add `m` into the store, if it exists first remove it and its mappings each man page may have aliases besides the name determined by its

find a man page by its name, everything following the last dot (.) in name, is taken as the section of the man page we return the man

Build a minimal RawManpage for test use.

Create a new (or open an existing) writable database and return a Store.

Helper to build a ParsedManpage with the conventional source path.

Build a fake Batch object matching openai.types.Batch shape.

Build a minimal PreparedFile for testing.

Yield a path to a fresh temp SQLite file, cleaned up on exit.

Remove options whose flag set is a subset of (or equal to) another's. Covers both strict-subset (e.g. {Z} vs {-M, Z}) and exact-match duplica

Copy and restructure manpages from *src_dir* to *dst_dir*. Only sections listed in ``SECTIONS_TO_KEEP`` are copied. Section directories are r

Build a fake BatchExtractor with prepare/finalize/batch_provider.

Like ``run_batch``, but collects per-file results into a list.

Build a mock batch provider.

Create a minimal source directory with man1/ and man8/.

Zero-everything summary, used when the histogram is unavailable. Old reports written before option-count tracking existed migrate to

Return the serving Store for the current Flask app. Production keeps one read-only CachingStore per worker process. Debug/dev keeps the old p

Convert one LLM option dict to a models.Option. Uses the "lines" field to slice the description from original_lines.

Append an entry to the db_events log.

Return the score of the mapping from *src* to *dst*, or None.

Normalize known LLM type mistakes in a raw option dict before validation. Fixes observed in the top-1000 corpus run: - ``has_argument: null``

Extract response content and token usage from codex JSONL output. Looks for ``item.completed`` events whose ``item.type`` is ``"agent_message

Return the --db path or raise a UsageError if not set. When *must_exist* is True, also verify the file is present on disk.

Run patched ``mandoc -T markdown <gz_path>`` to get markdown directly.

Unified dispatcher for all execution modes. - ``batch_size`` set → batch mode (requires ``BatchExtractor``). - ``jobs > 1`` → parallel mode v

Build description text from line range [start, end] (1-indexed, inclusive). The first line is treated as the flag line. Remaining lines form the

* Mark a function for special use by Sizzle * @param {Function} fn The function to mark

Record a batch outcome and flush to disk. If an entry with the same ``batch_idx`` already exists (e.g. from an earlier "submitted" re

Like ``run``, but collects per-file results into a list.

Reconcile subcommand mappings using declared subcommands from LLM-extracted pages. For each parent manpage that has a non-empty ``subcommands

Remove blacklisted top-level sections from manpage text. Returns (filtered_text, removal_counts).

Group work items into batches, respecting batch_size as a request count limit. Each work item stays whole (all its chunks in one batch). A batch

Fix common LLM mistakes in option fields. Returns (short, long, has_argument, positional, nested_cmd, prefix).

Read a dotted metric path, returning 0 if any segment is missing.

Create a ParsedManpage from a source path like 'distro/release/1/name.1.gz'.

Write report.json to the run directory.

* Support testing using an element * @param {Function} fn Passed the created div and expects a boolean result

Split text into numbered chunks at section boundaries. Each chunk gets line numbers corresponding to the original text. If the text is small

Application factory.

Return recent db_events, newest first. If *event* is given, only rows matching that event type are returned.