Hostile Verification Demo

Scenario

An AI system records events using ChainOfFact. Each event is hashed, timestamped, and appended immutably.

Later, the record is examined by a hostile reviewer.

The reviewer does not trust:

• The system operator
• The application owner
• ChainOfFact itself

Verification is performed independently.

Outcome 1 — MATCH

The recorded events are internally consistent.

• Hashes match
• Timeline is continuous
• No conflicts detected

Result: MATCH

Evidence exists and verifies.

Outcome 2 — MISMATCH

One or more recorded events were altered after being recorded.

• Hash comparison fails
• Chain integrity breaks

Result: MISMATCH

Tampering is detected.

ChainOfFact does not speculate about intent or impact.

Outcome 3 — ABSTAIN

Two or more conflicting records exist for the same event.

• Neither record can be proven dominant
• Both pass local validation

Result: ABSTAIN

A collision is detected.

ChainOfFact refuses to resolve the conflict.

Outcome 4 — UNKNOWN

No evidence exists for a claimed event.

• No record was submitted
• No hash exists
• No timestamp exists

Result: UNKNOWN

Absence is preserved explicitly.

ChainOfFact does not infer missing facts.

What This Demonstrates

• Evidence survives hostile review
• Tampering becomes visible
• Conflicts are not resolved narratively
• Missing data is not filled

What ChainOfFact Refuses to Do

• Judge correctness
• Assign blame
• Explain outcomes
• Certify compliance
• Ensure safety

ChainOfFact records facts.

Verification is independent.

Interpretation is external.

Independent Verification

All ChainOfFact evidence bundles include:

• Raw event records
• Hash chains
• Manifests
• Offline verification instructions

Verification does not require:

• ChainOfFact services
• Network access
• Trust in the operator