Stake-weighted median, 10% per-validator cap, 50% quorum, 0.25 outlier distance. Designed for N-validator consensus; SN81 v1 ships a single primary validator while the second is being brought online.
Validator mesh — protocol design · Reliquary
mesh · finney · sn81
Validator mesh
Stake-weighted median, 10 % per-validator cap, 50 % quorum, 0.25 outlier distance (5 % auto-gate). Designed for N-validator consensus; SN81 v1 ships a single primary validator while the second is being brought online.
FIG.03consensus rules · what the aggregator enforces
Stake-weighted median. Per-completion verdicts from each validator are aggregated viastake_weighted_medianover the validators' acceptance scores — not a mean, so a single rogue score can't drag the result.
Stake cap. No validator counts for more than 10 % of the capped mesh total — even a 90 %-stake whale buys exactly the same voice as a 10 %-stake honest node.
Quorum. 50 % of capped mesh stake must be online before any verdict aggregates. Below that, the window stalls rather than letting a minority rubber-stamp a result.
Outlier gate. Validators whose verdict diverges from the median by more than 0.25 in score-dict euclidean distance are flagged (the dashed source in FIG.01). A per-window outlier rate above 5 % auto-gates the validator for 12 windows.
v1 reality
FIG.01 is the protocol DESIGN. SN81 v1 runs a single primary validator, so the cap / quorum / outlier rules are implemented but only become observable signals once the second validator comes online. Until then the dashboard's per-window panel shows the primary validator's verdict bundle as the canonical record.