# 3330 - What the negative controls say: synthesis of the injury negative-control program (rounds 1 and 2)

Generated 2026-09-30 03:14 Central Daylight Time by `07_analysis_code/3330_negative_control_summary.py`, which reads the outputs of 3326 and 3331-3348 (round 1) and 3380-3385 (round 2) and refits nothing. Rows: `3330_negative_control_summary.csv`. Figure: `08_figures/3330_negative_control_summary.png`. Calibration preview: `project_results/negative_controls_3330/3330/calibration_preview.csv` / `.json`. Round-2 changes: `qa/negative_controls_20260925/3330_ROUND2_UPDATE.md`; the pre-round-2 outputs are kept in `qa/negative_controls_20260925/snapshots/3330_pre_round2/`.

## Bottom line

- Coverage: after round 2 16 of the 16 hierarchy contexts (15 pooled + U.S. rotavirus 2006, untied) have at least one fitted negative control (round 1: 13; the new ones are Canada Rotavirus (Excl. Quebec), Quebec Rotavirus, Czechia MMR). 10 have an injury control that is informative by the uniform 3330 class (round 1: 6); new: Hong Kong PCV, England Rotavirus, Northern Ireland Rotavirus, Colombia Varicella.
- Uniform 3330 class, any fitted row (the rule's descriptive tally): 7 contexts show a bias signal on at least one control (U.S. PCV (Core), U.S. Rotavirus 2006 (Core), Sweden PCV, Hong Kong PCV, England Rotavirus, Brazil Maternal DTaP, Colombia Varicella), 7 only consistent controls (U.S. RotaShield (Core), Japan MMR (Core), Germany PCV, Scotland Rotavirus, Northern Ireland Rotavirus, Canada Rotavirus (Excl. Quebec), Tanzania Schedule), 2 only uninformative ones (Quebec Rotavirus, Czechia MMR). Round 1 alone: 4 with a bias signal. Because this tally grows with the number of categories a source reports, the reading rows are tallied beside it (specified in the amendment): 4 contexts with a bias signal on a reading row (U.S. PCV (Core), U.S. Rotavirus 2006 (Core), Hong Kong PCV, Brazil Maternal DTaP), 9 consistent (Japan MMR (Core), Sweden PCV, Germany PCV, England Rotavirus, Scotland Rotavirus, Northern Ireland Rotavirus, Canada Rotavirus (Excl. Quebec), Colombia Varicella, Tanzania Schedule), 2 uninformative (U.S. RotaShield (Core), Czechia MMR).
- Review I5, descriptive: 5 fitted rows are UNINFORMATIVE by the width rule although their 95% interval excludes 1 (1 in hierarchy contexts: U.S. PCV (Core), Unintentional injury (V01-X59), ages 0-5 (3383) 2.59 (1.15-5.76)). They change no class.
- U.S. PCV: the 3326 rule's verdict stays MIXED, which draws neither 'drift' nor 'autism-specific' (only IDEA hearing is precise enough to count, and it leans at least as far as autism). Round 2's NCHS injury deaths on the state x cohort cells are UNINFORMATIVE by width (4.4x autism's interval) but exclude 1, 2.59 (1.15-5.76), with a failed future+3 placebo (0.18 (0.10-0.39)).
- Round 2 in one line each: England and Sweden now have informative CONSISTENT primary injury controls (STATS19 1.00 (0.87-1.16); Socialstyrelsen all injuries 1.06 (0.98-1.13)); Hong Kong's spurious 'effects' run in both directions (SEN harm-direction signals from a rule written with the estimate known, plus road casualties 0.66 (0.48-0.92)); Brazil has a second, independent register with a protective signal (SIM deaths 0.36 (0.15-0.79)); Canada excl. Quebec has a same-system asthma control (0.71 (0.48-1.06), CONSISTENT); Colombia's transport control is CONSISTENT only by a hair (1.36 (0.99-1.93)) and its autism estimate depends on the lockdown years; Germany, Czechia and Scotland still have no informative control on the block's own or closest cells (all prior-dominated).
- Calibration preview (documented primary set C_INJ, unchanged, n = 5): exp(mu) = 0.96 (0.51-1.87), tau = 0.48 (0.09-1.26); 16 of 16 calibrated hierarchy intervals include 1 and 10 are more than twice as wide as uncalibrated. Specified sensitivity S7 (C_INJ plus the 7 informative round-2 injury controls, n = 12): exp(mu) = 0.93 (0.72-1.19), tau = 0.29 (0.00-0.62), se(mu) = 0.11; 16 of 16 calibrated intervals include 1 and 7 are more than twice as wide. S7b (every eligible round-2 control, n = 14): tau 0.32; S8 (C_INJ without members that fail their own placebo or gate, n = 3): tau 0.28, exp(mu) 1.44. A preview, not a result.

