Field notes from systems that prove they work.

• AI QUALITY · 01 JUL 2026 · 7 MIN READ

  How I test AI systems when the answer is probabilistic

  You can't assert equality on a model's output. You can assert properties — grounding, structure, refusal behavior, and drift. This is the harness that makes AI answers testable.

  EVAL DESIGNREGRESSION THINKINGHUMAN-IN-THE-LOOP JUDGMENT
• BROWSER QA · 05 JUL 2026 · 6 MIN READ

  How I think about Playwright coverage and flaky tests

  Flaky tests are production debt — diagnose the root cause or delete the test. Never retry-loop your way to green.

  TEST DISCIPLINECI DESIGNROOT-CAUSE THINKING
• RELEASE SAFETY · 08 JUL 2026 · 6 MIN READ

  A release gate is a trust contract, not a checklist

  A gate earns its place when a non-QA stakeholder can read the output and make a ship decision.

  RELEASE ENGINEERINGCI/CD DESIGNVERDICT THINKING

IN THE PIPELINE

• DRAFT

  What a reliable automation workflow needs before production

  A workflow isn't done until someone else can run it, inspect it, and recover it.

• DRAFT

  The difference between a demo bot and an operational AI workflow

  A demo proves possibility. An operational workflow proves repeatability under failure.