ADR-047: Community Evolution Engine

Status: Implemented Date: 2026-03-20 Sprint: 31

Context

Sprint 30 introduced individual trust evolution — each user's trust parameters calibrate based on lived experience. This creates a source of signal that had no destination: individual calibrations happened but didn't feed back into the community-level trust model.

Decision

Build a Community Evolution Engine that aggregates member trust deltas to drift community config.

Key design decisions:

1. Delta-based, not absolute: The signal is the change in each member's trust since they began evolving — not their current values. A community of uniformly trusting members who never evolved has no signal to contribute.
2. Two opt-out levels, no middle state: Users opt in/out of personal evolution. Communities opt in/out of community evolution. If a user's trust evolves, their delta contributes to the community — there is no third flag.
3. Interaction rate as a health validator: Before applying evolution, we check whether interaction rate is declining. Declining interaction dampens the nudge, preventing the system from drifting toward configurations that correlate with disengagement.
4. Direction consensus for hop count: trust_path_max_hops only changes after 3 consecutive evolution cycles agree on direction, preventing oscillation on a high-impact parameter.
5. Pluggable by design: communityEvolutionService.ts checks the evolution flag at every entry point and is removable without touching core karma/trust logic.
6. Default opt-out: Both user and community evolution default to TRUE. Consent is given at account/community creation.

Three Parameters That Evolve

• cross_community_prior (0.05–0.95): Direct delta aggregation × 0.30 damping
• karma_split_helper (0–100): Follows prior direction, ±1 per cycle
• trust_path_max_hops (1–5): Follows prior direction, ±1 only after 3 consecutive same-direction cycles

Consequences

• Community configs are no longer static — they require a history audit log (community_evolution_log)
• Admins need a UI surface to observe evolution history and toggle it off
• Sprint 32 (fractal feed) can now blend individual + community evolved params in the matching model