Benchmark overview

Find current results, evidence definitions, and benchmark commands.

Use these pages in order. Current results lists the latest eligible measurements. Latency evidence maps each claim to its source and timer. Run benchmarks explains how to reproduce a result.

Current results

Read eligible measurements and the scope of each result.

Latency evidence

Check each claim’s status, measurement boundary, and proof.

Run benchmarks

Use the maintained commands to reproduce a result.

How to read the evidence

Current results includes eligible measurements. Latency evidence also records historical, experimental, candidate, estimated, rejected, and unverified claims. Each dated report preserves the full method and links its sanitized artifact.

Reproducibility defines the checks for source revisions, configurations, failures, cleanup, and publication.