# Feedback

Coherence feedback is a reversible information Action. It separates evidence from interpretation and offers choices.

## Required structure

1. **Scope:** entity, context, period, and purpose.
2. **Observation:** what behavior/events were considered.
3. **Comparison:** observed estimate, target, tolerance, and direction of drift.
4. **Confidence:** plain-language level and why it is limited.
5. **Impact:** practical consequence, without moral judgment.
6. **Options:** inspect, collect more evidence, adjust behavior/representation, change an authorized target, decline, or escalate.
7. **Controls:** visibility, evidence view, correction, history, and appeal.

Example:

> Rowan is configured to challenge assumptions. Across five planning responses, it raised no substantive counterpoint; the current directness estimate is +7 against a +35 target (confidence 0.78). You can ask Rowan to challenge the next plan, inspect the five observations, collect more evidence, or leave the profile unchanged.

Do not say “Rowan is submissive” or “you are incoherent.” Do not present a number without pole meanings and evidence scope. If confidence is low, ask or observe rather than prescribe.

Avatar/Source feedback is opt-in and targets generated representation or chosen aspirations. Realm feedback goes to authorized governance/facilitation views and protects participant privacy.