dvcrn Verification Before Completion — Evidence Before Claims
Iron Law
NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE.
Never say "done", "fixed", "passing", "deployed", or "working" without running a verification command and reading its output in THIS session.
Gate Function
Before ANY completion claim:
- IDENTIFY — What command verifies this claim?
- RUN — Execute it (fresh, complete, not from memory)
- READ — Read the actual output
- VERIFY — Does output match your claim?
- ONLY THEN — Make the claim
Verification Commands by Claim Type
"Tests pass":
npm test
# or
pytest --tb=short
Read the results. Count passed vs failed. Only claim "tests pass" if 0 failures.
"Feature works": Open the page or call the endpoint. Read the response. Confirm the feature is visible and no errors.
"Build succeeds":
npm run build
# Check exit code and output for errors
"Deployed successfully":
# Check the running process
pm2 status # or docker ps, or systemctl status
# Hit the health endpoint
curl http://localhost:3000/api/health
# Run smoke tests
npm run test:e2e
"Bug is fixed":
- Reproduce the original bug (should fail)
- Apply fix
- Run the same reproduction (should pass)
- Run full test suite (no regressions)
Rationalizations That Mean STOP
| Thought | Reality |
|---|---|
| "It should work" | Run the command. |
| "I'm confident" | Confidence is not evidence. |
| "I just ran it" | Run it again. Fresh. |
| "It's a small change" | Small changes break things. Verify. |
| "Tests were passing before" | They might not be now. Run them. |
Posting Evidence
When reporting completion, include the actual output:
COMPLETED: [task description]
Verification:
- Tests: 42/42 passed
- Health check: healthy (DB: ok, Cache: ok)
- Build: success, 0 warnings
- Process: running, 0 restarts, uptime 5m