Background
Jose's mainnet analysis (Explorer V2 data) shows that leader liveness accounts for 80%+ of consensus failures. The data reveals that specific operators consistently fail their leadership turns, and committee size
amplifies the problem — 7 out of 10 operators show more round changes than a committee of 4 would expect.
This isn't random churn. It's a structural pattern: certain operators are poor leaders, and the protocol keeps assigning them leadership turns at the same rate regardless.
Proposal
Introduce a lightweight reputation mechanism:
- Track leadership failures per operator (round changes triggered as leader)
- After N consecutive/cumulative failures, skip that operator's next leadership turn
- Round-robin rotation continues normally — the skipped operator re-enters the queue and gets their next turn eventually
This preserves eventual fairness (no operator is permanently excluded) while reducing the blast radius of known-bad leaders in the near term.
Properties
- No slashing or staking changes — purely a scheduling adjustment
- Deterministic — all honest operators should reach the same skip decision given the same view
- Self-healing — a skipped operator with a transient issue (restart, network blip) isn't penalized forever
- Backward-compatible — can be introduced as a protocol upgrade
Background
Jose's mainnet analysis (Explorer V2 data) shows that leader liveness accounts for 80%+ of consensus failures. The data reveals that specific operators consistently fail their leadership turns, and committee size
amplifies the problem — 7 out of 10 operators show more round changes than a committee of 4 would expect.
This isn't random churn. It's a structural pattern: certain operators are poor leaders, and the protocol keeps assigning them leadership turns at the same rate regardless.
Proposal
Introduce a lightweight reputation mechanism:
This preserves eventual fairness (no operator is permanently excluded) while reducing the blast radius of known-bad leaders in the near term.
Properties