## Interpretation rule (documented) and the round-2 amendment

Rule file `qa/negative_controls_20260925/3330_rule.json` (sha256 `7b853bb05d4b2c8df1938e7a79ccd6e63a6daa5b7a40c8722f5ffe212a0dff6a`), written 2026-09-29 21:29 CDT (2026-09-30T02:29Z), before 07_analysis_code/3330_negative_control_summary.py existed and before any pooled quantity of the negative-control program (systematic-error mean or SD, calibrated interval, harmonized-class count, figure) was computed by anyone. The per-row estimates and per-system verdicts were known when it was written (they were read to design it); what it fixed in advance is which rows are carried, the calibration sets and estimator, and the download priorities. It re-classifies nothing: every row keeps its source's own verdict under that source's documented rule (column `verdict_source`). The uniform 3330 class (`harmonized_class`) is descriptive: UNINFORMATIVE if the 95% interval's log width is at least 0.8 x 2.7171 (prior-dominated) or more than 2.5x the matching autism interval's; otherwise BIAS SIGNAL if the interval excludes 1; otherwise CONSISTENT. The per-context summary (bias signal on at least one control / consistent / uninformative) is a descriptive tally of that class.

Round 2 was added under the dated amendment `qa/negative_controls_20260925/3330_amendment_round2.json` (sha256 `2562329cdea367610ee32d47ae678a016b651290a5dd5d1c43a6eea33aa65277`), written 2026-09-30 02:54 Central Daylight Time (2026-09-30T07:54Z), before 07_analysis_code/3330_negative_control_summary.py was extended for round 2 and before any round-2 3330 quantity was computed: no round-2 row had been passed through 3330's harmonize(), no per-context tally, calibration set, systematic-error estimate, calibrated interval, CSV row or figure containing a round-2 row had been produced by 3330. Like the original rule it was written with every round-2 estimate and verdict known; it fixed which round-2 rows are carried and how, that round-1 rows are superseded only where a source says so (none does, so every round-1 row is carried beside round 2), the new calibration sets S7, S7b and S8 with their expected members (asserted), and the descriptive additions (the I5 'wide, excludes 1' tally and the reading-row tally). C_INJ and S1-S6 are computed exactly as on 2026-09-29 from round-1 rows; the script asserts that they reproduce the earlier preview.

## Rows carried

549 rows (452 fitted) from 25 sources, 204 of them from round 2: 3326 9, 3331 12, 3332 50, 3333 72, 3334 2, 3335 12, 3336 4, 3337 1, 3338 2, 3339 1, 3340 22, 3341 17, 3342 5, 3343 9, 3344 4, 3345 18, 3346 4, 3347 63, 3348 38, 3380 59, 3381 94, 3382 30, 3383 9, 3384 2, 3385 10. 21 rows dropped as duplicates (`duplicates_dropped.csv`: 3336 refits of 3335, the 3338 hearing refit of 3337, and the 3347 England rows that equal 3333 cell set S; no round-2 row duplicates another). Uniform classes over fitted rows: BIAS SIGNAL 28, CONSISTENT 195, UNINFORMATIVE 229 (round 2 alone: BIAS SIGNAL 12, CONSISTENT 81, UNINFORMATIVE 87).

Columns: r and 95% interval are the source's posterior median and CrI (Wald CI for the NSCH/SLAITS prevalence ratios, placebo-calibrated interval for the CASEMIS jumps, conformal interval for SDID); ESS is the pooled importance-sampling ESS where the source sampled and the Kish effective number of grid points where it used grid quadrature (column `ess_basis`); the matching autism estimate is autism refitted on the same cells when the source did that, otherwise the published estimate (column `autism_basis`); DDD is autism relative to the control (fitted log-ratio where the source fitted one, otherwise the contrast of independent posteriors; column `ddd_basis`). The CSV keeps the 43 columns and conventions that 3370 reads; its amendment-1 loader was run on it read-only: all 6 round-2 files classified from the harmonized rows, 0 unclassified; 139 round-2 NC records (25 PRIMARY and 58 SECONDARY pass the numerical rule); excluded by its classifier: composite 2, drowning 12, duplicate 21, immunization 4, not_fitted 24, sudi 2.

