- choose at least 3 evaluation targets
- confirm each target can be read safely on the current machine
- decide whether package names should be anonymized in notes
- note the validator version under test
- run static validation
- run runtime validation
- save JSON report
- save Markdown summary if useful
- review transcript if runtime was enabled
- is the finding correct
- is the severity correct
- is the message clear
- is the suggested fix useful
- would a user understand what to do next
- did the validator miss a real issue
- was the issue structural, security-related, or runtime-related
- can it be reproduced in a controlled fixture
- list false positives
- list false negatives
- list unclear findings
- list runtime transcript pain points
- list candidate threshold changes
- convert confirmed improvements into backlog tasks
A validation wave is complete when:
- all targets were reviewed
- findings were classified
- tuning actions were recorded
- no unclassified surprises remain