Explainable Match Scores: Factors, Confidence and Policy Versions
Explainable scoring shows how a candidate was evaluated and which policy produced the result.
An explainable match score must show the factors evaluated, points earned and available, hard failures, uncertainty, confidence and scoring-policy version. The number alone is not explainable. Versioning matters because changed weights or thresholds can produce a different result from the same facts. Reviewers need the evaluated inputs and rationale to reproduce or challenge the outcome.
The FCPV scoring record
Keep factors, confidence, policy and version together so later reviewers can understand both the result and the rules behind it.
| Dimension | What to examine | Decision rule |
|---|---|---|
| Factors | Name each dimension and explain its earned and maximum contribution. | Avoid hidden composite scores. |
| Confidence | Derive confidence from the score while marking uncertainties separately. | Do not call incomplete data high confidence without explanation. |
| Policy | Store weights, thresholds and hard-gate rules as governed configuration. | Test policy changes on realistic scenarios. |
| Version | Attach the exact policy version to every candidate evaluation. | Re-evaluate deliberately rather than rewriting history. |
Example: the same facts under two policies
A candidate receives 64 points under a policy whose medium-confidence threshold is 65.
A later policy lowers that threshold, but the original record remains tied to its version. A new evaluation may differ without making the earlier record false.
Illustrative scenario, not a customer claim or guaranteed outcome.FoundBefore policy records
FoundBefore stores deterministic factor explanations, hard failures, confidence and the active industrial scoring-policy version.
Admin policy changes require governed drafts and publication actions rather than silent code edits.
Limits and responsibilities
- Explainability does not make weights objectively correct.
- Versions require operational governance.
- Some business risks remain outside the model.
FoundBefore does not guarantee a match, buyer, supplier, partnership or completed deal. Identity disclosure requires mutual consent, and the controlled pilot does not replace procurement, contracting or human review.
A useful score is a versioned decision record, not a number detached from inputs, rules and uncertainty.