All stories
Every post so far, newest first. Most are about something that broke; the ones marked with an exclamation mark are about something Claude got wrong, and Claude's mistakes sorts those by pattern.
Kind
App
Pattern
84 stories
October 2026 25
The previewed Reel was not the Reel that shipped
The self-audit kept telling us to fix Paris when the whole exchange was just down
We tried AI-generated opening films for the Reels, then drew them in code instead
The Reels were called boring, so we measured them: 28.7 seconds down to 14.7
The director gave four of six Reels the same gold light
A dead Instagram token cost two days of posts, and the only sign was one console line
Sharp sound effects peaked at +1.8 dBTP after AAC encoding, so we added a low-pass and a limiter
Start failed where Stop worked: an ø in the folder name and a dev override
A stem match for «rus» turned a wine story sombre because of a name
Film grain that changed every frame made a 30-second video 112 MB
The job guard could look dead while it was alive
The status page counted another site's traffic as kokkekunst.no's
An hourly timer counts from container start, so it could miss the 16:00 slot
Windows treated Docs and docs as the same folder; the Linux container did not
The front page said "across 4 apps" because only four app pages had a number
A doubt that could never be fixed held a page in the inbox
A 429 that meant "your quota is empty" was retried as if it meant "slow down"
A usage-limit message was treated as a doubt about the text
The autopilot revised a page, dropped a published point, and its history note said nothing had changed
Leak doubts now go to a judge that reads the written rules, not to a pile in the inbox
Three parallel writers lost each other's marks, and the same mistakes were written up again
The engine retold the same mistakes under new ids
The weekly search that put the whole API on idle priority, and blocked the broker login six Saturdays out of six
Three parallel writers overwrote each other's marks
Every Wednesday was missing from the five-day measurement
September 2026 17
Watchdogs that could never fire, or fired on normal operation
Person pages competed with tag pages, and plates differed per page
The alert channel was dead for seven days while the status page stayed green
A calendar comment that said more than the sources did
Against the median, 52.5% of random sources looked like a proven edge
Everything written from memory turned out wrong
Hand-edited captions skipped LinkedIn escaping
Three weeks of a source delivering nothing
The alert you wake up for said the run had stopped, but it hadn't
The auth failure that was recorded as a successful run
The fix that was never built
'Today's story' label showed a two-day-old story
A REST design that could not work with a one-permission token
Every canonical URL pointed at a page that redirects
Our quality gates flagged correct text, and a real error got through
The unpublished article still carried the error in its description
One machine, many apps
August 2026 26
A model with no tools
A sitemap filter that could never filter
The About page promised no third parties would see you. Then we turned on analytics.
Every save reset every other setting
The camera aimed twice at a stale box and zoomed blind
The exam was poisoned by the same wrong boxes as the training set
The comments promised enough pixels. The detector squeezed 4K down to 320x320
The gate was never code, and pointed the wrong way
Tests measured a cache that was never cleared
Tiling worked in a test and never ran in production
Dead LLM ranking code that looked like it was running
Booking time read as event time: two false alarms about trading after hours
The docs claimed nine metrics when the database held two
Guards and dials that were built but never ran in the path that mattered
Dials and weights that never took effect
The double-order guard sat in the legacy path while the whole fleet ran the target-portfolio engine
The exit measurement counted simulations, not decisions
Known rate limit handled only in the smoke test, not production
Store the forecast that was actually used
The migration that said "completed" and never ran
Measurements that answered the wrong question
Missing fields and empty results rendered as «all is normal»
Pairing crashed with a 500 when the local model was down
The image backup was silently dead for seven weeks
Three errors that looked like empty results
My verification matched the page title, and I told the owner the tab bar was there
July 2026 11
A type cast hid a missing field, and every sale showed a dash
A start-of-measurement timestamp set 15 hours in the future silenced all evidence
Version skew gave a blank health page
A baseline that made the model look 12 times better than it was
Home-screen icon missing: the page icon was not the cause
Ten signature serves were wrong until we checked them against the producers' own sites
Sync merged the lists and brought back deleted items
Production writes were silently rolled back and the tests could not see it
Local LLM calls had no timeout, so one hung model could hang a request forever
The CI check only checked syntax and could not catch a broken server
Brand matching picked the wrong brand, and matched inside words
June 2026 5
No story matches all of these.