[P2][RES-02] Claim extraction, contradiction and ranking

Programme: #1
Epic: #25

## Objective
Transform retrieved sources into evidence-backed claims, contradictions and explainable rankings without flattening uncertainty or duplicated reporting.

## Scope
- Claim extraction with bounded evidence spans and source/revision/time references.
- Distinguish fact, source assertion, interpretation and inference.
- Resolve entity/variant/time scope before merging claims.
- Source-independence and syndication/derivation detection.
- Contradiction clusters with evidence from credible competing sources.
- Confidence/uncertainty representation based on evidence quality and gaps.
- Ranking against explicit user constraints with explainable factors and commercial-conflict flags.
- Recompute/invalidate through DATA-03 when source state changes.

## Research rules
- Model agreement does not replace source evidence.
- Contradictory credible evidence remains visible rather than being averaged away.
- Ranking should optimize explicit user goals, not affiliate/engagement incentives.

## Acceptance criteria
- [ ] Every material claim can enumerate supporting/contradicting evidence spans.
- [ ] Duplicated/syndicated coverage is not counted as independent corroboration.
- [ ] Variant/time mismatches are detected before claims are merged.
- [ ] Ranking explanations expose constraints, missing facts and commercial conflicts.
- [ ] Source changes invalidate/recompute affected claims/rankings.
- [ ] Evaluation measures evidence support, contradiction retention and ranking constraint satisfaction separately.

## Dependencies
- RES-01
- DATA-02
- DATA-03

**First phase:** P2
**Maturity target:** P5
**Owner:** connectors-domains

No files, tests, or entry points are named. Start by reading RES-01, DATA-02, and DATA-03 to understand the dependency boundaries and source-state flow. Done means implementing and evaluating evidence-backed claims, contradiction retention, explainable constraint-based rankings, and invalidation when sources change.

Written by the indexing model from the issue text.