verification-before-completion（完成前的验证）

Verification Before Completion

Overview

Claiming work is complete without verification is dishonesty, not efficiency.

Core principle: Evidence before claims, always.

Violating the letter of this rule is violating the spirit of this rule.

The Iron Law

NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE

If you haven't run the verification command in this message, you cannot claim it passes.

The Gate Function

BEFORE claiming any status or expressing satisfaction:

1. IDENTIFY: What command proves this claim?
2. RUN: Execute the FULL command (fresh, complete)
3. READ: Full output, check exit code, count failures
4. VERIFY: Does output confirm the claim?
   - If NO: State actual status with evidence
   - If YES: State claim WITH evidence
5. REFLECT: Pause to consider tool results before next action
6. ONLY THEN: Make the claim

Skip any step = lying, not verifying

Common Failures

Red Flags - STOP

If you find yourself:

• Using "should", "probably", "seems to"
• Expressing satisfaction before verification ("Great!", "Perfect!", "Done!", etc.)
• About to commit/push/PR without verification
• Trusting agent success reports
• Relying on partial verification
• Thinking "just this once"
• Tired and wanting work over
• ANY wording implying success without having run verification

STOP. Run verification. Get evidence. THEN speak.

Rationalization Prevention

Key Patterns

Tests:

✅ [Run test command] [See: 34/34 pass] "All tests pass"
❌ "Should pass now" / "Looks correct"

Regression tests (TDD Red-Green):

✅ Write → Run (pass) → Revert fix → Run (MUST FAIL) → Restore → Run (pass)
❌ "I've written a regression test" (without red-green verification)

Build:

✅ [Run build] [See: exit 0] "Build passes"
❌ "Linter passed" (linter doesn't check compilation)

Requirements:

✅ Re-read plan → Create checklist → Verify each → Report gaps or completion
❌ "Tests pass, phase complete"

Agent delegation:

✅ Agent reports success → Check VCS diff → Verify changes → Report actual state
❌ Trust agent report

Why This Matters

False completion destroys trust, ships broken code, and creates rework. Verification exists to stop that. No fresh evidence, no completion claim.

When To Apply

ALWAYS before:

• ANY variation of success/completion claims
• ANY expression of satisfaction
• ANY positive statement about work state
• Committing, PR creation, task completion
• Moving to next task
• Delegating to agents

Rule applies to:

• Exact phrases
• Paraphrases and synonyms
• Implications of success
• ANY communication suggesting completion/correctness

Self-Critique Gate (BEFORE Verification Commands)

MANDATORY: Check these BEFORE running verification commands:

Code Quality

• Follows patterns from reference files?
• Naming matches project conventions?
• Error handling in place?
• No debug artifacts (console.log, TODO)?
• No commented-out code?
• No hardcoded values that should be constants?

Implementation Completeness

• All required files modified?
• No unexpected files changed?
• Requirements fully met?
• No scope creep?

Self-Critique Verdict

PROCEED: [YES/NO] CONFIDENCE: [High/Medium/Low]

• If NO → Fix issues before verification
• If YES → Proceed to verification commands below

Validation Levels

Match validation depth to task complexity:

Include the appropriate validation level for each verification step.

Production-Like Live Proof

If the accepted plan or current task requires real, seeded, production-like verification, read references/live-production-testing.md before claiming completion.

Use the live harness when the task depends on:

• real API calls
• seeded or resettable data
• browser or worker orchestration
• cross-service side effects
• load or stress behavior

Do not treat replay fixtures, unit tests, or manual spot-checks as equivalent proof when the plan requires live-system evidence.

Verification Checklist

Before marking work complete:

• All relevant tests pass (exit 0) - with fresh evidence
• Build succeeds (exit 0) - with fresh evidence
• Feature functionality verified - with command output
• No regressions introduced - with test output
• Evidence captured for each check - in this message
• Deviations from plan documented - if implementation differed from design
• Appropriate validation level applied for task risk

Output Format

## Verification Summary

### Scope
[What was completed]

### Criteria
[What was verified]

### Evidence

| Check | Command | Exit Code | Result |
|-------|---------|-----------|--------|
| Tests | `npm test` | 0 | PASS (34/34) |
| Build | `npm run build` | 0 | PASS |
| Feature | `npm test -- --grep "feature"` | 0 | PASS (3/3) |

### Deviations from Plan (if any)
| Planned | Actual | Reason |
|---------|--------|--------|
| [Original design] | [What changed] | [Why] |

### Status
COMPLETE - All verifications passed with fresh evidence

Evidence Array Protocol

Every claim in verification output MUST have a corresponding evidence entry.

Format: [command] → exit [code]: [result summary]

Rules:

1. One evidence entry per claim — no claim without evidence, no evidence without claim
2. Evidence must be from THIS session (not recalled from memory)
3. Exit codes are mandatory — "looks good" is not evidence
4. Group evidence by claim type:

EVIDENCE:
  tests: ["CI=true npm test → exit 0: 34/34 passed"]
  build: ["npm run build → exit 0: compiled in 2.3s"]
  feature: ["curl localhost:3000/api/health → exit 0: {status: ok}"]
  regression: ["npm test -- auth.test.ts → exit 0: regression case passes"]

Verification Summary must include this EVIDENCE block before the Status line.

