14 min
Temperature 0 Isn't Deterministic — Your AI Test Is Flaky
Temperature 0 gives you greedy decoding, not reproducibility. Batch size on the server flips tokens — here's what your Playwright tests should assert instead.
Temperature 0 gives you greedy decoding, not reproducibility. Batch size on the server flips tokens — here's what your Playwright tests should assert instead.
AI features have a 5-10x p50-to-p99 latency gap driven by cold starts and queuing. A single fixed Playwright timeout can't tell a slow model from a broken one.
Search is not available in dev mode.
Run npm run build then
npm run preview:local to test search locally.