Check categories

1. CONSISTENCY — router (index.md) domain counts vs each domain-{domain}.md row count; every concept domain has a sub-index; broken concept links classified as concept_known / source_resolvable / truly_broken (only the last is a real finding); terminology drift and contradictory rules across articles.
2. FRESHNESS — sources with git commits after their compiled_at; concepts aged past N days; orphan concepts (all sources deleted).
3. COVERAGE — raw .md under scan_dirs absent from _meta.json.files; ghosts (meta entries with no file); per-directory coverage %. wiki_exclude sources are not counted as gaps.
4. CONNECTIONS (full-scan safety net for compile's incremental LINK) — concept pairs sharing 2+ sources but not cross-linked; merge candidates (very short) and split candidates (very long).
5. GAPS — terms implied across articles but lacking their own concept; domain imbalance; glossary misses.
6. SIZE — concepts >8 KB (WARN) / >16 KB (ERROR, propose a split).

Output

Write <bundle>/indexes/lint-report.md in output_language with a summary table (issues/suggestions per category) and a section per category listing findings. Use the dialect's callout style for emphasis (> [!warning] in obsidian, plain > **Warning:** in okf-pure).

Post-lint actions

Suggest concrete next steps: re-compile stale scopes (/wiki-compile --scope {dir}), pick up uncovered files (/wiki-compile --full), reset the LINK baseline on recurring connection gaps (/wiki-compile --full-link), split oversized concepts, and resolve contradictory rules by checking sources.

Notes

• Lint is read-only on concept articles; it only writes the lint report.
• Run after each major compile cycle. Git tracks report history.