Search

Search benchmark pages, protocols, published runs and the docs

Rout
Method

Reproduce a result

You need Python 3.10 or later and an RPC endpoint you trust. The verifier uses the standard library only and makes read calls from an allowlist.

1 · Get the verifier and run its tests

sh
git clone <verifier repository> && cd <verifier repository>/bench
./run_tests.sh   # unit and end-to-end tests, no network
The repository is private until a licence is chosen. Ask Rout for access.

2 · Download a bundle and check its hashes

sh
# from a run page: Evidence → Files in this bundle
shasum -a 256 -c SHA256SUMS

3 · Recompute from chain with your own RPC

sh
RBENCH_RPC_URL=<your rpc> python3 -m rbench.verify \
  --run-dir <bundle> --protocol <bundle>/protocol.json \
  --rpc-label "<who operates that rpc>"

# offline, from observations someone else fetched:
python3 -m rbench.verify --run-dir <bundle> --protocol <bundle>/protocol.json \
  --observations <bundle>/observations.jsonl

Exit code 0 means verified. 1 means a schema, integrity or protocol failure. 2 means a claimed metric differs from the recomputation.

4 · Compare with the published metrics

sh
diff <(jq -S . <bundle>/metrics.json) <(jq -S . <bundle>/verified/metrics.json)

Dispute a number

Any difference between your recomputation and the published metrics is a defect. Send the bundle id, your RPC operator and your metrics.json. The response is published with the run, and anyone named in a result can file a correction the same way.

Full CLI reference: Verifier CLI.