
Third rejection, same verdict: the September 2025 QRG had quietly changed what low-value means
Three low-value content rejections, each with a clean ads.txt and an empty policy breakdown. The hub listings were real. The diagrams were in place. A fifteen-agent research review found what three audit cycles had missed: the September 2025 Quality Rater Guidelines had changed the criteria the audits were running against.
Thirteen Days, One Ready Draft: The Pipeline Had No Way to Surface the Silence
Between 30 June and 13 July the ukcalculators drafter processed ten cron cycles and produced exactly one ready draft. The root cause was a state-machine error: `draft.py` treated failed-cluster index entries as permanently consumed. Without a per-cycle ready-draft count, the silence was indistinguishable from thin ground until the backlog had already compounded.
Verification without correction isn't a gap in the pillar cron — it's the design
The 30 June mortgage draft had 10 of 13 sentences clear the sentence-level verifier against live gov.uk pages. The three that failed dropped it from auto-publish. That is not a limitation — it is the design.
161 agents ran the SEO audit. The skeptics killed half the findings before they reached a PR.
An eight-dimension ultracode audit ran 161 agents and roughly 7.7 million tokens against captainrandom.co.uk. The parallelism is a speed feature. The adversarial-verify pass is the quality feature — it killed roughly half the findings before they touched a branch.