Link-check population cross-check limits

Seal uses a family-fixture link guard that clones six sibling repositories under the checkout’s .family/ directory. Because the checker recursively walks the checkout root, the occurrence counts in its summary include Markdown links in those clones. Those counts are diagnostic output, not a population assertion: legitimate source edits may add or remove repeated occurrences without changing the set of targets that must be checked.

test/linkcheck.test.mjs compares the targets reported by scripts/linkcheck.mjs with targets reconstructed from repository sources. This is a separate-source cross-check, not a separately implemented population oracle.

The two routes overlap in these bug classes: recursive file enumeration, Markdown/HTML source filtering, JSON/YAML data filtering, JSON string traversal, path-string recognition, URL skipping, and HTML attribute extraction. A defect in one of those shared rules can hide a target from both routes. The check therefore measures agreement between the product extractor and this overlapping reconstruction; it does not establish complete target population coverage.

It still catches some product-only omissions, including a target dropped after extraction and before check(). It also requires the checker to finish and report zero broken targets. Broader assurance requires a source with a genuinely distinct file list and extraction model.