# Source-facing UX

The default interface uses questions and behavior, not psychometric terminology.

## Required views

- **Profile:** desired dimensions with pole meanings, target, tolerance, scope, source, and expiry.
- **Current state:** contextual estimate, confidence, evidence coverage, and last update; `unknown` is explicit.
- **Drift:** target distance, tolerance-band drift, direction, zone, and boundary status shown separately.
- **Edit:** plain-language choices, scenarios, sliders, direct numeric entry for experts, inheritance diff, and rehearsal.
- **Explain:** evidence allowed for this viewer, method/version, alternative explanations, policy, Authority, and selected response.
- **History:** target/state/confidence over time with configuration changes and context breaks; never a human leaderboard.
- **Recommendation:** accept, modify, decline, collect more evidence, correct evidence, appeal, and expiry.
- **Character creation:** natural language/preset/scenario mapping to an inspectable draft.
- **Realm creation:** governed culture/process choices with participant/privacy effects.
- **Mediation:** participant-consented mismatch and temporary compact, with private details suppressed.

## Visualization rules

Bipolar meters answer “where are target, tolerated range, observed estimate, and boundaries?” A trend answers “is the estimate or target changing?” An interaction overlay answers “where do the participants' legitimate targets differ?” Radar charts MAY compare several dimensions for one purpose but MUST show missing/confidence and MUST NOT imply an overall score.

Every control exposes scope, Authority, viewer, retention, and whether a change affects desired profile, generated representation, or actual human behavior. Avatar controls use first-person aspiration language and place “decline/disable inference” beside recommendations.