Backend Reward Snapshot¶

The current production validator path uses backend-directed reward snapshots. The old capacitorless, relay SOTA, and local winner modes are archived under AutoML-Zero Archive.

Snapshot Endpoint¶

GET https://autoresearch.bitsota.com/api/v1/reward-snapshot

The snapshot is the coordination point between:

accepted autoresearch submissions;
Pool reward publication;
Merkle claim packages;
validator weight setting.

Validator Weight Policy¶

Validators should read:

{
  "reward_policy": {
    "validator_weights": {
      "mode": "targets",
      "targets": []
    }
  }
}

Supported backend modes in the client are:

Mode	Meaning
`local`	Do not override local behavior. Not the production target.
`burn_uid0`	Set all weight to UID `0`.
`targets`	Set explicit UID/hotkey targets from the backend.

Production should use explicit targets and resolve hotkeys dynamically.

Safety Rules¶

Run only one weight-owning process per validator hotkey.
Do not run legacy relay/local SOTA weight setters in production.
Do not hardcode the contract UID; resolve the contract hotkey each run.
Dry-run the backend weight setter before enabling live weight submission.

See Validator Guide and Public Autoresearch Validator Runner.