# Product Decisions: 40 Concrete Answers

1. **What is Coherence?** Kiduna's configurable alignment Capacity for defining legitimate desired states, observing contextual actual states, measuring drift and boundary risk, and selecting authorized feedback, adjustment, mediation, repair, or escalation.
2. **Minimum object?** A versioned binding containing entity reference, profile reference, policy references, current state reference, Sentinel reference, Authority reference, Consent reference, visibility, and version.
3. **Dimension?** A versioned operational construct with one meaning, a scale, applicable entity/scope types, observable indicators and counter-indicators, evidence rules, validity limits, and prohibited uses.
4. **Dimension scope?** All three. System-library dimensions are portable; Realm dimensions are governed locally; entity dimensions are private/local. Stable IDs and versions prevent name collisions.
5. **Who can create one?** Anyone may draft. Publication requires the Authority for its intended namespace: Kiduna governance for system dimensions, Realm governance for Realm dimensions, or the controlling Source/authorized steward for an entity-private dimension.
6. **Valid dimension?** It is bipolar or ordered, behaviorally observable, single-construct, falsifiable, context-aware, distinguishable from identity/diagnosis, supported by explicit evidence rules, calibrated for its use, and bounded by prohibited uses.
7. **Target?** An authorized desired point or interval for a dimension in a scope and time period.
8. **Why not zero?** Zero is only a coordinate. A legitimate purpose can require a candid Ally, formal Realm, cautious deployment, or forceful emergency Actor. The target expresses intent, not statistical normality.
9. **Tolerance range?** Inclusive interval considered aligned enough that no drift response is needed. It can be asymmetric around the target.
10. **Drift?** Signed distance outside the tolerance interval. Target distance is reported separately.
11. **Extreme?** A dimension-defined absolute region requiring heightened review independent of local target. Unusual is not automatically extreme or unsafe.
12. **Hard boundary?** A policy rule or absolute scale region that cannot be crossed under the applicable Authority tier; local profiles cannot neutralize it.
13. **Current-state measurement?** Permissioned observations are normalized, weighted by reliability, contextual relevance, and time decay, then aggregated with contradiction and coverage penalties.
14. **Enough evidence?** The dimension's minimum count, minimum independent contexts, coverage, agreement, and confidence threshold must all pass. Defaults are three observations across two contexts for non-consequential feedback; consequential intervention requires at least five across three contexts unless deterministic Action evidence proves a boundary event.
15. **Uncertainty?** A `0…1` confidence value plus evidence count, coverage, agreement, recency, missing-data status, and reasons. Confidence is not probability that a person “is” a trait.
16. **Observation persistence?** Append-only, contextual, purpose-bound records with minimized evidence references, retention class, visibility, Consent, and expiry. Raw prompts are not retained by default.
17. **Temporal decay?** `weight = reliability × relevance × 0.5^(ageHours / halfLifeHours)`. The dimension sets half-life; observations expire when weight or policy retention reaches zero.
18. **Inheritance?** Resolve constraints by intersection, policies by priority, and targets by an explicit ordered override chain. Every effective field records its source.
19. **Inherited conflict?** Higher-order constraints win. For targets, a valid more-specific authorized target overrides; if two equal-priority targets conflict, the engine emits `resolution-required`, uses the last known valid profile or safest non-controlling fallback, and forbids automatic intervention based on the unresolved dimension.
20. **Source changes?** Through a prepared versioned configuration Action showing scope, inherited effects, permissions, expiry, and diff. Draft changes are reversible; published Realm changes follow governance.
21. **Sentinel monitors?** Events, authorized evidence features, derived states, Action preflight contracts, target drift, policy/boundary conditions, intervention outcomes, calibration, and unresolved conflicts.
22. **Automatic intervention?** Only for an AI/Actor/Automation when confidence and zone thresholds pass, policy explicitly permits the response, Authority is current, the action is reversible, and it does not widen scope. A deterministic boundary hit may be constrained immediately.
23. **Merely recommend?** Low/medium confidence; human-related feedback; material profile changes; Realm culture changes; contested evidence; and any response outside standing Authority.
24. **Escalate?** Hard-boundary hit, high-confidence escalation zone, repeated failed correction, ambiguous high-consequence conflict, unavailable required Authority, or an appeal/repair condition defined by policy.
25. **Entity differences?** Realms receive aggregate facilitation/governance recommendations; Allies may receive prompt modulation; Actors receive Action preflight constraints; Avatars receive representation assistance and optional Source-facing feedback only.
26. **Presence?** Presence states intended behavior. Coherence measures actual behavior against explicit targets and may supply a bounded temporary modulation; durable instruction change is separate.
27. **Wisdom?** Wisdom supplies relevant knowledge/evidence. Coherence records remain structured and purpose-bound; only consented reusable summaries enter retrieval.
28. **Abilities?** Abilities contain executable procedures. Coherence invokes Skills but validates their outputs and never treats skill possession as Permission.
29. **Automations/Actors?** Coherence performs synchronous preflight for consequential Actions, can narrow/deny/pause under policy, and evaluates outcomes asynchronously. It cannot grant new authority.
30. **Connections/tool use?** Preflight verifies exact Connection scope, data, purpose, target, resource limits, and Coherence constraints. Revocation/narrowing is immediate; widening uses the ordinary Permission flow.
31. **Character creation?** Plain language, archetypes, scenario choices, comparisons, cards, and optional sliders compile into a draft profile; the Source reviews concrete behavioral examples and publishes it.
32. **Realm creation?** Founders/governors select operating norms, decision posture, privacy, risk, conflict, and experimentation targets; the system exposes unusual choices and boundary interactions before governed publication.
33. **Mediation?** The interaction Sentinel overlays participant and relationship targets, identifies mismatches without fault, proposes translation/pacing/turn-taking or a temporary negotiated profile, requests Consent, and expires the compact.
34. **Feedback?** Plain-language claim, evidence scope, confidence, impact, target comparison, reversible options, and inspect/decline controls. No judgmental labels.
35. **Unusual targets?** Any in-scale target is valid unless it conflicts with a higher-order constraint/boundary. The engine assesses distance from that target, never from zero.
36. **Never configurable away?** Human agency and Consent, privacy and data-purpose limits, non-deception requirements, explicit governance, authority boundaries, prohibition on unauthorized resources/tools, and system safety/legal constraints.
37. **Anti-social-scoring design?** No global human score, no cross-Realm default, no hidden ranking, no identity labels, purpose limitation, contextual dimensions, minimum evidence, access controls, expiry, correction, deletion where lawful, and a schema that contains no `overallScore`.
38. **Privacy/Consent?** Measurement requires declared purpose, lawful Authority, visible scope, minimum data, allowed observers/viewers, retention, portability rule, and revocation. Human observations do not cross Realms without explicit Consent or a specific legal/governance basis.
39. **Engineering contract?** Draft 2020-12 schemas, pure merge/measure/assess/response functions, versioned REST-style APIs, CloudEvents-compatible Events, synchronous preflight hooks, asynchronous measurement workers, and append-only audit Records.
40. **Operational Skills?** `coherence-profile-builder`, `coherence-assessor`, `coherence-sentinel`, `coherence-calibrator`, `coherence-feedback`, `coherence-mediator`, `coherence-repair`, and `coherence-scale-author`.