Verified Performance
Every number below is backed by a reproducible benchmark or a contractual performance threshold. No marketing estimates.
Core Decision Engine
Pure computation, no I/O. Measured with Criterion on standard hardware.
| Scenario | Threshold |
|---|---|
| Resolve — 1 evidence, 1 policy | < 50 µs |
| Resolve — 100 evidences, 1 policy | < 1 ms |
| Resolve — with conflict detection | < 2 ms |
API Latency (p95)
End-to-end HTTP request including authentication, resolution, persistence, and response serialisation.
| Scenario | Threshold |
|---|---|
POST /decisions/resolve | < 10 ms |
GET /decisions/{id} | < 5 ms |
GET /decisions/{id}/why | < 15 ms |
GET /decisions (paginated) | < 10 ms |
System
Scenario Threshold Cold start (binary → ready) < 2 s Health check response < 1 ms
Methodology
Core benchmarks use Criterion (statistical micro-benchmarking for Rust). API thresholds are contractual performance targets validated under load testing. All benchmarks are reproducible from the source repository.
Regression Policy
Any performance regression exceeding 10% against the main branch baseline is treated as a blocking defect.
Performance is a structural property, not a feature.