Validator benchmarks

Micro-benchmarks for Validator.validate() / validate_batch() over representative contract shapes, with two baselines: a hand-written Pydantic model (what you would write without pycharter) and the Validator's own compiled model called directly (isolates the validate() wrapper overhead from Pydantic itself).

Numbers are machine-dependent — treat relative ratios, not absolute times, as the signal. Shapes differ in field count, so compare each benchmark to the baselines, not to each other.

Results

Benchmark	Median	Mean	Ops/s	Rounds
test_baseline_raw_pydantic	0.69 µs	0.70 µs	1,433,298	83,633
test_baseline_generated_model	0.69 µs	0.70 µs	1,423,971	137,761
test_validate_rule_heavy	0.78 µs	0.79 µs	1,260,107	136,557
test_validate_flat	1.13 µs	1.14 µs	876,162	34,392
test_validate_nested	1.41 µs	1.43 µs	701,315	77,054
test_validate_coercion_heavy	2.30 µs	2.32 µs	430,836	33,732
test_validate_batch_1k_flat	1.08 ms	1.81 ms	553	607

Environment

• CPU: Intel(R) Core(TM) Ultra 7 255H
• Python: 3.13.9
• Captured: 2026-06-07T02:27:38.013453+00:00
• pytest-benchmark: 5.2.3

Regenerate

pytest benchmarks --benchmark-json=/tmp/bench.json
python scripts/render_benchmarks.py \
    --input /tmp/bench.json --output docs/operations/benchmarks.md

Benchmarks live in benchmarks/ — outside the default pytest run and never part of CI gates. A workflow_dispatch GitHub Actions job (benchmarks.yml) runs them on demand and uploads the JSON report as an artifact.