Scoring & Validation

Where This Is Documented

A validator scores miners off-chain and then submits a normalized weight vector via the set_weights extrinsic, bounded by a minimum interval between submissions and a cap on how much weight any one validator can put on any one miner. Yuma Consensus aggregates those vectors on-chain at the end of each epoch, stake-weighted, and discounts validators whose vectors sit far from the consensus — which is what makes honest scoring the profitable strategy rather than merely the virtuous one.

The exact algorithm — bonds, clipping, the trust and consensus terms, the two-pass re-weighting and the numbers it produces — changes with the runtime. Read it at the source:

What You Still Decide

The protocol decides how weights are aggregated. It does not decide what you score, or how. On subnets that leave evaluation open, that is your entire competitive position: the reference datasets you curate, the judge model you run, how you combine quality with latency, and how much you smooth scores across epochs so a single bad epoch does not eliminate an otherwise strong miner. Those choices are covered in Building Reputation.