For Auditors ARCHIVE

For Auditors: 5-Minute Verification Checklist

This page provides a quick verification path for external auditors evaluating MathLedger's governance claims.

5-Minute Verification Checklist

Complete these steps to verify the system's core claims:

Step 1: Open the Hosted Demo (30 seconds)

1. Navigate to /demo/
2. Confirm the demo loads without errors
3. Verify the demo banner shows the latest demo-capability version (currently v0.2.10-demo-reliability). The demo may lag behind the CURRENT documentation version; this is intentional.

Step 2: Run the Boundary Demo (90 seconds)

1. In the demo, click "Boundary Demo: Same Claim, Different Authority"
2. Observe the 4-step sequence:

1. Key insight: Same mathematical truth, different outcomes based on trust class declaration

Step 3: Verify with Self-Test (60 seconds) — Recommended

The verifier has built-in test vectors that work regardless of demo state:

1. Open the Evidence Pack Verifier
2. Click "Run self-test vectors"
3. Confirm the banner shows SELF-TEST PASSED (3 vectors)
4. Observe the results table:

This proves the verification algorithm works correctly and tamper detection functions as designed.

Step 3 (Alternative): Download from Demo

If you want to verify a pack generated by your own demo session:

1. After running the boundary demo, click "Download Evidence Pack"
2. Save the JSON file locally
3. Upload to the Evidence Pack Verifier
4. Click "Verify" and confirm PASS

Ready-to-Verify Examples (No Demo Required)

If the demo is unavailable or you want to verify without running it, use these pre-generated examples:

Download Examples

The file /v0.2.14/evidence-pack/examples.json contains:

Verification Steps

1. Download examples.json
2. Copy the content of examples.valid_boundary_demo.pack to a new file
3. Open the Evidence Pack Verifier
4. Upload the pack JSON
5. Click Verify and observe PASS
6. Repeat with tampered_ht_mismatch.pack and observe FAIL

Why This Matters

These examples prove:

• The verification algorithm is deterministic (same input always produces same hash)
• Tamper detection works (any modification is detected)
• No demo required (auditors can verify offline using the same code)

What This Version Claims

This version (v0.2.14) demonstrates:

• Deterministic canonicalization of reasoning artifacts (RFC 8785-style)
• Cryptographic binding via H_t = SHA256(R_t || U_t)
• Trust class separation where ADV claims are excluded from R_t
• Replay verification using the same code paths as live execution
• Self-auditing capability via the evidence pack verifier

What This Version Does NOT Claim

1. No formal verification — FV (Formally Verified) claims always return ABSTAINED. There is no Lean/Z3/Coq integration.

1. No multi-model consensus — Only one partitioner template is active. The "multi-model arena" is not operational.

1. No learning loop — RFL (Reinforcement from Ledger) is not active. The system does not learn from past decisions.

1. No production hardening — This is a demonstration, not a production system. Do not use for real financial or legal decisions.

1. No guaranteed uptime — The hosted demo may be unavailable. Local execution is always possible.

Demo Reliability Note

The hosted demo runs on shared infrastructure with known limitations:

• "Server lost proposal state" — If you see this error, refresh the page and retry. This is a demo hosting limitation (stateless container restart), not user error.
• Evidence pack download timing — The boundary demo sequence is scripted and does not persist state for download. Use the verifier self-test for reliable verification.
• Intermittent 503 errors — Network tab may show failed requests to analytics endpoints. These do not affect demo functionality.

Key Documentation Links

Field Manual (Why It Exists)

The Field Manual (fm.tex/pdf) is an internal constraint artifact, not marketing documentation.

How auditors should use it:

• Cross-reference claims — If a feature is claimed, check FM for caveats
• Look for "TODO" and "OBLIGATION" — Explicit acknowledgments of gaps
• Compare versions — FM changes show what was addressed between releases
• Trust gaps over features — The gaps we document are more honest than features we claim

Local Verification (Alternative Path)

If the hosted demo is unavailable, verify locally:

git clone https://github.com/helpfuldolphin/mathledger
cd mathledger
git checkout v0.2.14-governance-explanation
uv run python demo/app.py
Open http://localhost:8000

The local demo produces identical evidence packs that can be verified with the same auditor tool.

Audit Closure Records

This project maintains authoritative closure matrices that document all BLOCKING, MAJOR, and MINOR audit findings with evidence of resolution.

Raw audit transcripts are preserved in the repository for completeness, but are not part of the primary evaluator path. Use closure matrices for authoritative status.

Questions?

For technical questions about this verification process, see the Hostile Demo Rehearsal document which addresses common skeptical questions with prepared answers.

Contact

• ismail@mathledger.ai (primary)
• ismail.abdullah.23@cnu.edu (backup)