Skip to content

Evidence over Vibes

Prefer tests, logs, readiness checks, user reports, and git state over “feels done.”

Evidence is anything a skeptical teammate could verify. Vibes are unmeasured taste. Both matter in craft; only evidence closes issues and drains fires.

Which evidence? That depends on the quality regime: deterministic compute (correctness / contracts), interactive products (journeys / a11y / Web Vitals), or generative systems (traces / evals / feedback). The wrong green check is still a vibe.

Where does the oracle live? Prefer the quality trace: DocSlime PRODUCT / REQUIREMENTS / TESTING plus lightweight BDD Given/When/Then scenarios mapped to proof. No scenario and no doc contract → you are guessing.

Agents are fluent at sounding finished. Evidence is how you refuse false peaks — including “all unit tests passed” on a stochastic agent, or “playground looked fine” on a billing pipeline.

  • Name the regime first (Quality regimes)
  • Walk the quality trace: requirement → BDD scenario → evidence map (test-it, check-readiness, DocSlime TESTING.md)
  • Use observe-it when prod is silent — golden signals for services; RUM/vitals for web; LLM traces/scores (e.g. Langfuse) for generative — and feed learning back into experience/requirements
  • Don’t merge on greenwashed CI you didn’t understand

Craft and Harden; practices test / observe / readiness / DocSlime; Quality regimes; Quality trace