Interface ReputationLedger

All Known Implementing Classes:
InMemoryReputationLedger

public interface ReputationLedger
Reputation oracle derived from AgentLiabilityRecord history (TNS-298). Surfaces "is this agent trustworthy?" to gating logic (multi-agent orchestrators, payment brokers, scope issuers) without forcing the caller to walk audit records every time.

Implementations

  • InMemoryReputationLedger — derives directly from a LiabilitySink on every query; default, fine for fleets with manageable record volume.
  • External impls cache scores against a TTL or maintain a streaming aggregate; the SPI is narrow on purpose.

Score policy

Each implementation owns its weighting policy — what counts as a "miss" (out-of-scope action? failed outcome? both?) and how LiabilityClass weighs into the aggregate. The framework's default ledger uses simple "pass / fail" counting with class-weighted misses. Custom ledgers swap in domain-specific scoring (insurance underwriters use a different model than internal multi-agent orchestrators).
Since:
0.10.0 (TNS-298)