AI Dev
template

App evaluation plan

Turn the approved PRD into repeatable acceptance, boundary, and production checks with recorded evidence.

PRD version/commit:

Acceptance matrix

IDPRD criterionTest levelFixture/accountExpected resultEvidence
AC-01Human/browser

Required boundary cases

  • ☐ Signed-out user requests a protected page.
  • ☐ User A requests User B's record by a guessed identifier.
  • ☐ Required input is missing or invalid.
  • ☐ Empty state appears before the first record.
  • ☐ External service fails or returns slowly.
  • ☐ Repeated submission does not create unintended duplicates.
  • ☐ Narrow viewport and keyboard route work.

AI feature evaluation, if applicable

  • Input set and source:
  • Expected properties, not preferred wording:
  • Unacceptable outputs:
  • Grounding/source check:
  • Cost/latency to record:
  • Human reviewer:

Do not tune a prompt on one attractive example. Keep a small fixed set of ordinary, edge, and failure cases, then compare changes against the same set.

Test commands

  • Static/check:
  • Unit:
  • Browser/E2E:
  • Production smoke:

Result record

  • Commit:
  • Environment:
  • Date:
  • Passed:
  • Failed:
  • Decision: ship / fix / rollback / narrow scope