Docs·4ff474d·Updated Mar 14, 2026·43 ADRs
Back
ADR-036accepted

ADR-036: Private Feedback Model — Trust Without Rating Economies

ADR-036: Private Feedback Model — Trust Without Rating Economies

Date: 2026-02-25 Status: Accepted Deciders: Development Team

Context

After implementing the trust score formula (50 + karma_contribution + feedback_contribution), we needed a mechanism for users to submit quality feedback after a completed interaction. The feedback.feedback database table and avg_feedback_score field in reputation.trust_scores were already designed for this.

The key design question was: should feedback ratings be visible to other users?

The Social Gossip Problem

In human societies, peer reputation forms through quiet word-of-mouth — not public scorecards. When people whisper "she's reliable" or "he follows through," that reputation propagates through trust networks organically. No one publishes a leaderboard.

Public rating systems disrupt this natural dynamic in several ways:

1. Performative behavior: When ratings are visible, people optimize for the score rather than the interaction. The goal becomes "get 5 stars" rather than "be genuinely helpful." This distorts the signal we're trying to measure.

2. Asymmetric power dynamics: A negative public review can damage someone disproportionately — especially in small communities where the population is limited and reputation is everything. A helper who gives a bad experience once shouldn't carry a public 1-star for years.

3. Rating chilling effect: Users avoid giving honest negative feedback out of fear of retaliation ("I gave them 2 stars, now they'll give me 2 stars"). This compresses all feedback toward 5 stars, making the signal useless.

4. Rating economy emerges: Public scores create a secondary currency. Users start tracking their scores, comparing scores, and feeling anxiety around them. This is the Uber/Airbnb problem — the rating number becomes an identity artifact divorced from actual helpfulness.

5. Gaming and reciprocal inflation: When ratings affect visibility or status, communities develop implicit norms of reciprocal 5-star exchange that defeat the measurement purpose entirely.

Decision

Feedback ratings are private inputs to the trust score algorithm. They are never exposed via API or displayed in any UI.

  • Users submit a 1–5 star rating after each completed interaction
  • The rating feeds into avg_feedback_score in the trust formula
  • The trust score itself influences feed curation and trust distance filtering
  • No user can ever see another user's rating or their own average feedback score
  • The feedback endpoint (POST /reputation/feedback) is authenticated but the aggregated result is never returned to clients

The trust score (50–100) is surfaced in the trust path display, but it's presented as a holistic "trust distance" indicator — not decomposed into "this person's average rating is 3.7 stars."

Analogy

Credit scores work the same way: lenders use them to make decisions, but individuals don't see each other's credit scores. This enables honest assessment without social comparison or anxiety.

Consequences

Positive

  • Honest signal: Users can rate honestly without fear of retaliation
  • Organic reputation: Trust shifts feel natural, not punitive — you don't know why your feed changed
  • No rating anxiety: Users aren't watching a number
  • No gaming: No visible score to optimize, so behavior optimizes for the interaction instead
  • Simpler UI: No rating display component or history view needed

Negative

  • Less transparent: Users can't see why their feed weight is changing
  • Harder to debug: If feedback data is affecting results incorrectly, users can't surface the issue
  • Some user expectation mismatch: Users familiar with Uber/Airbnb may expect to see ratings

Neutral

  • Karma is still visible: Public karma score and interaction count remain surfaced on profiles
  • Trust distance still shown: The "trust path" badge (direct / 2nd degree / community) gives directional information without rating numbers

Implementation

  • services/reputation-service/src/database/feedbackDb.ts — insert/query layer
  • services/reputation-service/src/routes/reputation.ts — POST /reputation/feedback (authenticated, no GET exposed)
  • apps/frontend/src/components/UpcomingPanel.tsx — star picker appears inline after match completion, silently submits, no confirmation of the rating value shown
  • apps/frontend/src/lib/api.ts — reputationService.submitFeedback() — fire-and-forget pattern (errors swallowed client-side)

References

  • ADR-035: Karma Allocation Strategy and Trust Score Abstraction
  • ADR-011: Reputation Decay System
  • services/reputation-service/src/services/trustScoreStrategy.ts