Rewards And Claims¶
Current rewards are driven by accepted autoresearch outcomes and backend reward snapshots. The old relay SOTA reward-mode docs are archived under AutoML-Zero Archive.
Current Model¶
flowchart LR
Accepted[Accepted submissions] --> Snapshot[Backend reward snapshot]
Snapshot --> Pool[Pool reward publication]
Pool --> Root[Merkle root]
Root --> Contract[Merkle claim contract]
Snapshot --> Weights[Validator weight policy]
Weights --> Chain[Bittensor SN94 weights]The backend reward snapshot is available at:
It may include:
- accepted task outcomes;
- best-result state;
- idea rewards for centerless tasks;
- reward policy;
- validator weight targets.
What Counts¶
For standard and centerless tasks, the validator's observed replay metric is
the canonical result. Claimed metrics are informational.
For peer_evaluation tasks, the coordinator finalizes status through peer
consensus once the threshold is reached.
Only accepted outcomes can feed reward accounting.
Claims¶
Pool consumes backend reward data, builds Merkle leaves, publishes roots, and serves claim proofs. The miner hotkey identifies work and claim lookup. The published recipient coldkey receives the claim.
Claim publication is not instant. A submission can be accepted before a claim package exists.
Validator Weights¶
Validators that opt into backend-directed weights read the backend reward
snapshot and set weights according to reward_policy.validator_weights.
Current intended production allocation:
The hotkey must be resolved dynamically through the live metagraph.
No Guarantee¶
Participation does not guarantee any reward, emission, ranking, validator acceptance, Merkle claimability, or future economic benefit. Read the Disclaimer.