Performance

Kernite uses Python by design for early iteration speed, but optimization decisions should be benchmark-driven.

Benchmark Harness

Run:

uv run python benchmarks/benchmark_execute.py --iterations 20000 --warmup 500

The script measures per-scenario latency and throughput for evaluate_execute without external dependencies.

Measured on February 19, 2026 with Python 3.11.10:

Scenario	Throughput (req/s)	p50 (ms)	p95 (ms)	p99 (ms)
governed_no_matching_policy	74,683.11	0.013209	0.014667	0.017791
governed_missing_field_denied	44,207.49	0.022334	0.025002	0.031842
governed_approved	46,366.11	0.021292	0.024209	0.031425
out_of_scope_approved	72,028.81	0.013375	0.015085	0.022833

These values are environment-specific. Use the same harness in CI/staging/prod-like environments for objective comparison over time.