Fetch and decompress a raw manpage. Returns a ``RawManpage`` or ``None`` if not stored.

List source paths that start with *prefix*. Uses a range scan on the ``manpages`` primary key index.

Return distinct section directories for a distro/release. Extracts the third path component from source keys matching ``distro/releas

Create a provider for the given model string.

Pre-process a manpage without calling LLM. Raises SkippedExtraction if the manpage is too large.

Raises ValueError if data is missing 'options' or options have wrong types.

Return ``<run_dir>/<kind>/<stem><suffix>``, creating the subdir.

Map a provider exception to CONTENT_FILTER or PROVIDER_ERROR. Azure / OpenAI surface content-policy refusals as a 400 with ``code='co

Build an ExtractionStats from a PreparedFile's prep-phase metrics.

Apply DB cleanup and per-file logging; bucket decisions for the caller.

Normalize HTML entities mandoc -T markdown emits. `&nbsp;` becomes a plain space — its non-breaking semantics aren't needed downstream and AS

Build a real Store backed by in-memory SQLite with standard test fixtures.

Build a summary from a list of per-file option counts. Empty input yields :meth:`empty` rather than ``None`` — the field is always pr

Strip parent-name prefix and deduplicate subcommand names. LLMs sometimes return ``["git-add", "git-commit"]`` instead of ``["add", "commit"]

Print and return summary stats for a list of timings.