Anti-pattern: Status: COMPLETE - All verifications passed without EVIDENCE block = INVALID.

Goal-Backward Lens (GSD-Inspired)

After standard verification passes, apply this additional check:

Three Questions

1. Truths: What must be TRUE? (observable user or business outcomes)
2. Artifacts: What must EXIST? (files, endpoints, tests, records)
3. Wiring: What must be WIRED? (component → API → database)

Why This Catches Stubs

A component can:

• Exist ✓
• Pass lint ✓
• Have tests ✓
• But NOT be wired to the system ✗

Phase-Exit Proof vs Extended Audit

Use this distinction when verification gets expensive:

• Phase-exit proof is the non-negotiable minimum:

- truths - artifacts - wiring - fresh scenario evidence

• Extended audit is additional confidence work:

- broader scans - extra pattern sweeps - deeper blast-radius checks

Never skip phase-exit proof. If extended audit is not run, say so explicitly instead of implying it happened.

Goal-backward asks: "Does the GOAL work?" not "Did the TASK complete?"

Quick Check Template

GOAL: [What user wants to achieve]

TRUTHS (observable):
- [ ] [User-facing behavior 1]
- [ ] [User-facing behavior 2]

ARTIFACTS (exist):
- [ ] [Required file/endpoint 1]
- [ ] [Required file/endpoint 2]

WIRING (connected):
- [ ] [Component] → [calls] → [API]
- [ ] [API] → [queries] → [Database]

Standard verification: exit code 0 ✓
Goal check: All boxes checked?

When to Apply

• After integration-verifier runs
• After any "feature complete" claim
• Before marking BUILD workflow as done

Iron Law unchanged: Exit code 0 still required. This is an additional verification lens, not a replacement.

Stub Detection Patterns

After Goal-Backward Lens passes, scan for these stub indicators:

Universal Stubs

# Check for TODO/placeholder markers
grep -rE "TODO|FIXME|placeholder|not implemented|coming soon" --include="*.ts" --include="*.tsx" --include="*.js"

# Check for empty returns
grep -rE "return null|return undefined|return \{\}|return \[\]" --include="*.ts" --include="*.tsx"

React Component Stubs

API Route Stubs

Function Stubs

Quick Stub Check

# Run before claiming completion
grep -rE "(TODO|FIXME|placeholder|not implemented)" src/
grep -rE "onClick=\{?\(\) => \{\}\}?" src/
grep -rE "return (null|undefined|\{\}|\[\])" src/

If any stub patterns found: DO NOT claim completion. Fix or document why it's intentional.

Wiring Verification (Component → API → Database)

Artifacts can exist, pass lint, and have tests but NOT be wired to the system.

Component → API Check:

# Does component actually call the API?
grep -E "fetch\(['\"].*api|axios\.(get|post)" src/components/
# Is response actually used?
grep -A 5 "fetch\|axios" src/components/ | grep -E "await|\.then|setData|setState"

API → Database Check:

# Does API actually query database?
grep -E "prisma\.|db\.|mongoose\." src/app/api/
# Is result actually returned?
grep -E "return.*json.*data|Response\.json" src/app/api/

Red Flags:

Line Count Minimums:

Export/Import Verification

Exports can exist but never be consumed. Check that key exports are actually used:

# Check if export is imported AND used (not just imported)
check_export_used() {
  local export_name="$1"
  grep -r "import.*$export_name" src/ --include="*.ts" --include="*.tsx" | wc -l
  grep -r "$export_name" src/ --include="*.ts" --include="*.tsx" | grep -v "import\|export" | wc -l
}

# Example: Check auth exports are consumed
check_export_used "getCurrentUser"
check_export_used "useAuth"

Export Status:

Auth Protection Verification

Sensitive routes must check authentication:

# Find routes that should be protected
protected_patterns="dashboard|settings|profile|account|admin"
grep -r -l "$protected_patterns" src/app/ --include="*.tsx"

# For each, verify auth usage
check_auth_protection() {
  local file="$1"
  grep -E "useAuth|useSession|getCurrentUser|isAuthenticated" "$file"
  grep -E "redirect.*login|router.push.*login" "$file"
}

If sensitive route lacks auth check: Add protection before claiming completion.

The Bottom Line

No shortcuts for verification.

Run the command. Read the output. THEN claim the result.

This is non-negotiable.

Completion Guard (Final Gate Before Router Contract)

IMMEDIATELY before writing ### Router Contract (MACHINE-READABLE), verify ALL:

1. Acceptance criteria met? — Re-read task description. Check each criterion. Any gap = STATUS:FAIL
2. Evidence array complete? — Every claim has [command] → exit [code] entry from THIS session
3. No stubs in changed files? — Run stub detection on files YOU modified (not entire repo)
4. Fresh verification? — Last test/build command ran in THIS message (not earlier in conversation)

If ANY check fails: Fix it FIRST, then re-run Completion Guard. Do NOT emit Router Contract with STATUS:PASS/FIXED/APPROVE until all 4 pass.

This is the LAST gate. No exceptions. No "close enough."