Consensus Update
🚀 Scoring, rewards, and version visibility
We just merged a validator upgrade that tightens miner evaluation and makes reward distribution more fair and transparent.
What changed
- Baseline-delta scoring: miners are now scored on how much they improve over the unmerged base model, instead of using
1 / loss - Top-3 rewards: rewards are now split proportionally across the top 3 miners instead of going all-in on a single winner
- Config cleanup: evaluation settings now live in a dedicated locked
evaluationsection, including a new configurablescore_window - Better fallback behavior: if previous on-chain weights are just a uniform fallback, we now recompute against the current miner set instead of reusing stale values
- Version reporting: validator startup now logs the build version and git SHA, so operators can immediately see what commit/tag their image is running
Why this matters
This should make validator behavior stricter, reward allocation smoother, and ops/debugging a lot easier.