## Hierarchy contexts: tallies, reading rows and every bias-signal row

| context (own-data autism) | uniform tally, any row (rounds 1+2) | reading rows only | round 1 only | wide, excludes 1 (I5) |
|---|---|---|---|---|
| U.S. PCV (Core) (1.17 (0.97-1.39)) | BIAS SIGNAL on at least one control: IDEA hearing impairment (3326) 1.41 (1.05-1.91) | BIAS SIGNAL on at least one control | BIAS SIGNAL on at least one control | Unintentional injury (V01-X59), ages 0-5 (3383) 2.59 (1.15-5.76) |
| U.S. RotaShield (Core) (0.87 (0.36-1.83)) | CONSISTENT (no control signals) | UNINFORMATIVE (no control is informative) | CONSISTENT (no control signals) | - |
| U.S. Rotavirus 2006 (Core) (0.95 (0.76-1.19)) | BIAS SIGNAL on at least one control: IDEA hearing impairment (3326) 1.41 (1.07-1.82); IDEA traumatic brain injury (3331) 2.37 (1.09-4.58) | BIAS SIGNAL on at least one control | BIAS SIGNAL on at least one control | - |
| Japan MMR (Core) (0.85 (0.28-2.32)) | CONSISTENT (no control signals) | CONSISTENT (no control signals) | CONSISTENT (no control signals) | - |
| Sweden PCV (1.01 (0.79-1.56)) | BIAS SIGNAL on at least one control: Lumbar or Pelvic Fracture (S32) (3381) 2.02 (1.09-3.58); Shoulder or Upper-Arm Fracture (S42) (3381) 1.12 (1.02-1.21) | CONSISTENT (no control signals) | CONSISTENT (no control signals) | - |
| Germany PCV (1.03 (0.27-3.96)) | CONSISTENT (no control signals) | CONSISTENT (no control signals) | CONSISTENT (no control signals) | - |
| Hong Kong PCV (0.89 (0.66-1.16)) | BIAS SIGNAL on at least one control: SEN physical + visual (3342) 1.50 (1.09-2.22); SEN physical disability (3342) 1.78 (1.23-2.90); Road-Traffic Casualties (Police-Reported) (3385) 0.66 (0.48-0.92) | BIAS SIGNAL on at least one control | BIAS SIGNAL on at least one control | - |
| England Rotavirus (0.94 (0.86-1.02)) | BIAS SIGNAL on at least one control: Poisoning (T36-T65) (3380) 0.41 (0.19-0.85) | CONSISTENT (no control signals) | UNINFORMATIVE (no control is informative) | - |
| Scotland Rotavirus (1.55 (0.38-5.31)) | CONSISTENT (no control signals) | CONSISTENT (no control signals) | CONSISTENT (no control signals) | - |
| Northern Ireland Rotavirus (1.01 (0.88-1.16)) | CONSISTENT (no control signals) | CONSISTENT (no control signals) | UNINFORMATIVE (no control is informative) | - |
| Canada Rotavirus (Excl. Quebec) (0.99 (0.32-1.89)) | CONSISTENT (no control signals) | CONSISTENT (no control signals) | NO CONTROL | - |
| Quebec Rotavirus (1.07 (0.74-1.48)) | UNINFORMATIVE (no control is informative) | NO READING ROW (analogue only) | NO CONTROL | - |
| Czechia MMR (1.04 (0.32-3.66)) | UNINFORMATIVE (no control is informative) | UNINFORMATIVE (no control is informative) | NO CONTROL | - |
| Brazil Maternal DTaP (0.77 (0.51-1.14)) | BIAS SIGNAL on at least one control: All Injuries (3332) 0.51 (0.29-0.83); Falls (3332) 0.38 (0.15-0.86); All Unintentional Injuries (V01-X59) (3382) 0.36 (0.15-0.79) | BIAS SIGNAL on at least one control | BIAS SIGNAL on at least one control | - |
| Colombia Varicella (0.76 (0.51-1.07)) | BIAS SIGNAL on at least one control: Transport + Accidental Injuries (3385) 1.33 (1.01-1.96) | CONSISTENT (no control signals) | CONSISTENT (no control signals) | - |
| Tanzania Schedule (1.12 (0.54-1.69)) | CONSISTENT (no control signals) | CONSISTENT (no control signals) | CONSISTENT (no control signals) | - |

