Known limitations

These behaviors reflect the current implementation:

• The dependency graph code exists internally but is not emitted by the indexing engine, so generated output does not contain a dependency graph.
• On a cached or update run, index.md is built from newly parsed files rather than every cached file. It is not a complete view of unchanged files.
• When a source file is deleted, its cache entry is pruned but its existing generated Markdown in output/files/ is not removed.
• parse --detail selects full CST output only when its value is exactly full; invalid values currently fall back to summary output instead of failing validation.

Use a clean output directory and a non-cached parse when a complete, current output set is required.