# K9b – bradbury2020 (opus-S3 rows 051–248) vs Supplementary File 006

Source: /home/claude/drive/supplements/bradbury2020_File006.txt (S3, S4, S8, S9). The txt was checked against the source docx (ije-2018-05-0670-File006.docx): all 606 HR strings identical.
Method: all 198 rows parsed and compared by script (HR, CI, cases, participants, mean intake = dose_point, P-trend, is_calibrated, is_main_model); flags and notes read manually.

## Discrepancies
| estimate_id | field | in database | in paper | severity |
|---|---|---|---|---|
| none numeric | – | – | – | – |
| per-increment rows (055, 060, 065, 070 … 248; 40 rows) | is_main_model | no (is_calibrated=yes) | Every per-increment trend in the paper uses re-measured 24-h means (Methods p. 251), so no uncalibrated trend exists. KODEBOK 4.2b (new) makes calibrated rows main when a paper publishes only calibrated trends. Bradbury publishes uncalibrated categories, so the current coding is defensible, but this is a Fable decision point for the continuous contrast | minor (decision) |

Calibration flags: category rows = no, per-increment rows = yes, consistent with the opus-F rows (005, 010, 015 …) and the Methods. Confirmed.

## Table 3 vs S3 discrepancy (row 055 / 026)
- S3 minimally adjusted red+processed per 50 g/d: 1.30 (1.10–1.53). Main Table 3 first row: 1.26 (1.12–1.42).
- S3 prints 1.26 (1.12–1.42) for red meat alone (row 065), and **1.30 (1.10–1.53) is also exactly the S3 minimally adjusted processed meat ≥2/week cell** (row 074).
- The S3 interval (log width 0.33) is about 40% wider than the fully adjusted 1.17 (1.04–1.32) (width 0.24), while Table 3's 1.26 (1.12–1.42) (width 0.24) fits the sequential series. Probably a copy error in S3, not in Table 3. Recommend treating 055 as the suspect value. Both rows are correctly recorded as printed and cross-referenced in notes.

## Checked and confirmed
- S3: minimally adjusted (m1) and fully adjusted excluding diet-changers (m2, subgroup excl_diet_change_illness) columns for 5 exposures: 50 rows. Mean intakes 21/52/64/76 (excl. column 21/52/64/75; red 7/34/44/54) correct.
- S3 fully adjusted all-participants column = Fig 1 rows 001–025 (opus-F): all 25 values identical (not re-coded, correctly).
- S4 men/women (48 rows): all correct incl. women-specific means (e.g. poultry 2/19/29/40, fish 6/22/29/37). Red+processed per 50 g by sex = rows 047/048 (Text), so not missing.
- S8 colon/rectum and S9 proximal/distal colon (100 rows): cases, HR, P-trend all correct; participants are whole-cohort per category (same in both subsites), correctly recorded.
- Row 127 (women, processed ≥2/wk, 1.03 (0.79–1.24)): asymmetric CI already flagged as a probable print error. Agree.
- adjustments m1/m2 match S3/S4/S8/S9 footnotes (stratified age, deprivation, sex, region; + waist, height, smoking, alcohol, education, PA, family history, NSAID, vit D, folate supplements, women's reproductive/HRT). adj_bmi=no correct (waist only).
- dose_conversions c6–c45 match the printed means.
- Exposure: red_total = red + processed (sum of frequencies); red_unprocessed = beef, lamb, pork (UKB touchscreen excludes processed); poultry/fish includes_processed=no correct.

All 198 rows confirmed numerically.

## Missing estimates
- Suppl Table S5 (fish types, oily and non-oily): 4 categories × (minimally adjusted, fully adjusted, fully adjusted excl. diet-changers) + per 25 g/d for the minimally adjusted and excl. columns = 14 rows per type, **28 rows**. Only the fully adjusted per-25 g values exist (049/050, from Text). Fish subtypes are negative-control material (R1); coding them is recommended for completeness (Jin's fish subtypes were coded).
- bradbury2020_File005.txt was not in this assignment and was not checked.

## Paper-internal inconsistencies
- Table 3 vs S3 minimally adjusted red+processed trend (see above).
- S4 women processed ≥2/wk CI asymmetric (probable typo).
- S4 coffee women cell "0.97 (0.82-1.15" is missing a parenthesis (not a meat row).