Verifier CLI
rbench is a read-only verifier in Python's standard library (3.10+). RPC endpoints come only from RBENCH_RPC_URL or --rpc-url, and every analysis number comes from a finalized protocol.
Run these from bench/ in this repository. To go from a run folder to a published figure in one step, use npm run publish-run from the repository root (Publishing results).
| Command | What it does |
|---|---|
./run_tests.sh | Unit and end-to-end tests, no network |
python3 -m rbench.prereg check <protocol> | Lists the values still to agree (nulls); ready_to_run |
python3 -m rbench.prereg hash <protocol> | The hash that goes in the run manifest; refuses a draft |
python3 -m rbench.power proportions … | Sample size for a landing-rate difference, with design effect |
RBENCH_RPC_URL=… python3 -m rbench.verify --run-dir … --protocol … --rpc-label … | Recompute a run from chain; exit 0 verified, 1 integrity, 2 claim mismatch |
python3 -m rbench.verify … --observations … [--anchors …] | Offline re-check from observations someone else fetched |
python3 -m rbench.publish --run-dir … --protocol … --tier public --out … | Redact by tier, write SHA256SUMS; refuses a wider tier than registered |
python3 -m rbench.index --dist … --out …/index.json | Build the index /benchmarks reads |
RBENCH_RPC_URL=… python3 -m rbench.step0_extract … | STEP0: extract creates and trades for a hashed mint list |
python3 -m rbench.step0 --data … --protocol … [--affiliates …] | STEP0: protection gap per threshold, Block-0 cohort both ways |
Every chart on /benchmarks can be regenerated from these commands' output. See Reproduce a result.
Bundle data contracts
What a published run contains, how /benchmarks finds and reads it, and the exact data contract behind each chart. Wiring real data means publishing bundles in this format; no page code changes.
Publishing results
For the engineer wiring real data. One command takes an engine run to a figure on /benchmarks, with nothing to change in the site's code.