| name | confidence-score |
| version | 1.0.0 |
| description | Auto-score a page's confidence (high/medium/low) based on evidence levels,
backlinks, recency, and citation density. Also assigns a compile grade
(A/B/C/F) for the latest ingest event touching this page. Driven primarily
by dikw-compile and kos-patrol; exposed as a standalone command too.
|
| triggers | ["score this page","what's the confidence","grade the compile"] |
| tools | ["get_page","backlinks","list_pages"] |
| mutating | false |
Confidence Score Skill
Port of KOS v1 knowledge/agents/shared/quality-criteria.md scoring logic.
Confidence formula (high/medium/low)
Inputs:
E_max — highest evidence level on page (E0-E4)backlinks_in — count of pages linking TO this pageage_days — days since updatedcitation_density — inline citations / paragraphs
Thresholds:
| Confidence | Condition |
|---|
high | E_max ≥ E3 AND backlinks_in ≥ 2 AND age_days ≤ 90 AND citation_density ≥ 0.5 |
medium | E_max ≥ E2 AND backlinks_in ≥ 1 AND age_days ≤ 180 |
low | otherwise |
Auto-write confidence: <value> to frontmatter on every skill-driven page update.
Compile grade (per ingest event)
Same rubric as dikw-compile Phase 5:
| Grade | Ingest event outcome |
|---|
| A | 3+ impacted pages updated AND at least 1 new synthesis/decision |
| B | 1-2 impacted pages updated with precise strong links |
| C | Only source page created, no downstream propagation |
| F | Source page fails evidence-gate |
Written to source page frontmatter as compile_grade: A|B|C|F + compile_notes: <short>.
Rollup metrics (for kos-patrol)
Daily rollup computed by kos-patrol:
compilation_rate = pages_with_grade_A_or_B / total_pages_ingested_this_weekconfidence_distribution = {high, medium, low} countsstale_count = count where age_days > 180 AND status=active
KOS v1 reported ~63% compilation rate as of 2026-04-15. Maintain or improve.
CLI
bun run ~/Projects/jarvis-knowledge-os-v2/skills/kos-jarvis/confidence-score/run.ts <slug>
