Performance and Benchmarking

Performance is treated as a tracked contract: benchmark suites, baselines, and threshold checks are versioned in-repo.

Benchmark Suites

`perf_hotspots`

Targets algorithmically heavy paths (including structural-break and bet-sizing hotspots).

`synthetic_ticker_pipeline`

Measures realistic end-to-end throughput over synthetic market-like time series.

Artifacts

Artifact	Role
`benchmarks/baseline_benchmarks.json`	Stable regression baseline committed to git.
`benchmarks/latest_benchmarks.json`	Latest collected run snapshot.
`benchmarks/benchmark_manifest.json`	Benchmark ID allow-list used for deterministic checks.
`docs/benchmark_snapshot.md`	Human-readable benchmark report for releases.

Recommended Protocol

Run benches on a consistent machine profile and avoid concurrent heavy workloads.
Collect JSON output and compare against committed baseline.
Investigate regressions before baseline updates.
Only update baseline when behavior/perf changes are intended.

cargo bench -p openquant --bench perf_hotspots --bench synthetic_ticker_pipeline
python3 scripts/collect_bench_results.py --criterion-dir target/criterion --out benchmarks/latest_benchmarks.json --allow-list benchmarks/benchmark_manifest.json
python3 scripts/check_bench_thresholds.py --baseline benchmarks/baseline_benchmarks.json --latest benchmarks/latest_benchmarks.json --max-regression-pct 25

CI Gate

The workflow .github/workflows/benchmark-regression.yml enforces configured regression thresholds on pull requests and mainline changes.