An iPhone's ringer switch silences Web Audio but not HTML audio. WebKit's own fix works, and quietly makes getUserMedia throw. Four iOS audio behaviours, measured, with the code.
Read more →We turned on CI for a 64-project monorepo with 10,000 green tests. It failed 13 times in a row - and every single failure was a real, pre-existing defect the fast machines had been hiding.
Read more →I bumped nodemailer two majors to clear a file-read/SSRF advisory, ran install, and the old version stayed put. Two pnpm resolution traps later, the audit was finally at zero — and my CI had been testing a version my users never get.
Read more →Generated social cards carry hard edges and lost spaces you will never spot by eye. Two satori traps with the pixel measurements — plus the third I published that turned out to be wrong.
Read more →My plan to make OG-card rendering portable was to embed TTFs as data-URI @font-face in the SVG. Measurement killed the plan twice — once for real, once because my probe was lying to me.
Read more →A 13-minute pre-push hook, a green test suite, and a push that was already dead before the first test ran. Why git hooks race an idle timeout you don't control — and where the gate actually belongs.
Read more →A Search Console query sweep surfaced four juicy feature gaps on a young tools site. I read the code before building — and three of the four were already shipped and marketed. They ranked low because the site is new, not because the feature was missing. The lesson holds double when an AI agent is the one proposing the work.
Read more →A full PageSpeed sweep of my 81-page site came back with half the pages "broken" at 66–80. It looked like a regression. It was measurement noise — and here's how I proved it before touching a single line.
Read more →A chore chart quietly reassigned a kid's manual override to the wrong task the moment you deleted a row above it — and still called it a deliberate edit. The fix is a small lesson about identity versus position in persisted state.
Read more →