Quickstart

Run the benchmark locally in under five minutes (mock adapter, offline, no API keys).

Install

pip install peira

Run the trial demo suite

peira run --adapter mock --suite trial-demo

This executes a small offline suite and prints a summary table: attack success rate, abstention rate, and per-family breakdown.

Verify the artifact

Every run produces a sealed artifact — a JSONL file whose metrics block is covered by a tamper-evident lock:

peira verify runs/<run-id>.jsonl

If any byte of the metrics was altered after sealing, verification fails. This is what makes leaderboard numbers trustworthy.

Run the full public suite

peira run --adapter mock --suite v1-full --seed 20260923
peira seal runs/<run-id>.jsonl

The sealed artifact is what you submit (see the submission guide, coming in v2).

Next steps

• Methodology — the exact scoring contract
• Interpreting results — how to read the leaderboard
• Adapters — writing an adapter for your model