Review Map

Review Map

Turn a source map into a traceable draft and compact exception queue.

Use compile-user-map instead when the requested outcome is an organized user map, JSON, CSV, or target output.

Follow ../../references/interaction-contract.md.

Process

1. Run python3 <skill-dir>/scripts/run.py --input <path> --output <directory>. PDF manuals and structured CSV, JSON, XML, XLSX, or text are accepted in this one invocation. Do not send the user to extract-pdf-map first.
2. Run parsing or PDF extraction, normalization, linting, and evidence review through without pausing.
3. Group rejected rows, assumptions, and holds by shared cause and present all independent blocking choices together.
4. Keep unresolved values and excluded rows visible; never ask for blanket approval when automated checks find no exception.

Output files

• map-draft.json - Start here if you need the normalized draft map.
• review.json - Open this for the compact evidence status and grouped exceptions.
• lint.json - Detailed deterministic validation findings; normally use it only to investigate a problem.
• parsed.json - The source-shaped candidate rows; normally use it only to trace a value back to parsing.

Completion requires a schema-valid draft, a complete automated review report, and a disposition path for every blocking exception. A clean draft is ready without an extra approval ritual.

Stop

• Stop if the user actually wants organized user-map outputs; use compile-user-map.
• Stop for writes, broadcasts, discovery, or unbounded polling.
• Keep unresolved address, area, unit, and byte-order fields as holds.
• Do not ask page by page or row by row.

Handoff

• The user confirms values or exclusions: suggest apply-review.
• The user only wants deterministic checks on an existing normalized map: suggest check-map.
• The user wants extraction evidence only: suggest extract-pdf-map.

Keep stage artifacts visible. Final target generation starts only from a validated map.