A code-first record of automation pipelines, small web apps, paper trading experiments, and AI-assisted development. Screenshots, commands, and failure notes are the main material.
Defaulting actor isolation to the main actor synthesizes an isolated deinit only for classes that do not write one. The classes that wrote nothing are the dangerous ones — the exact opposite of intuition.
I read five store product page A/B tests on the same day. All five treatments were the identical lever — synthesized captions — and the results spread from -42% to +325%. And my power calculation from days earlier was 9x wrong because it used pre-dedupe impressions.
MediaWiki API booleans are true if the parameter is present, whatever its value. exintro=0 cut article bodies to 292 characters, and verdicts built on those 292 characters permanently blacklisted 43 perfectly good topics.
25 apps were live on the store; my monitor was watching 22. What the missing five had in common wasn't a bug — it was naming. They broke a convention I invented, so they never made it into the candidate list, and nothing ever said so.
I rewrote a new product's gate twice. The first version was a bar my best-performing app couldn't clear. The second compared the wrong unit — a number measured across all traffic sources, used as a search-only threshold. I fixed both while the observation was still zero, so it isn't post-hoc bar moving. It is still twice wrong.
In August I fixed the 404 you get when you list a playlist you just created. Four months later the same spot killed a run with a 409. Same cause, different call: the guard was attached to 'the read', not to 'before propagation finishes' — and both forks were carrying the exact same half.
I hand-filled a missing entry into a canonical list of 46 apps. Two days later I regenerated that file and the entry was gone, replaced by a different app. Still 46. A check that counts will pass a swap like this forever.
A block test that swaps user-agents returns 403 under every condition, because the edge verifies by IP, not by UA. The verdict has to come from the origin access log.
I shipped 227 pages and pre-registered a +30% bar on 28-day sessions. Then the apps I never touched grew 4.5x in two weeks. The threshold wasn't wrong — the comparator was.
My dashboard judges bots by output freshness against an expected interval. When that interval doesn't match the real schedule, every idle day lights up as late.
The bot published successfully every day and every log was green. But a five-format schedule had shrunk to three posts. The cause was a single return at the top of a function.
Each function was correct and the tests passed. The defect was which label sat on which function. In English they barely separate as Stop and End — the collision was born in translation.
Exactly one mode picked its answer on the client, from its own catalog. When the catalog updated the pool changed, and per-device pinning held each of those different answers in place. It wasn't a shareable daily — it was a private puzzle.
MissingTranslation only checks whether the key exists in the locale file. A value that is still the English original passes. Users in five languages were reading half the app in English, and every metric said fully translated.
Even with a viewBox, rasterizers honor the SVG's intrinsic size and pad the rest of the canvas. File size and resolution both look correct, so nothing caught it until I opened the file and looked.