filinglisted

# filing ## Why this exists A whole class of confident false negatives has one root cause. In one session, four separate "we don't have that" assertions were all wrong: a log, two record entries, and a batch of scanned reports. Every one existed. The cause was not carelessness. It was that **hundreds of files sat in the inbox referenced by zero scripts and indexed in zero recall sources.** Artifacts with no canonical home and no index produce confident false negatives, which is the most expensive kind of wrong: it looks like an answer. ## The one rule that matters **The index is the point, not the tidy tree.** Moving a file to a prettier path fixes nothing on its own. What failed was RECALL. So: - Every artifact gets an `archive/INDEX.jsonl` row **even when it cannot be classified**. - `archive/unsorted/` is a first-class destination, not a failure state. - An indexed unsorted file is findable. A perfectly-filed unindexed file is not. If you ever have to choose between placing something correctly and indexing it at all, index it. ## Procedure ### 1. Dry run first, always ```bash python3 scripts/pipeline/file_intake.py --verbose ``` Dry run is the default and `--apply` is required to write anything. Read the subject and label distribution before applying. If the distribution looks wrong, the signatures are wrong; fix `SIGNATURES` / `OCR_SIGNATURES` in the script rather than hand-filing. ### 2. Apply ```bash python3 scripts/pipeline/file_intake.py --apply