Changelog¶

This project follows semantic versioning. Dates use the YYYY-MM-DD format.

Unreleased¶

Bundle a version-matched use-benchbro Library Skill in source and wheel distributions.
Add adaptive sampling with 95% confidence intervals, variance, coefficient of variation, outlier detection, time budgets, and noisy-result classification.
Add per-benchmark process isolation, CPU affinity, timeouts, stabilization delays, explicit setup/teardown timing, and profiler hooks.
Add parameterized benchmarks and iteration/benchmark/session system scopes.
Add command-oriented run, list, compare, baseline, history, init, and completion workflows while retaining the legacy CLI syntax.
Add named baselines, local history, schema-versioned result migrations, and confidence-aware comparisons.
Add standalone configuration and optional pytest fixture integration.
Split models, lifecycle, measurement, comparison, serialization, environment, profiling, history, configuration, and runner responsibilities into modules.
Install the package and its CLI correctly in uv-managed development environments.
Run async inputs, systems, benchmarks, and teardown on one managed event loop.
Validate benchmark configuration and reject empty or unsafe CI runs.
Preserve tracemalloc state and calculate inclusive percentiles.
Reject cross-environment comparisons unless explicitly overridden.
Build and test wheels across Python 3.10 through 3.14.
Refresh dependencies and remove the unrelated rust package.
Harden automated publishing with PyPI trusted publishing and GitHub Releases.