# Raw-data verification (our QC record — NOT a reviewer input)

This is an internal check that the raw-data workbook given to the AI reviewers
(`runs/inputs/raw-data.xlsx`) is the genuine analysis dataset behind the published paper.
It is **not** part of the reviewer input set and must not be shown to the reviewing
instances. It exists to support our later fact-check and the note's reproducibility claims.

The workbook is the journal's shared file (`Sheet1` = 1694 participants × 135 variables,
plus a `Dictionary` sheet). Computed directly from `Sheet1`, the published headline numbers
reproduce exactly:

| Check | Computed | Published |
|-------|----------|-----------|
| Primary outcome, TXA | 251 / 845 = 29.7% | 29.7% |
| Primary outcome, placebo | 297 / 846 = 35.1% | 35.1% |
| Primary outcome analyzed N | 1691 (3 missing PPH) | 1691 / 1694 = 99.8% |
| Safety population, TXA | 837 | 837 |
| Safety population, placebo | 845 | 845 |
| Serious AE (`Severe_side`), TXA | 4 / 837 = 0.5% | 0.5% |
| Serious AE (`Severe_side`), placebo | 4 / 845 = 0.5% | 0.5% |

This confirms `Sheet1` is the analysis dataset and fixes the group coding: **`GroupNo1` 1 =
TXA, 0 = placebo**; `safepopulation == 1` is the safety population (1682 = 837 + 845);
`safeGroupNo1` is populated for all 1694 rows.

## Why this folder no longer stages reproduction inputs

An earlier version of this work split the workbook into `Rdata1.xlsx` and `Vital.xlsx` to
let a reviewer re-run the **authors' own R code** from the supplementary appendix. Per
Eric's decision, the appendix (`zhal089636.ww.pdf`) is **withheld** from the AI reviewers —
they reproduce results independently from the raw data, which is a stronger test of the
affordance. The split files and the appendix were therefore removed from `runs/inputs/`.
The authors' undisclosed package versions and the fact that they never deposited the exact
`Rdata1.xlsx`/`Vital.xlsx` named in their code remain reproducibility limitations worth a
sentence in the note.