Conflicting Claims¶
Placeholder — this page needs to be written.
Scientific and legal literature routinely contains contradictory claims. A graph that silently resolves conflicts by picking a winner destroys information. This framework represents disagreements explicitly.
What counts as a conflict¶
Two relationships conflict when they share the same subject and object entities and the same (or inverse) predicate, but differ in:
- Polarity — one asserts the relationship holds; another asserts it does not.
- Magnitude — one claims a drug increases a biomarker; another claims it decreases it.
- Conditionality — the relationship holds under one set of conditions but not another.
How conflicts are stored¶
Conflicting edges are stored as separate relationship nodes, each with its own
provenance. A conflict_group tag links them so that query interfaces can surface
the disagreement rather than arbitrarily returning one result.
Resolution strategies¶
The framework does not resolve conflicts automatically. Options for downstream users:
- Surface all — return all conflicting claims with provenance; let the user judge.
- Weight by confidence — rank claims by confidence score, but show the others.
- Flag for review — mark the conflict group for human curation.
- Domain rules — a domain pipeline may define rules (e.g. "prefer randomized trials over case reports") that influence display order without discarding minority claims.