Trust Scoring (Explainable Signals)
CueCrux surfaces explainable trust signals so you understand why evidence was selected without publishing raw weights that can be gamed. Signals differ by object and by mode (light, verified, audit).
Signals by Object
Domains
- Reputation prior (official/peer‑reviewed vs low‑credibility venues)
- Retraction and venue‑risk trends
- Licence clarity and robots compliance
- Uptime/availability and change frequency
- Historical contradiction rate
Artefacts
- observed_at (recency) and re‑verification timestamp
- content hash (BLAKE3) and provenance ledger status
- licence_ok flag and domain lineage
- duplication/similarity and near‑duplicate handling
- reuse count across independent orgs
- counterfactual presence and resolution
Users/Orgs
- Verified identity/KYB (where applicable)
- Artefact reuse and contradiction rates
- Successful receipt ratio and dispute outcomes
- Policy strikes and remediation
Mode Thresholds (Qualitative)
- Light: fast answers; minimal provenance checks; diversity encouraged but not enforced.
- Verified: full provenance checks; QUORUM (MiSES) coverage; ≥2 independent domains when available; contradiction surfacing.
- Audit: deterministic replay; counterfactuals mandatory; signed receipt required.
Diversity & Counterfactuals
- Domain diversity aims for independent corroboration; bursts from the same domain are damped.
- Credible counterfactuals are shown and reduce confidence until reconciled.
Trust Report Subscores
UI shows bars for: domain diversity, recency, licence, contradictions, and source authority. Exact numeric weights are not published and change over time.
Anti‑Gaming & Appeals
- Signals change; WatchCrux audits penalise circular reuse, same‑IP clusters, and excessive same‑domain citations.
- Appeal path: open a ticket with receipts, propose corrections (licence updates, better sources). Accepted corrections improve future runs.