# Lead scoring validation evidence run

Date: 2026-08-28

## Commands

```bash
deno fmt --check docs/seo/evidence/lead-scoring-validation/*.ts
deno check docs/seo/evidence/lead-scoring-validation/*.ts
deno test docs/seo/evidence/lead-scoring-validation/*_test.ts
deno run --allow-write \
  docs/seo/evidence/lead-scoring-validation/run_fixture.ts
deno run --allow-read --allow-write \
  docs/seo/evidence/lead-scoring-validation/render_manifest.ts
```

## Reviewed result

- 16 synthetic candidate records.
- 10 records accepted for holdout evaluation.
- 6 records rejected across five fail-closed reason codes.
- Review capacity: top 4 records.
- Precision at 4: 0.75.
- Recall at 4: 0.60.
- Lift at 4: 1.50 against the fixture base rate.
- Pairwise ROC AUC: 0.72.
- Evidence coverage: 0.975.
- Automated tests: 25 passed, 0 failed.

## Artifact integrity

- Manifest SHA-256:
  `ecbc6633dbe9448912a8f9bd4f26808cee3139165538bab1a75b7918893ac9ce`
- SVG SHA-256:
  `3d0c1779f697a2b2490518c98ecab780390dfe6f844e2205eda5d06d0ae19d50`
- SVG rendered at 1280×720 and visually inspected: no clipping, overflow, or
  missing labels.

These values describe the included synthetic fixture only. They demonstrate
deterministic metric calculation and validation behavior; they are not a model
benchmark, recommended score threshold, or expected conversion result.