Link Validation (ev-link-validation)

Phase 4 validates URL health after by-type collection, live star verification, Phase 2B benchmark-source verification, and adversarial audit.

Type-First Evidence Lake Contract (#1148)

The evidence lake is type-first. The primary working set is evidence/by-type/<canonical-evidence-type>.md. Legacy evidence/tier_*.md files may still exist as coexistence artifacts, but they are not the semantic routing key.

Coexistence Shim

evidence/scripts/validate_sources.py intentionally remains a temporary coexistence URL-health shim. It reads evidence/by-type/*.md (primary, type-first) and evidence/tier_*.md (compatibility-only) by default (#1786).

python3 evidence/scripts/validate_sources.py

For benchmark rows, include every row/source URL plus catalog-level sourceUrl, methodologyUrl, harnessUrl, attestor URLs, and any candidate manifest methodology/attestor links in the URL-health pass.

For a small sample:

python3 evidence/scripts/validate_sources.py --limit 10

Targeting an intake's candidate rows (not yet in the lake)

A pre-registry intake's candidate rows aren't in evidence/by-type/ yet, so the default lake scan won't see them. Point the validator at them directly instead:

# Plain URL list, one per line
python3 evidence/scripts/validate_sources.py --urls /tmp/intake-urls.txt

# JSON/YAML candidate manifest — bare URL list, {"url": ...} objects, or an
# object with a urls/candidates/entries/rows list of either shape (same
# shape generate_source_dump.py's --candidate-manifest accepts, see
# ev-pipeline)
python3 evidence/scripts/validate_sources.py --manifest /tmp/intake-candidates.json

--urls and --manifest are mutually exclusive; --limit N and --report <path> apply in every mode.

Output

Write Firecrawl validation findings to the verification report and append a summary to the source report. Benchmark-source candidates remain candidate-only until Phase 2B findings, adversarial findings, and link-health results are reviewed by a human. For #1418 scratch multi-target peer-review partitions, validate the repeated URL once per reviewed skill row in peer-review.md when the source legitimately covers each target; the repetition is expected, not duplicate-noise by itself. Do not commit generated validation reports, scratch manifests, or generated partitions without human approval. Hand only live, correctly scoped rows to ingestion.