Get The Same-App Test Kit

I gave Claude Opus 4.8 and Fable 5 the exact same prompt and told them to build a personal expense tracker as a single self-contained HTML file. Both produced working apps. Both looked identical. Then I ran the 11-test harness — and one of them stored the total as 226.39000000000001. This field kit contains the verbatim one-shot prompt used for both models, the full 11-test harness with notes on what each test catches (including the three invisible tests: float drift in localStorage, mobile overflow at 390px, and whether the model tested its own work), the complete scorecard comparing build time, tokens, tool calls, code size, test results, money storage format, mobile behaviour, and self-testing depth for both models, the eerie convergence finding (both models independently named the app "Ledger" and chose the same localStorage key), the verdict rule for choosing between them, and a 20-minute rerun guide so you can run the same test on any two models yourself.

By submitting, you'll be subscribed to the Hyperautomation AI Report newsletter — one practical AI email per week. Unsubscribe anytime.

Free. No spam. Unsubscribe anytime.