Peer Scoring

Overview

Peer scoring converts measurable events into a decision signal. A protocol may reward valid message delivery, topic participation, low latency, or long-lived connections, while penalizing invalid messages, duplicate spam, broken promises, or excessive resource consumption. The resulting value can influence Peer Reputation, connection pruning, gossip forwarding, request selection, and the threshold for a Peer Ban.

Scoring is normally local. Two nodes can observe the same peer differently because they receive different traffic, use different parameters, or operate under different network conditions. In pubsub systems, scores may also be topic-specific: a peer can perform well on one message stream and poorly on another. This makes the score useful for local defense but unsuitable as a universal statement about the peer.

Design requires careful resistance to gaming. Attackers may farm positive behavior before launching an attack, exploit score-reset mechanisms by changing identity, or send low-cost valid traffic to offset harmful actions. Honest peers behind congested links can be penalized for latency or delivery failures. Caps, decay, separate penalty categories, and cryptographically verifiable evidence reduce these weaknesses.

Operators should monitor score distributions, threshold crossings, and the reasons behind penalties. Sudden broad score drops may indicate a software incompatibility or network event rather than mass misconduct. Scoring should preserve Node Connectivity and diversity while protecting validation and bandwidth resources. Consensus-critical decisions must still be based on protocol rules, not on whether a message came from a highly scored peer.

Production teams benefit from retaining each scoring signal, weight, cap, decay rule, threshold, and action taken after a score change. The record is especially useful because attackers can game simplistic formulas and honest peers can be punished for congestion or asymmetric connectivity. Safeguards should test score behavior under failure and monitor whether pruning reduces diversity or creates eclipse exposure.