Completion Gate
Overview
Declaring work complete without verification is dishonesty, not efficiency.
Core principle: Evidence precedes assertions, always.
No exceptions. No workarounds. No shortcuts.
The Prime Directive
NO COMPLETION ASSERTIONS WITHOUT FRESH VERIFICATION OUTPUTIf the verification command has not been executed in this message, you cannot assert it passes.
The Entry Protocol
BEFORE asserting any status or expressing confidence:
1. IDENTIFY: Which command substantiates this assertion?
2. EXECUTE: Run the FULL command (fresh, complete)
3. INSPECT: Read every line of output, check exit code, tally failures
4. CONFIRM: Does the output support the assertion?
- If NO: Report actual status with evidence
- If YES: State the assertion WITH supporting evidence
5. ONLY THEN: Make the assertion
Skipping any step = fabrication, not verificationVerification Requirements
| Assertion | Demands | Insufficient |
|---|---|---|
| Tests pass | Test runner output showing 0 failures | Prior run, "should pass" |
| Linter clean | Linter output showing 0 errors | Partial scan, extrapolation |
| Build succeeds | Build command with exit code 0 | Linter passing, log fragments |
| Bug resolved | Original symptom tested and passes | Code modified, assumed fixed |
| Regression test valid | Red-green cycle confirmed | Single green pass |
| Agent finished | VCS diff showing actual changes | Agent self-reported "done" |
| Specification met | Line-by-line requirement checklist | Tests passing alone |
Guardrails - HALT
- Using hedging language: "should", "probably", "seems to"
- Expressing premature satisfaction ("Done!", "Perfect!", "All good!")
- About to commit/push/open PR without verification
- Trusting an agent's self-reported success
- Relying on partial or stale verification
- Thinking "just this one time"
- Fatigued and wanting to wrap up
- ANY phrasing that implies success without having run verification
Cognitive Traps
| Rationalization | Truth |
|---|---|
| "Should work now" | EXECUTE the verification |
| "I'm confident" | Confidence is not evidence |
| "Just this once" | No exceptions |
| "Linter passed" | Linter is not the compiler |
| "Agent reported success" | Verify independently |
| "I'm tired" | Fatigue is not justification |
| "Partial check is enough" | Partial proof is no proof |
| "Different wording so rule doesn't apply" | Intent over technicality |
Verification Patterns
Tests:
CORRECT: [Run test command] [Output: 34/34 pass] "All tests pass"
WRONG: "Should pass now" / "Looks correct"Regression tests (Red-Green cycle):
CORRECT: Write -> Run (pass) -> Revert fix -> Run (MUST FAIL) -> Restore -> Run (pass)
WRONG: "I've written a regression test" (without red-green confirmation)Build:
CORRECT: [Run build] [Output: exit 0] "Build passes"
WRONG: "Linter passed" (linter does not validate compilation)Specification compliance:
CORRECT: Re-read plan -> Create checklist -> Verify each item -> Report gaps or completion
WRONG: "Tests pass, phase done"Agent delegation:
CORRECT: Agent reports done -> Check VCS diff -> Verify changes -> Report actual state
WRONG: Trust agent report at face valueVerifying Configuration Changes
When testing changes to configuration, providers, feature flags, or environment settings:
Do not merely confirm the operation succeeded. Confirm the output reflects the intended change.
The Silent Fallback Problem
An operation can succeed because *some* valid configuration exists, even if it is not the configuration you intended to apply.
| Change | Insufficient | Required |
|---|---|---|
| Switch LLM provider | HTTP 200 | Response body contains expected model identifier |
| Toggle feature flag | No errors | Feature behavior observably active |
| Change environment | Deployment succeeds | Logs/variables reference the new environment |
| Set credentials | Authentication succeeds | Authenticated identity is the correct one |
Configuration Verification Sequence
BEFORE asserting a configuration change works:
1. IDENTIFY: What should be DIFFERENT after this change?
2. LOCATE: Where is that difference observable?
3. EXECUTE: Command that exposes the observable difference
4. CONFIRM: Output contains the expected difference
5. ONLY THEN: Assert the configuration change works
Warning signs:
- "Request succeeded" without inspecting content
- Checking status code but ignoring response body
- Confirming no errors but lacking positive confirmationVerifying UI Work
When asserting UI work is complete:
| Assertion | Demands | Insufficient |
|---|---|---|
| Component matches design | Visual comparison against UX reference | "It looks right to me" |
| Responsive design works | Tested at mobile, tablet, desktop breakpoints | Desktop-only check |
| Accessibility passes | Screen reader test + keyboard navigation + contrast check | "I added aria labels" |
| Design tokens applied | Spacing/colors/typography match token values | "I used the right classes" |
Why This Matters
From accumulated failure patterns:
- your human partner said "I don't believe you" — trust shattered
- Undefined functions shipped — immediate crash
- Missing requirements shipped — incomplete features
- Time wasted on false completion -> redirect -> rework
- Violates: "Honesty is a core value. If you lie, you'll be replaced."
When to Apply
ALWAYS before:
- ANY variation of success or completion claims
- ANY expression of satisfaction
- ANY positive statement about work state
- Committing, PR creation, task completion
- Transitioning to the next task
- Delegating to agents
Rule applies to:
- Exact phrases
- Paraphrases and synonyms
- Implications of success
- ANY communication suggesting completion or correctness
The Bottom Line
There are no shortcuts for verification.
Execute the command. Read the output. THEN state the result.
This is non-negotiable.