This document captures the benchmark methodology, acceptance gates, and latest measured results for the PrivaDEX DarkPool matching engine.
- Primary artifact:
build/engine_bench_summary.json - Execution path:
benchmarks/engine_bench.cpp - Test gate wiring: CTest benchmark gate (
p99 < 150 ms)
- Workload size: 16-order batch matching path.
- Measurement count:
N=100iterations. - Reported statistics: mean, p95, p99.
- Output contract: machine-readable JSON summary plus exit-code gate.
- Serialization mode in path: SEAL compression with zstd preference when available.
- Primary performance gate:
p99 < 150 ms. - Mean and p95 are tracked for drift and regression visibility.
- Gate is enforced by benchmark exit code in CI/local ctest workflow.
| Metric | Value | Gate | Status |
|---|---|---|---|
| iterations | 100 | fixed window | PASS |
| mean_ms | 38.5274 | informational | PASS |
| p95_ms | 41.502 | informational | PASS |
| p99_ms | 45.164 | < 150 ms |
PASS |
- Current p99 is substantially below gate, leaving meaningful tail-latency headroom.
- The benchmark confirms post-audit hardening did not break latency targets.
- This benchmark is local/off-chain matching performance and does not represent on-chain finality latency.