| context | control (source; cells / design) | cells | r (95% interval) | P(r>1) | source verdict | 3330 class | matching autism (basis) | DDD autism / control | future+3 |
|---|---|---|---|---|---|---|---|---|---|
| U.S. PCV (Core) | IDEA hearing impairment (3326; core-v2 observed cells (joint U.S. likelihood)) | 1381 | 1.41 (1.05-1.91) | 0.990 | SIMILAR | BIAS SIGNAL (r > 1 (harm direction)) | 1.17 (0.97-1.39) (all core cells) | 0.79 (0.56-1.12) | PASS 0.97 (0.63-1.74) |
| U.S. PCV (Core) | IDEA orthopedic impairment (3326; core-v2 observed cells (joint U.S. likelihood)) | 1261 | 1.43 (0.91-2.26) | 0.937 | UNINFORMATIVE | UNINFORMATIVE | 1.17 (0.97-1.39) (all core cells) | 0.81 (0.54-1.23) | PASS 0.59 (0.39-1.05) |
| U.S. PCV (Core) | IDEA visual impairment (3326; core-v2 observed cells (joint U.S. likelihood)) | 1073 | 1.45 (0.74-2.72) | 0.855 | UNINFORMATIVE | UNINFORMATIVE | 1.17 (0.97-1.39) (all core cells) | 0.90 (0.61-1.34) | PASS 0.95 (0.54-2.07) |
| U.S. PCV (Core) | IDEA traumatic brain injury (3331; core-v2 observed cells) | 472 | 1.65 (0.77-4.29) | 0.892 | UNINFORMATIVE | UNINFORMATIVE | 1.11 (0.80-1.49) (same cells) | 0.33 (0.20-0.86) | PASS 1.14 (0.45-3.78) |
| U.S. PCV (Core) | IDEA traumatic brain injury (3331; core-v2 extended cells) | 977 | 0.84 (0.46-1.35) | 0.248 | UNINFORMATIVE | UNINFORMATIVE | 1.06 (0.91-1.22) (same cells) | 1.44 (0.74-2.85) | PASS 1.20 (0.55-3.45) |
| U.S. PCV (Core) | Unintentional injury (V01-X59), ages 0-5 (3383; M1: Unintentional injury (V01-X59), ages 0-5 (state x birth-cohort Lexis cells, deaths...) | 1107 | 2.59 (1.15-5.76) | 0.989 | UNINFORMATIVE | UNINFORMATIVE (r > 1 (harm direction)) [wide, excludes 1] | 1.17 (0.97-1.39) (all core cells) | 1.03 (0.69-1.52) | FAIL 0.18 (0.10-0.39) |
| U.S. RotaShield (Core) | IDEA traumatic brain injury (3331; core-v2 observed cells) | 472 | 0.69 (0.20-2.21) | 0.286 | CONSISTENT | UNINFORMATIVE | 1.01 (0.41-2.13) (same cells) | 1.58 (0.45-3.75) | n/a (PCV-path placebo only) |
| U.S. RotaShield (Core) | Unintentional injury (V01-X59), ages 0-5 (3383; M1: Unintentional injury (V01-X59), ages 0-5 (state x birth-cohort Lexis cells, deaths...) | 1107 | 0.82 (0.17-3.13) | 0.366 | CONSISTENT | UNINFORMATIVE | 0.87 (0.36-1.83) (all core cells) | 1.31 (0.48-2.84) | n/a (PCV-path placebo only) |
| U.S. Rotavirus 2006 (Core) | IDEA hearing impairment (3326; core-v2 observed cells (joint U.S. likelihood)) | 1381 | 1.41 (1.07-1.82) | 0.991 | not classified (the 3326 rule covers the PCV r only) | BIAS SIGNAL (r > 1 (harm direction)) | 0.95 (0.76-1.19) (all core cells) | 0.76 (0.60-1.03) | n/a (PCV-path placebo only) |
| U.S. Rotavirus 2006 (Core) | IDEA traumatic brain injury (3331; core-v2 observed cells) | 472 | 2.37 (1.09-4.58) | 0.985 | BIAS SIGNAL | BIAS SIGNAL (r > 1 (harm direction)) | 0.75 (0.55-1.01) (same cells) | 0.68 (0.39-1.29) | n/a (PCV-path placebo only) |
| Japan MMR (Core) | All unintentional injuries (3341; 9 birth cohorts 1988-1996, deaths at ages 1-4, national) | 9 | 0.86 (0.46-1.61) | 0.307 | CONSISTENT | CONSISTENT | 0.85 (0.28-2.32) (Kohoku, same cohorts) | 0.86 (0.28-2.35) | PASS 0.77 (0.30-1.57) |
| Japan MMR (Core) | All injuries (3347; the 9 Honda cohorts (national)) | 9 | 0.89 (0.48-1.67) | 0.353 | CONSISTENT | CONSISTENT | 0.85 (0.28-2.32) (published (other cells)) | 0.95 (0.27-3.09) | PASS 0.73 (0.29-1.46) |
| Japan MMR (Core) | Accidents (不慮の事故) (3385; nine Kohoku birth cohorts 1988-1996, Yokohama deaths at ages 15-19) | 9 | 1.00 (0.41-2.21) | 0.500 | UNINFORMATIVE | CONSISTENT | 0.85 (0.28-2.32) (Kohoku, same cohorts) | 0.84 (0.27-2.35) | PASS 0.68 (0.24-1.57) |
| Sweden PCV | Unintentional injuries (3347; closest to the autism cells) | 34 | 0.71 (0.32-1.62) | 0.197 | CONSISTENT | CONSISTENT | 1.01 (0.79-1.56) (published (other cells)) | 1.45 (0.60-3.46) | PASS 0.70 (0.32-1.73) |
