The five categories
| Category | Tests | What is validated |
|---|---|---|
| Token validation | 31 | Schema compliance, signature verification, expiry enforcement, delegation attenuation |
| State machine coverage | 28 | All DFA transitions exercised; unreachable-state rejection |
| Physical constraint enforcement | 24 | Velocity, force, geofence, e-stop binding |
| Escalation logic | 22 | All Δ factors; timeout handling; M-of-N quorum |
| Ledger integrity | 52 | Hash chain, TEE attestation, append-only, verification, rollback |
Exit criteria
An implementation is certified SINT-compliant if:≥ 154/157 tests pass (≥ 97.5%)
All 52 ledger integrity tests pass (100% mandatory)
Zero state-machine coverage failures
Running the suite
The benchmark suite
Separate from the conformance suite, seven benchmarks (BM-*) measure performance characteristics. These are not certification gates but ship with every release.
| Benchmark | Code | Measures | Target |
|---|---|---|---|
| Policy Gateway T0 | BM-PG-T0 | T0 decision latency under load | p50 ≤ 1ms |
| Policy Gateway T2 | BM-PG-T2 | T2 decision latency with sponsor notify | p50 ≤ 5ms |
| Policy Gateway T3 | BM-PG-T3 | Human-in-loop round-trip | SLA-configurable |
| End-to-End MCP | BM-E2E-MCP | request → decision → ledger | p99 ≤ 25ms |
| End-to-End ROS 2 | BM-E2E-ROS2 | request → decision → actuation → commit | p99 ≤ 50ms |
| Audit Ledger | BM-AUDIT | Sustainable throughput | ≥ 1,000 events/sec |
| CSML Accuracy | BM-S1-S2 | Spearman ρ vs. ROSClaw | ≥ 0.85 |
BM-PG-* benchmark includes a flood test (10,000 ESCALATE/min). Gateway must maintain T0 p50 ≤ 2ms under flood. This is the protocol’s DoS resilience test.
Hardware target matrix
| Platform | CPU | RAM | Gateway total | Ledger write |
|---|---|---|---|---|
| Jetson Orin NX | ARM Cortex-A78 8-core | 16 GB | ≤ 50ms | ≤ 5ms |
| Intel NUC 13 | Core i7-1365U | 32 GB | ≤ 60ms | ≤ 3ms |
| Raspberry Pi CM4 | Cortex-A72 4-core | 4 GB | ≤ 70ms | ≤ 8ms |
| Mac Mini M4 (dev) | Apple M4 | 16 GB | ≤ 20ms | ≤ 1ms |
Certification bundle
On passing, the suite emits a certification bundle — a JSON document suitable for submission to external auditors, internal compliance teams, or standards bodies. The bundle contains:- Per-invariant evidence (I-T1 through I-G3)
- Per-tier benchmark results with confidence intervals
- Full test log with timestamps
- SHA-256 hash of the test harness at runtime
- Gateway version and commit hash
- Reproducibility metadata (hardware, OS, runtime versions)
docs/reports/certification-bundle-summary.md.