Verify Before Claim

NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE.

Usage Template

Prompt

Use verify-before-claim. Before saying this is done, identify the proof command, run it fresh, read the output, and state the actual status with evidence.

Use Case

• Preventing unsupported claims that code, docs, tests, or workflows are fixed, passing, complete, or working.

Expected Result

• The agent reports command output, exit code, pass/fail count, and any remaining unverified risk.

Output Example

• Command: npm test, Exit code: 0, Result: 42 passed, 0 failed, plus residual risks.

Verification Case

• No completion language appears unless fresh evidence is shown in the same response.

Verified Effect

• The agent shifts from confidence-based status claims to evidence-based status reports.

Success Metrics

• A fresh verification command or inspection is run in the current session.
• Output includes command, exit code or observable evidence, result, and residual risk.
• No completion claim is made when evidence is missing, stale, or failing.

When to Use

• Before saying "done", "fixed", "complete", "passing", "working"
• Before committing or creating a PR
• Before expressing satisfaction with results
• When the user asks "is it working?"
• Any time the agent feels the urge to say "it should work"

The Gate Function

BEFORE claiming any status:

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. ONLY THEN: Make the claim

Skip any step = lying, not verifying.

Common Failures

Red Flags — STOP

• Using "should", "probably", "seems to" about completion
• Expressing satisfaction before verification ("Great!", "Perfect!", "Done!")
• About to commit/push/PR without verification
• Trusting agent success reports without independent check
• Relying on partial verification

Expected Value Thinking (From Poker Psychology)

"The biggest bluff is convincing yourself you have control when you don't." — Maria Konnikova

Apply to verification:

The verification mindset:

1. Separate what you CAN verify from what you CAN'T
   - CAN: Code compiles, tests pass, output matches expected
   - CAN'T: Edge cases, production behavior, user experience

2. Focus on EXPECTED VALUE, not single results
   - "Tests pass once" ≠ "Tests will pass reliably"
   - "Works in my environment" ≠ "Works in all environments"

3. Be skeptical of success
   - Good decisions can have bad outcomes
   - Bad decisions can have good outcomes
   - KEY: Evaluate the PROCESS, not just the result

Single-Source Product Claims

Keynotes, launch posts, demos, and roadmap statements can be useful evidence, but they are not enough for operational guidance. When the only source is a vendor statement:

Status: single-source
Source type: keynote / launch post / docs / independent test
Operational risk:
What must be verified next:

Do not turn a demo into a guarantee. Treat availability, scale numbers, benchmark claims, payment protocols, security properties, and adoption figures as unverified until checked.

The Biggest Lesson from Poker:

"Knowing when to fold is more important than knowing when to play."

In verification: Knowing when to say "I'm not sure" is better than claiming "it works" without evidence.

• ANY wording implying success without having run verification

Rationalization Prevention

SOP TDD Context

This skill is the verification component of the TDD-for-SOPs approach. Before writing a new SOP:

1. Define failure: What will the agent do wrong without the SOP?
2. Watch it fail: Verify the baseline error exists (RED)
3. Write the SOP: Address those specific failure modes
4. Watch it pass: Verify the agent now complies (GREEN)
5. Refactor: Close loopholes while maintaining compliance

Quality Gates

• [ ] Verification command identified and run
• [ ] Full output shown (exit code, failure count)
• [ ] Claim only made WITH evidence
• [ ] No "should", "probably", "I think" on completion status
• [ ] Vendor/demo/product claims are marked single-source unless independently verified
• [ ] Delegated user actions have mandate, receipt, and rollback evidence