| Sweden PCV | All Injuries (S00-T98 Excl. T78, T80-T88) (3381; SWE_A (reading): SWE_PCV_corrected cells) | 756 | 1.06 (0.98-1.13) | 0.932 | CONSISTENT | CONSISTENT | 1.01 (0.79-1.56) (same cells) | 0.97 (0.74-1.56) | PASS 1.00 (0.89-1.07) |
| Sweden PCV | Lumbar or Pelvic Fracture (S32) (3381; SWE_A (reading): SWE_PCV_corrected cells) | 126 | 2.02 (1.09-3.58) | 0.986 | UNINFORMATIVE | BIAS SIGNAL (r > 1 (harm direction)) | 1.47 (0.63-2.94) (same cells) | n/a | PASS 0.80 (0.35-1.57) |
| Sweden PCV | Shoulder or Upper-Arm Fracture (S42) (3381; SWE_A (reading): SWE_PCV_corrected cells) | 756 | 1.12 (1.02-1.21) | 0.986 | UNINFORMATIVE | BIAS SIGNAL (r > 1 (harm direction)) | 1.01 (0.79-1.56) (same cells) | n/a | FAIL 1.09 (0.97-1.16) |
| Germany PCV | Unintentional injuries (3347; ages 1-4 + 5-9 Lexis window) | 55 | 1.01 (0.50-1.81) | 0.515 | CONSISTENT | CONSISTENT | 1.03 (0.27-3.96) (published (other cells)) | 1.04 (0.23-4.65) | PASS 1.21 (0.69-2.28) |
| Germany PCV | All Injuries (S00-T98 Excl. T78, T80-T88) (3381; DEU_A (reading): pooled DEU_PCV, frozen 36 cells, 2000-2023 discharges) | 36 | 1.09 (0.31-3.91) | 0.553 | UNINFORMATIVE | UNINFORMATIVE | 1.03 (0.27-3.96) (same cells) | 1.05 (0.27-4.04) | NOT TESTABLE |
| Hong Kong PCV | SEN physical + visual + hearing (3342; grades P1-S6 x school years 2015/16-2019/20 (60 cells)) | 60 | 1.15 (0.90-1.47) | 0.893 | CONSISTENT | CONSISTENT | 0.89 (0.66-1.16) (same cells) | 0.86 (0.64-1.14) | PASS 0.895 (0.710-1.216) |
| Hong Kong PCV | SEN physical + visual (3342; grades P1-S6 x school years 2015/16-2019/20 (60 cells)) | 60 | 1.50 (1.09-2.22) | 0.992 | BIAS SIGNAL | BIAS SIGNAL (r > 1 (harm direction)) | 0.89 (0.66-1.16) (same cells) | 0.65 (0.45-0.90) | FAIL 0.581 (0.406-0.857) |
| Hong Kong PCV | SEN physical disability (3342; grades P1-S6 x school years 2015/16-2019/20 (60 cells)) | 60 | 1.78 (1.23-2.90) | 0.997 | BIAS SIGNAL | BIAS SIGNAL (r > 1 (harm direction)) | 0.89 (0.66-1.16) (same cells) | 0.55 (0.38-0.77) | PASS 0.685 (0.382-4.206) |
| Hong Kong PCV | All injuries (3343; ages 1-4 pooled + 5-9 x 2002-2018 (34 cells; deaths > 0)) | 32 | 0.89 (0.26-3.74) | 0.432 | UNINFORMATIVE | UNINFORMATIVE | 0.89 (0.66-1.16) (published (other cells)) | 1.00 (0.23-3.54) | FAIL 2.97 (1.19-7.96) |
| Hong Kong PCV | Road-Traffic Casualties (Police-Reported) (3385; WHO bands 5-9/10-14/15-19 x calendar year (bands_2009_2024)) | 48 | 0.66 (0.48-0.92) | 0.008 | BIAS SIGNAL | BIAS SIGNAL (r < 1 (protective direction)) | 0.89 (0.66-1.16) (published (other cells)) | 1.34 (0.86-2.05) | PASS 0.88 (0.56-1.64) |
| England Rotavirus | All unintentional injuries (V01-X59 excl. X30-X39) (3333; cell set S: school-age bands 5-9/10-14 (closest to the autism cells)) | 28 | 0.57 (0.25-1.22) | 0.074 | UNINFORMATIVE | UNINFORMATIVE | 0.94 (0.86-1.02) (autism refit (other cells)) | 1.64 (0.77-3.81) | FAIL 0.34 (0.15-0.79) |
| England Rotavirus | All unintentional injuries (V01-X59 excl. X30-X39) (3333; cell set E: extended Lexis cells (ages 1-4 single + 5-9)) | 113 | 0.80 (0.46-1.32) | 0.194 | UNINFORMATIVE | UNINFORMATIVE | 0.94 (0.86-1.02) (autism refit (other cells)) | 1.18 (0.71-2.06) | PASS 0.80 (0.45-1.44) |
| England Rotavirus | Injuries, foreign bodies, burns, frostbite (S00-T35) (3380; C1 HES Diagnosis, cell set S (2013-2025; ages 5-9, 10-14)) | 23 | 0.99 (0.57-1.57) | 0.489 | CONSISTENT | CONSISTENT | 0.89 (0.53-1.38) (same cells) | 0.90 (0.45-1.82) | PASS 1.12 (0.67-1.93) |
| England Rotavirus | Poisoning (T36-T65) (3380; C1 HES Diagnosis, cell set S (2013-2025; ages 5-9, 10-14)) | 23 | 0.41 (0.19-0.85) | 0.008 | BIAS SIGNAL | BIAS SIGNAL (r < 1 (protective direction)) | 0.89 (0.53-1.38) (same cells) | 2.15 (0.90-5.28) | PASS 2.53 (0.32-7.08) |
| England Rotavirus | Injuries, foreign bodies, burns, frostbite (S00-T35) (3380; C1 HES Diagnosis, cell set E (2013-2025; ages 1-4, 5-9, 10-14)) | 38 | 0.96 (0.69-1.29) | 0.384 | UNINFORMATIVE | UNINFORMATIVE | 0.94 (0.86-1.02) (published (other cells)) | 0.99 (0.72-1.39) | PASS 1.04 (0.72-1.55) |
