discovery(correlation): evaluate cross-source conflict detection

## Related epic

#5981 — discovery only; this issue does not block epic completion.

## What to build

Determine whether preserving and surfacing conflicting source claims improves trust enough to justify an entity-keyed conflict data model and user experience.

## Acceptance criteria

- [ ] Choose the first user-visible field and source pair, then define the baseline failure and primary outcome.
- [ ] Define claim identity, provenance, source confidence, freshness, and resolution semantics without overwriting disagreement.
- [ ] Derive field-specific numeric-spread thresholds rather than reusing unit-free constants.
- [ ] Specify user-visible states for unresolved, low-confidence, stale, and resolved conflicts.
- [ ] Evaluate UCDP/ACLED or another real source pair with positive and negative examples.
- [ ] Produce an explicit promote, defer, or reject decision and require fresh implementation review before coding.

## Blocked by

- #5987

Start with blocked issue #5987 and the related discovery epic #5981. Evaluate UCDP/ACLED or another real source pair using positive and negative examples, define the conflict semantics and user-visible states, and finish with an explicit promote, defer, or reject decision plus fresh implementation review before coding.

Written by the indexing model from the issue text.