Evidence Score

Score evidence strength claim-by-claim with a transparent rubric.

The evidence-score workflow quantifies support quality so you can separate strong findings from weak or uncertain ones.

How it works

The workflow applies a rubric-based scoring pass:

  1. Claim inventory — extracts claims from your artifact or topic summary.
  2. Evidence alignment — maps each claim to supporting sources.
  3. Scoring — computes support strength using source quality, consistency, and reproducibility cues.
  4. Risk flagging — marks weakly supported claims for downgrade, rewrite, or removal.

Scores are decision support, not absolute truth; always read associated risk notes.

Real examples

Example 1 (score an existing brief):

/evidence-score outputs/my-brief.md
bohr evidence-score outputs/my-brief.md

Example 2 (score by topic):

/evidence-score "enterprise copilots ROI claims"
bohr evidence-score "enterprise copilots ROI claims"

Expected output

Primary output file:

  • outputs/<slug>-evidence-scores.md

Typical structure:

  • scoring rubric section
  • claim-by-claim score table (0-100)
  • confidence and key-risk annotations
  • weak-support summary (claims under threshold)

Example snippet:

| Claim | Score | Confidence | Key Risks |
|---|---:|---|---|
| "Latency drops 35% with adaptive retrieval." | 82 | Medium | Small sample size; one benchmark family |
| "Approach generalizes across domains." | 41 | Low | Evidence limited to legal QA only |

Customization

You can customize:

  • thresholds — “treat anything below 60 as weak”
  • weights — “weight code-backed replication higher”
  • scope — “score only top-line claims from summary”
  • format — “include action recommendation per low-score claim”

When to use

Use this workflow before decision reviews, prioritization meetings, or whenever stakeholders need transparent confidence framing.

Use alongside /contradict when you need both quantitative support scoring and adversarial critique.