| England Rotavirus | Unintentional external causes (V01-X59 excl. X30-X39) (3380; C2 HES External Cause, cell set S (2013-2025; ages 5-9, 10-14)) | 21 | 0.98 (0.53-1.60) | 0.463 | CONSISTENT | CONSISTENT | 0.91 (0.52-1.43) (same cells) | 0.93 (0.44-1.98) | PASS 1.17 (0.68-2.15) |
| England Rotavirus | Unintentional external causes (V01-X59 excl. X30-X39) (3380; C2 HES External Cause, cell set E (2013-2025; ages 1-4, 5-9, 10-14)) | 35 | 0.98 (0.69-1.34) | 0.462 | UNINFORMATIVE | UNINFORMATIVE | 0.94 (0.86-1.02) (published (other cells)) | 0.96 (0.70-1.38) | PASS 1.10 (0.74-1.70) |
| England Rotavirus | All road casualties (3380; C3 STATS19 England, cell set S (2011-2024; ages 5, 6, 7, 8, 9, 10, 11, 12)) | 112 | 1.00 (0.87-1.16) | 0.511 | CONSISTENT | CONSISTENT | 0.97 (0.89-1.06) (same cells) | 0.97 (0.82-1.15) | PASS 0.99 (0.88-1.12) |
| England Rotavirus | All road casualties (3380; C3 STATS19 England, cell set E (2011-2024; ages 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12,...) | 195 | 1.06 (0.94-1.19) | 0.836 | CONSISTENT | CONSISTENT | 0.94 (0.86-1.02) (published (other cells)) | 0.89 (0.77-1.02) | PASS 1.04 (0.95-1.14) |
| Scotland Rotavirus | All unintentional injuries (V01-X59 excl. X30-X39) (3333; cell set S: school-age bands 5-9/10-14 (closest to the autism cells)) | 33 | 0.88 (0.24-3.07) | 0.423 | UNINFORMATIVE | UNINFORMATIVE | 1.55 (0.38-5.33) (autism refit (other cells)) | 1.74 (0.27-10.60) | PASS 0.66 (0.20-2.19) |
| Scotland Rotavirus | All unintentional injuries (V01-X59 excl. X30-X39) (3333; cell set E: extended Lexis cells (ages 1-4 single + 5-9)) | 90 | 0.87 (0.32-2.31) | 0.393 | CONSISTENT | CONSISTENT | 1.55 (0.38-5.33) (autism refit (other cells)) | 1.77 (0.32-8.78) | PASS 0.80 (0.28-2.22) |
| Scotland Rotavirus | Unintentional injuries (3347; closest to the autism cells) | 22 | 0.91 (0.25-3.25) | 0.445 | UNINFORMATIVE | UNINFORMATIVE | 1.55 (0.38-5.33) (published (other cells)) | 1.68 (0.25-10.41) | PASS 0.68 (0.20-2.30) |
| Scotland Rotavirus | All unintentional-injury admissions (3380; C4 PHS Admissions, cell set S (2015-2024; ages 5-14)) | 306 | 0.92 (0.26-3.05) | 0.446 | UNINFORMATIVE | UNINFORMATIVE | 1.23 (0.31-4.45) (same cells) | 1.34 (0.21-8.20) | PASS 0.93 (0.28-3.21) |
| Scotland Rotavirus | All unintentional-injury admissions (3380; C4 PHS Admissions, cell set E (2015-2024; ages 0-4, 5-9, 10-14)) | 960 | 1.11 (0.42-2.75) | 0.585 | CONSISTENT | CONSISTENT | 1.55 (0.38-5.33) (published (other cells)) | 1.40 (0.26-6.69) | PASS 1.05 (0.41-2.65) |
| Scotland Rotavirus | All road casualties aged 5-18 (3380; C5 STATS19 Scotland, cell set S (2010-2025; ages 5-18)) | 490 | 0.75 (0.22-2.42) | 0.317 | UNINFORMATIVE | UNINFORMATIVE | 1.55 (0.38-5.33) (same cells) | 2.06 (0.33-11.93) | PASS 0.68 (0.20-2.16) |
| Scotland Rotavirus | All road casualties aged 5-18 (3380; C5 STATS19 Scotland, cell set E (2010-2025; ages 5-18)) | 512 | 0.75 (0.22-2.42) | 0.318 | UNINFORMATIVE | UNINFORMATIVE | 1.55 (0.38-5.33) (published (other cells)) | 2.06 (0.33-11.92) | PASS 0.68 (0.21-2.16) |
| Northern Ireland Rotavirus | All unintentional injuries (V01-X59 excl. X30-X39) (3333; cell set S: school-age bands 5-9/10-14 (closest to the autism cells)) | 30 | 0.86 (0.25-2.76) | 0.400 | UNINFORMATIVE | UNINFORMATIVE | 1.01 (0.88-1.16) (autism refit (other cells)) | 1.18 (0.36-4.02) | PASS 1.70 (0.46-5.00) |
| Northern Ireland Rotavirus | All unintentional injuries (V01-X59 excl. X30-X39) (3333; cell set E: extended Lexis cells (ages 1-4 single + 5-9)) | 52 | 1.46 (0.51-3.51) | 0.778 | UNINFORMATIVE | UNINFORMATIVE | 1.01 (0.88-1.16) (autism refit (other cells)) | 0.69 (0.29-1.99) | PASS 1.98 (0.82-4.51) |
| Northern Ireland Rotavirus | All road casualties (3380; C7 PSNI Casualties, cell set S (2013-2024; ages 0-9, 10-16)) | 22 | 1.02 (0.37-2.04) | 0.520 | CONSISTENT | CONSISTENT | 0.65 (0.29-1.32) (same cells) | 0.64 (0.22-2.15) | PASS 1.16 (0.55-1.92) |
