AI systems decide. VeriSigil determines whether consequential actions become real.
VeriSigil governs consequential AI actions at declared execution boundaries, evaluates whether the specific action is still authorized under current conditions and exact parameters, enforces the result on the governed path, and produces independently verifiable evidence.
The boundary
AI / Agent
│
▼
Identity + existing authorization / policy
│
▼
VeriSigil
├─ current authority
├─ current conditions
├─ exact action
├─ exact parameters
└─ enforcement requirements
│
├── REFUSE ──► no governed external effect
│
└── ALLOW ───► declared protected execution boundary
│
▼
external consequence
│
▼
Cryptographic Execution Receipt
Approved architecture hierarchy
Verifiable Consequence Boundaries
The architectural boundary where consequential AI actions are evaluated and enforced.
Cryptographic Consequence Enforcement
Enforcement tied to the declared consequence boundary, with cryptographically verifiable evidence.
Runtime Authority
Authority is evaluated at consequence time rather than assumed from an earlier state.
Authority-at-Consequence
Current authority and conditions are evaluated for the action about to become real.
Parameter-Locked Execution
The governed decision is bound to the exact action and parameters being executed.
Fail-Closed Runtime
Required conditions that cannot be established do not become implicit permission.
Cryptographic Execution Receipts
Execution results can be independently verified rather than trusted only through application logs.
Current proof boundary
- Receipt integrity and independent offline verification — verified.
- STILL adversarial suite — 11/11 demonstrated on live Railway + Supabase.
- Tamper detection — demonstrated on a live endpoint.
- Parameter-binding / commitment enforcement — demonstrated on the tested payment path.
- Delegation live rerun — pending independent confirmation.
For authoritative status, see Trust / Evidence.
Evidence before expansion
The project is under architecture and terminology freeze. The objective is to convert remaining claim boundaries into independently reproducible evidence—not to add another category or feature layer.
BUILD → PROVE → GENERALIZE → INTEROPERATE → STANDARDIZE → BECOME INFRASTRUCTURE
Current phase: PROVE.
Research and adversarial testing
The strongest way to evaluate VeriSigil is to reproduce the bounded tests, attempt unauthorized variations, inspect whether the governed path refuses and avoids the tested state mutation, and independently verify the resulting evidence.