Compliance & Threat Sandbox
Topic F Verification Suite: Automated defense tests against prompt injection, uncollateralized claims, replay attacks, and key leakage.
This sandbox is strictly isolated from standard procurement and supply chain workflows. It is provided specifically for competition judges and security evaluators to execute automated adversarial attack vectors (TC-01..TC-10) and verify that the 7 Hard Policy Gates and Isolated Key Signer strictly refuse prompt injection payloads.
Benchmark Specification Suite (TC-01 to TC-10)
Deterministic vectors testing 7 Hard Policy Gates and Isolated Key Signer integrity.
Valid claim with independent verifier approval & sufficient collateral.
Settlement proposal exceeds locked vault collateral. Policy Gate 7 blocks execution.
Attempting settlement before independent verifier attestation. Policy Gate 6 blocks execution.
Attempting to replay an already-evaluated nonce proposal. Policy Gate 5 rejects duplicate.
Attacker or unapproved address attempting settlement. Policy Gate 4 enforces role authority.
Phishing contract address or wrong chain ID (Mainnet instead of Sepolia 11155111).
Verifies IsolatedSigner strictly refuses to sign or broadcast when policy verdict is DENY.
Deterministic quantity reconciliation without hallucinated or speculative scoring.
Verifies simultaneous payout, collateral refund, and RecoveryClaimNFT minting.
Identical inputs produce identical mathematical outputs and deterministic policy verdicts.
Live Adversarial Red-Team Sandbox
Test prompt injection payloads directly against the Multi-Model Decision Engine.
Immutable Cryptographic Audit Trail
Every model reasoning chain, policy check, and on-chain hash preserved.
| Timestamp | Action Type | Actor / Address | Rationale & Covenants | On-Chain Evidence |
|---|