| Northern Ireland Rotavirus | All road casualties (3380; C7 PSNI Casualties, cell set E (2013-2024; ages 0-9, 10-16)) | 24 | 1.16 (0.48-2.17) | 0.650 | UNINFORMATIVE | UNINFORMATIVE | 1.01 (0.88-1.16) (published (other cells)) | 0.88 (0.46-2.11) | PASS 1.12 (0.53-1.85) |
| Canada Rotavirus (Excl. Quebec) | Asthma incidence (CCDSS, 1-19) (3384; CAN_noQC province x fiscal year, ages 1-19 (the block's 166 cells)) | 166 | 0.71 (0.48-1.06) | 0.036 | CONSISTENT | CONSISTENT | 0.99 (0.32-1.89) (same cells) | 1.34 (0.47-2.20) | PASS 1.28 (0.72-1.93) |
| Quebec Rotavirus | no control held | | | | | | 1.07 (0.74-1.48) (own data) | | |
| Czechia MMR | All Injuries (S00-T98 Excl. T78, T80-T88) (3381; CZE_A (reading): CZE_MMR_counts model, 77 districts x ages 5-9 x 2019-2024) | 462 | 1.19 (0.32-4.48) | 0.602 | UNINFORMATIVE | UNINFORMATIVE | 0.91 (0.27-3.37) (autism refit (other cells)) | n/a | NOT TESTABLE |
| Brazil Maternal DTaP | All Injuries (3332; 27 UF x census bands 0-4/5-9/10-14/15-19, admissions 2021-2023 (census-cohort age)) | 108 | 0.51 (0.29-0.83) | 0.003 | BIAS SIGNAL | BIAS SIGNAL (r < 1 (protective direction)) | 0.77 (0.51-1.14) (same cells) | 1.52 (0.96-2.38) | FAIL 0.63 (0.43-0.90) |
| Brazil Maternal DTaP | Falls (3332; 27 UF x census bands 0-4/5-9/10-14/15-19, admissions 2021-2023 (census-cohort age)) | 108 | 0.38 (0.15-0.86) | 0.010 | BIAS SIGNAL | BIAS SIGNAL (r < 1 (protective direction)) | 0.77 (0.51-1.14) (same cells) | 2.52 (1.10-5.34) | FAIL 0.45 (0.20-0.89) |
| Brazil Maternal DTaP | Unintentional injuries (3347; ages 1-4 + 5-9 Lexis window) | 80 | 1.08 (0.79-1.40) | 0.701 | CONSISTENT | CONSISTENT | 0.77 (0.51-1.14) (published (other cells)) | 0.71 (0.44-1.18) | PASS 0.83 (0.52-1.20) |
| Brazil Maternal DTaP | All Unintentional Injuries (V01-X59) (3382; 27 UF x census bands 0-4/5-9/10-14/15-19, SIM deaths 2021-2023 (age at death)) | 108 | 0.36 (0.15-0.79) | 0.005 | BIAS SIGNAL | BIAS SIGNAL (r < 1 (protective direction)) | 0.77 (0.51-1.14) (same cells) | 1.92 (0.97-3.57) | FAIL 0.31 (0.15-0.57) |
| Colombia Varicella | School census hearing disability (3337; 28 regions x 3 levels x 2019-2024, observed cells) | 497 | 0.98 (0.66-1.49) | 0.450 | CONSISTENT | CONSISTENT | 0.75 (0.49-1.04) (same cells) | 0.79 (0.51-1.18) | PASS 1.48 (0.42-4.37) |
| Colombia Varicella | School census motor / physical disability (3338; 28 regions x 3 levels x 2019-2024, observed cells) | 502 | 0.87 (0.67-1.28) | 0.174 | CONSISTENT | CONSISTENT | 0.77 (0.51-1.08) (same cells) | 0.90 (0.53-1.30) | PASS 2.09 (0.30-7.98) |
| Colombia Varicella | School census visual disability (3338; 28 regions x 3 levels x 2019-2024, observed cells) | 501 | 0.90 (0.58-1.27) | 0.273 | CONSISTENT | CONSISTENT | 0.77 (0.51-1.08) (same cells) | 0.88 (0.56-1.45) | PASS 1.92 (0.39-6.40) |
| Colombia Varicella | Unintentional injury deaths (3347; reading cell set (not fitted)) |  | not fitted: (c) not fitted: no held age-specific population for 2007-2023 |  | UNINFORMATIVE | NOT FITTED | 0.76 (0.51-1.07) (published (other cells)) | n/a |  |
| Colombia Varicella | Transport Injuries (INMLCF, Non-Fatal) (3385; region x level x school year 2019-2024, observed cells (DANE band population)) | 500 | 1.36 (0.99-1.93) | 0.973 | CONSISTENT | CONSISTENT | 0.76 (0.51-1.07) (same cells) | 0.63 (0.39-1.02) | PASS 1.07 (0.35-3.28) |
| Colombia Varicella | Transport + Accidental Injuries (3385; region x level x school year 2019-2024, observed cells (DANE band population)) | 501 | 1.33 (1.01-1.96) | 0.979 | BIAS SIGNAL | BIAS SIGNAL (r > 1 (harm direction)) | 0.76 (0.51-1.07) (same cells) | 0.63 (0.40-0.99) | PASS 0.92 (0.34-2.72) |
| Tanzania Schedule | Pupils with albinism (3339; Std I-VII x 7 school years (2016, 2017, 2020-2024)) | 49 | 1.00 (0.69-1.28) | 0.510 | CONSISTENT | CONSISTENT | 1.12 (0.54-1.69) (same cells) | 1.06 (0.59-1.61) | PASS 1.038 (0.672-1.346) |

