Verify
You should not have to trust this site. Here is what can be checked, how, and what cannot be checked yet.
Status
The contracts are written and tested on a local chain. They are not deployed, so there is no on-chain record to verify yet. The commands below have been run only against local test chains.
Who decides what
| Record | Enforced by the contract | Recomputed by the verifier |
|---|---|---|
| Order of steps and population sizes | Yes | Yes |
| Best half survives, ties by lower id | Yes | Yes |
| Randomness round chosen before it is published | Yes | Yes |
| Randomness value matches drand | No | Yes |
| Hourly decisions and market inputs | No | Yes |
| Fitness scores | No | Yes |
| Genome hashes and fossils | Fossil must match its hash | Yes |
Where the contract does not enforce a record, the runner could write a false one. It could not do so unnoticed, because anyone can recompute it from public data.
Check a run
You need the open code, the published manifest and any Elysium RPC. No wallet and no funds.
npm run runner:verify -- --manifest manifest.jsonThis rebuilds each replicate from its seed, the drand beacon and public market candles, then compares every recorded thought, score, genome and fossil. It stops a replicate at its first mismatch and names it.
It also sends one thought per generation to ThoughtVerifier, a separate implementation of the brain in Solidity. Answering takes the EVM 6 to 13 million gas on the test cases, read for free with eth_call. ThoughtVerifier changes no experiment state and has no owner.
Check the wiring
The wiring is built from the paper’s own connectivity matrix.
python3 scripts/build-connectome.pyThe rebuilt files should match the fingerprints below.
- nodes.csv
2f82acd4e076e9a69d052f5316d535afe0b46233fd70439cbe132ac48d2c365a- edges.csv
5ba7d9aecfa6fef82195cbaca1b45a635d5ff2c53494dbc8a5fb7955588b4aba
Corrections
3 November 2026. The first run used a public copy of the wiring in which 192 of 205 rows of outgoing connections carried the wrong cell’s name. We found it by comparing that copy with the paper’s matrix. The wiring was rebuilt from the paper and the run repeated. Results made on the old copy are withdrawn.
What this cannot show
- Verification shows the records are consistent with public inputs. It does not show that the brains learned anything. That is what the statistical test is for.
- The result depends on the market data published by Hyperliquid. The verifier checks the data used against that source, not that the source is right.
- Genesis and connection weights use floating-point functions, which can differ between JavaScript engines. Verify with the Node version named in the protocol.