Deterministic execution boundary for AI systems.
AI can generate proposals.
Execution requires admissibility.
Binary verdict with an audit-ready evidence pack.
Probabilistic evaluation optimizes likelihood. Execution requires binary admissibility. Current systems treat both as the same inference problem.
A deterministic gate between AI reasoning and real-world execution. Uncertainty equals rejection.
Uncertainty = Rejection · Fail-closed by design
Deterministic admissibility is evaluated by a standards-based oracle (Eurocode / Kirchhoff / mandate bounds). This page shows the evidence pack format and decision boundary with fixed inputs for reproducibility. Live oracle endpoint is operational; demonstrated under NDA.
| Layer | What it validates | Method |
|---|---|---|
| Syntax / Tooling | Allowed operations and patterns | Rule matching, regex, schema |
| Policy / Transaction | Who can act, when, under what authority | Identity, timing, value bounds |
| Reality Layer (FCAL) | Physical and formal admissibility | Eurocode, Kirchhoff, formal limits |
All three layers are required. Only one is missing.