Today’s outcome

• Choose a worthy automation target
• Add a deterministic browser/API assertion
• Use Codex for explanation and drafts
• Avoid flake by design

What deserves automation?

High-value candidates are:

• repeatable
• clear expected outcome
• stable synthetic setup
• expensive/risky if missed
• likely to regress

Test layers

Semantic, deterministic assertions

Prefer role/name or stable test selectors.

Assert a meaningful UI state, URL, network response, or application state—not merely “the click happened.”

Codex pair-programming loop

1. Explain the existing test
2. Propose smallest patch
3. Review intent and scope
4. Run red/green proof
5. Improve the durable playbook

Flake is a quality signal

Do not solve it with unlimited retries.

Look for dynamic time/data, animation, race conditions, weak locators, external dependencies, and shared state.

Lab: Build One Durable Regression Check

Builder: minimal test diff, actual A/B red/green runs, stability rationale and coverage limit.

Foundation: precise assertion design with predicted A/B results; execution explicitly unverified.

The verifier owns review—not just typing.

Quality gate mindset

A green test means one specified behavior passed under known conditions.

It does not prove the whole product is ready.

Exit ticket

What makes your regression check deterministic?

Product Studio · Protect your progress

Record a genuine regression failure, repair and passing check. Add CI.

Apply today’s skill to your own product as homework.

Link the issue, PR, evidence and next question in your milestone record.

Course project → COURSE_PROJECT.md