# CAIN HERMES — Benchmarks

`artifacts/hermes/HERMES_BENCHMARKS.json` (reproduce: `python3 scripts/cain42_hermes/bench_hermes.py`). Measured
in-process on the build host; **not a production SLA**.

| Operation | n | p50 | p95 | p99 | mean |
|---|---|---|---|---|---|
| compile (1 rule) | 1000 | 0.232 ms | 0.624 ms | 1.802 ms | 0.311 ms |
| semantic preserve | 2000 | 0.011 ms | 0.021 ms | 0.070 ms | 0.015 ms |
| coverage analyze | 500 | 0.278 ms | 0.623 ms | 1.446 ms | 0.344 ms |
| certificate generate (compile+sign) | 500 | — | — | — | see JSON |
| integrity check | 2000 | 0.007 ms | 0.017 ms | 0.044 ms | 0.010 ms |
| drift detect | 2000 | 0.009 ms | 0.017 ms | 0.046 ms | 0.011 ms |

## Fleet scaling (compile a spec with N rules)

| Rules | Compile time |
|---|---|
| 10 | 2.29 ms |
| 100 | 44.29 ms |
| 1000 | 265.06 ms |

## Separation
All results are **local, in-process** benchmarks. No simulated or distributed results are presented as production.
