# Supplementary tables

Generated from the frozen corpus by `ingest/build_supplementary.py`.
They are derived on every build rather than hand-maintained, so a change
in the corpus changes them and cannot leave a stale copy behind.

## Derived here

- **Supplementary Table 1** — Per-species post-listing window and baseline assessment (43 rows)
- **Supplementary Table 6** — Post-listing series on both reporter sides (2654 rows)
- **Supplementary Table 7** — The AI toolchain: vendor, product, version basis, access dates, configuration and the human gate per step (3 rows), built from `../pipeline.json` by `../build_supplementary_table_7.py`
- **Supplementary Table 8** — Focal-set reconciliation by appendix (6 rows)
- **Supplementary Table 9** — Post-listing quantity series per unit dimension (334 rows)

## Also in this directory

- **`change_table_published_vs_current.csv` / `.md`** — every
  claim-bearing number in the submitted manuscript beside its value in
  the rebuilt corpus, with the sentence it appeared in. The published
  column is historical; the current column is recomputed on every build.
  Produced by ingest/build_change_table.py.
- **`species_narratives_no_citations.csv`** — the 343 machine-drafted
  species narratives with their citations removed. The narratives are
  gated: none has been signed by a reviewer, none is served, and no
  reported value derives from one. Their citations were never verified,
  and sampling found both references that resolve to nothing and
  references that resolve to real papers attached to claims those papers
  do not make — the second kind passes a reference-existence check, so
  only reading the source catches it. This rendering keeps the prose and
  drops the attribution. Produced by ingest/strip_narrative_citations.py;
  the original text remains in `../tables/species_narratives.csv`.

## Elsewhere in the deposit

- **Supplementary Table 5** — results traceability: `../results_traceability.csv`

## Not derivable, and why

These need an author. They are named rather than stubbed: a placeholder
table looks finished and survives review, which is worse than an absence.

- **Supplementary Table 2** — The seven audit questions must be reproduced verbatim. The wording is the instrument itself and does not exist in the database; it comes from the authors' questionnaire.
- **Supplementary Table 3** — Step-to-stage mapping is an editorial statement about the Methods, not a derivation. It is written in the manuscript text and belongs there.
Two entries that used to sit here have moved up, and this note records why
rather than deleting them, because "not derivable" is a claim about the
project and it changed.

- **Supplementary Table 4** — was listed here as needing the outcome-tier /
  adoption-tier partition of Q1-Q7, a judgement only a subject author can
  make. The partition was made, the table is deposited as
  `supplementary_table_4_step_question_mapping.csv`, and `../pipeline.json`
  carries its SHA-256 so a reader can tell whether the mapping drawn in the
  figures is the mapping deposited. This note was stale from the day the table
  landed. On 22 October 2026 it was rebuilt on the manuscript's questions
  (Sections 4.1 to 4.7): the first deposited version mapped an earlier
  question set, whose Q2 was conservation status where the manuscript's Q2 is
  listing. The lead author confirmed the producing steps the same day, and
  `../pipeline.json` derives its per-step bindings from this table
  (`../../response/apply_st4_manuscript_questions.py`).
- **Supplementary Table 7** — was listed here as needing vendor build strings
  and access dates held only by the authors. On 1 October 2026 those values
  were recovered from the project's own operational record and written into
  `../pipeline.json`, so the table is derived like the others by
  `../build_supplementary_table_7.py`. What no vendor publishes is reported as
  absent rather than left blank: neither service exposes a build string for
  the model actually served, and the version column says so.