LATEST · October 21, 2026
A viral photo-scoring video has people drawing conclusions about what women want. Tinder uses actual user behavior to match people. So whose model produced the scores everyone is debating?
READ REPORT →
KEEP READING ↓
FILE 02 · October 20, 2026
Theo says Gemini 4 Argon got more than half its claims about his code wrong in nearly every test. Another coder reports 115 regressions in four hours, followed by four hours of repairs. Google's own overnight demo kept two of six proposed optimizations after review. No matched test here measures the engineer's cleanup time.
READ REPORT →
FILE 03 · September 28, 2026
Three public Claude Code bug reports. Three destructive actions outside the boundary the users thought they'd set. Different models, versions, environments — so this is a pattern report, not one bug behind all three.
READ REPORT →
PUBLIC FILE 04 SMOKE DESK
FILE 04 · September 28, 2026
ChatGPT-authenticated Codex users got 401s naming an API key they said they never configured. The central GitHub issue hit 104 thumbs-up and 96 comments. OpenAI later called it a full Codex outage.
READ REPORT →
FILE 05 · September 23, 2026
Two months of user revolt over Opus 5's long answers, scope creep and overengineering — and now the awkward coda: Anthropic says Opus 5.5 improves writing and communication, in response to one of the most common areas of Opus 5 feedback.
READ REPORT →
FILE 06 · September 23, 2026
Fresh weekly allowance: gone in about 44 minutes. A paused Goal fired off 2,466 more continuation turns. A separate run: 1,545 responses and 242.3M telemetry tokens. Shared cause and billing formula: still unknown.
READ REPORT →
FILE 07 · September 22, 2026
You asked for GPT-6 Astra. The response came back labeled GPT-5.6 Luna. Several public captures show the same swap, and OpenAI hasn't publicly explained the routing contract.
READ REPORT →
THE METHOD Upvotes, complaint threads, bug reports, public traces and release artifacts are leads, not conclusions. We log the pattern, separate the user revolt from the product claim, and keep every source on the table.
← CURRENT PAPER
