Break Your Assumptions v0.1.0 - It works. Until it doesn't.

It works. Until it doesn't.

First public release: six complete, deterministic, free, dependency-free experiments.

• Payment déjà vu, The last ticket, The lying clock, The invisible difference, Offline twins, The perfect backup.
• Each lab has a prediction, bounded conditions (with no-failure controls where meaningful), an original failure, an invariant, a chronological trace, a repair, a plain-language explanation, explicit model limits, inspectable source, and a standalone Node.js test download.
• Runs by opening index.html directly; also works from a local server and under a GitHub Pages repository subpath. No paid service, login, analytics, or third-party asset.

Changes from the unpublished starter build

• Simulations now reject out-of-range or wrongly typed conditions instead of running unbounded.
• Exported tests cover every offered condition (invariant tests plus negative controls), state plainly that they verify the bundled reference until an adapter is connected, and are proven to fail when an adapter keeps the original bug.
• Changing a condition now re-runs the version you last ran (original or repaired) instead of silently switching back to the original.
• The ticket trace now shows the stale read that causes the oversell.
• Results are announced through a concise live region rather than re-announcing the whole log; progress and reset are announced too.
• Landmark, ARIA, and scrollable-region fixes; minimum text size raised to 11 px.
• Header and footer documentation links point to the GitHub repository instead of raw Markdown files.
• A continuous-integration failure on Node.js 24 (a test that only understood Node.js 22's TAP output) was found by the first Actions run and fixed without weakening any assertion.
• Example test files are generated by npm run examples and verified to match the exporter.

Verification summary

See docs/RELEASE_CHECKS.md for exactly what was run and what was not.

Limits

These are teaching models, not production-grade reliability tooling. A downloaded test checks your application only after you connect an adapter. Only Chromium was used for browser checks.

Licenses

Code: MIT. Original prose: CC BY 4.0.