# Finding: unsigned machine-drafted prose and references

**Status:** withheld pending expert review. The material exists in the deposit
and is not served by the application.

## What Reviewer 1 found

Reference lists were identical across species pages. The front-end binding fell
back to a shared query instead of the per-species identifier. The underlying
per-species reference data were not themselves corrupted; the rendering layer
was.

## What the rebuild does

There are now **no** reference lists on species pages until a named subject
expert has reviewed and signed them. The same gate applies to narrative
sections. Pretending the section was repaired while still serving unsigned
machine output would repeat the defect under a different name.

343 narrative sections and their reference bindings exist in the deposit
(`tables/species_narratives.csv`). None carries an approved review
(`tables/narrative_review.csv` and `tables/profile_review.csv` are empty). The
count of withheld material is disclosed on every affected species page.

The application role cannot read unsigned narratives or unsigned reference
bindings. It can read counts of pending review, which is enough to disclose the
withholding without leaking the text.

## Why withholding rather than caveat

Machine-drafted text that nobody has checked is not displayed with a disclaimer.
It is not displayed. A caveat next to unreviewed prose invites the reader to
treat the prose as evidence with an asterisk. Withholding states the standard
instead.

## Related

- Application: `/changes#unsigned-prose`
- Deposit: `tables/species_narratives.csv` (343 rows), review tables (0 rows)
- API views: `arctic_api.species_narrative`, `arctic_api.species_reference`,
  `arctic_api.species_pending_review`