Core components
7 weighted signals
How VeriScore Works
Inside the VeriScore Confidence Engine
VeriScore is not a black-box label. It is a weighted confidence framework that combines model consensus, source quality, temporal validity, and uncertainty controls into one audit-ready number.
Output range
0-100 confidence
Review context
Evidence first
Section 1
1) Signal Collection Across Independent Paths
Each verification run collects independent signals before scoring. We avoid single-model dependence by combining structured outputs from multiple model families and citation trails from trusted sources.
- Provider agreement is measured as weighted consensus, not simple majority votes.
- Source authority is calculated from provenance quality and credibility history.
- Counter-evidence is preserved to avoid overconfident false positives.
Section 2
2) Weighted Score Construction
The score is composed through a fixed weighting policy so teams can reason about why confidence moved up or down between revisions of a claim.
- Source authority and model consensus carry the highest influence in the final score.
- Temporal validity down-weights stale or superseded references.
- Nuance and uncertainty penalties reduce inflated confidence on ambiguous phrasing.
Section 3
3) Human-Readable Confidence Bands
Scores are emitted with banded interpretation so editorial and compliance teams can operationalize policy thresholds instead of guessing from raw numbers alone.
- High-confidence TRUE claims include clear source and reasoning chains.
- Mixed and partial outcomes preserve unresolved ambiguity explicitly.
- Low-confidence results trigger escalation for deeper review before publication.
Run VeriScore On Your Own Claims
Use live verification to see score breakdowns, compare claim revisions, and export evidence-backed reports your team can defend publicly.