case-intakelisted

# Konsil Case Intake Input: a case directory (e.g. `cases/<case-id>/inbox/`) containing whatever the user has — PDFs, photos of lab printouts, prior reports, pathology, a DICOM disc copy. Output: a structured case file the board can trust, or a refusal to proceed with targeted questions. **Language rule:** the case file and all questions to the user are written in the language of the source documents / user dialogue. ## Step 1 — Inventory List every file. Classify: lab report / imaging report / pathology / clinical note / imaging pixels (DICOM/JPEG of scans) / other. **Pixels are never interpreted here** — note their existence for the `imaging-3d` skill and move on. Record document dates; unreadable or undated files go on the missing-data list. ## Step 2 — Extraction with verification - **Labs:** extract analyte, value, unit, and the reference range *printed on that report* (ranges are lab-specific — never substitute remembered ranges). Map analytes to LOINC via the NLM Clinical Tables API (see `literature-protocol`). Plausibility check every value against its printed range and physiologic possibility; photo-extracted values are lower-trust — re-read the image a second time independently and reconcile; any mismatch or implausible value goes to the user for confirmation before it enters the case file. - **Imaging reports:** extract findings verbatim-faithful (measurements with dates), impression, and recommendations. Do not editorialize. - **Cross-document checks:** recomp