## What the negative controls say, context by context

**U.S. PCV (core, pooled; own-data autism 1.17 (0.97-1.39)).** The injury control on the core's own cells is IDEA traumatic brain injury, published in 472 of the 1,469 core-v2 cells: 1.65 (0.77-4.29) (UNINFORMATIVE; its interval is 2.8x as wide as autism's 1.11 (0.80-1.49) on the same cells), and it moves to 0.83 (0.42-1.59) on the complete-series cells and 0.84 (0.46-1.35) on the extended single-age cells (UNINFORMATIVE). The IDEA disability categories that the same likelihood uses as negative controls lean at least as far as autism: hearing impairment 1.41 (1.05-1.91) (3326 class SIMILAR; its interval excludes 1), orthopedic 1.43 (0.91-2.26) (uniform width ratio 2.52, just over the 2.5 limit against autism on all core cells; against autism on the control's own cells it would fall below it: a borderline, review M6), visual 1.45 (0.74-2.72); 4 of the 4 IDEA control medians (TBI included) lie above autism's 1.17. Outside the core, the NSCH legacy cohorts give parent-reported brain injury 1.12 (1.02-1.23) (prevalence ratio per +10 pp; BIAS SIGNAL) against autism 1.10 (0.93-1.32) in the same children; in the modern NSCH cohorts both outcomes sit below 1 (brain injury 0.81 (0.63-1.04) for PCV 4+, 0.85 (0.68-1.06) for the combined series), and the SLAITS toddler-injury control is 1.01 (0.94-1.08) (CONSISTENT). The national benchmark (WHO U.S. deaths, national PCV7 step) is 0.99 (0.66-1.34) (CONSISTENT). Round 2 (3383) fitted NCHS unintentional injury deaths at ages 0-5 on state x birth-cohort Lexis cells of the same PCV path: 2.59 (1.15-5.76), P(r > 1) 0.989, class UNINFORMATIVE (4.4x autism's interval) although the interval excludes 1, and its own future+3 placebo FAILs (0.18 (0.10-0.39)): the PCV path shifted three cohorts also 'finds' a large effect on these deaths, in the other direction. The complete-series and ages 1-5 cell rules give 2.47 (1.10-5.76) and 2.69 (1.23-5.76) (both with failed placebos), the age-band rule 1.08 (0.50-2.78), all external causes 1.49 (0.66-3.60); the head-injury fit is degenerate (pooled ESS 1.0; carried as not fitted). Cohort-aligned DDD on the core's own units (autism relative to injury deaths): 1.03 (0.69-1.52). Reading: on the U.S. state x cohort cells the design returns non-vaccine associations of this size in both directions, and every control precise enough to read leans at least as far as autism, so the core's lean cannot be attributed to autism specifically. The 3326 rule's own verdict remains MIXED, which draws neither conclusion.

**U.S. RotaShield (core, pooled; own data 0.87 (0.36-1.83)).** IDEA TBI on the same cells gives 0.69 (0.20-2.21), prior-dominated (CONSISTENT under 3331; 3330 class UNINFORMATIVE); hearing 1.41 (0.48-3.16); the round-2 NCHS injury deaths 0.82 (0.17-3.13) (CONSISTENT by 3383, prior-dominated, 3330 class UNINFORMATIVE). The controls carry no information about this context's bias.

**U.S. rotavirus 2006 (inside the U.S. likelihood, untied; own data 0.95 (0.76-1.19)).** The same fit returns a harm-direction 'effect' on IDEA TBI, 2.37 (1.09-4.58) (BIAS SIGNAL), while autism on the same cells is 0.75 (0.55-1.01); hearing 1.41 (1.07-1.82), orthopedic 1.01 (0.70-1.46), visual 1.16 (0.78-1.71). A design that gives a rotavirus 'effect' on brain injury and hearing loss is most plausibly picking up state x cohort drift in IDEA identification, which fits the lead decision to keep RV2006 untied from the pool. The round-2 death cells cannot test it: rotavirus-2006 exposure is 0 on every death cohort, so 3383's 1.17 (0.33-4.49) is the prior (NOT IDENTIFIED).

**Japan MMR (Kohoku, pooled; own data 0.85 (0.28-2.32)).** National unintentional injury deaths at ages 1-4 of the same nine birth cohorts give 0.86 (0.46-1.61) (CONSISTENT); the secondary outcomes are 7 CONSISTENT, 1 UNINFORMATIVE; the independent 3347 refit of all injuries is 0.89 (0.48-1.67). Round 2 (3385) used Yokohama City accident deaths of the same cohorts at ages 15-19 (the only ages the city's cause series covers): 1.00 (0.41-2.21), UNINFORMATIVE by 3385 (b: thin (minimum cohort count 6.2 < 16)); the uniform 3330 class, which has no thinness criterion, reads it CONSISTENT. The injury medians sit where autism's does, so the controls neither support nor undercut reading the Kohoku estimate as an MMR effect.

**Sweden PCV (pooled; own data 1.01 (0.79-1.56)).** Round 2 (3381) put Socialstyrelsen patient-register injuries on the block's own cells: all injuries 1.06 (0.98-1.13), head injuries 1.04 (0.94-1.15), fractures 1.04 (0.97-1.11) (all CONSISTENT; 3381 reading: CONSISTENT), against autism 1.01 (0.79-1.56) on the same cells; the all-injury interval is 0.21x autism's, so this is an informative control. The uniform 3330 class, which has no thinness criterion, flags lumbar or pelvic fracture (S32) 2.02 (1.09-3.58); shoulder or upper-arm fracture (S42) 1.12 (1.02-1.21) (3381: UNINFORMATIVE, thin), so Sweden counts as 'bias signal on at least one control' in the any-row tally but not on its reading row. The external-cause series on 3-year bands (beside the reading) give all accidents 1.03 (0.98-1.08) (SVOV) and 0.99 (0.89-1.06) (inpatient only; future+3 FAIL). The round-1 national WHO deaths 0.71 (0.32-1.62) test only the national step (analogue).

**Germany PCV (pooled; own data 1.03 (0.27-3.96)).** Round 2 (3381) fitted Destatis hospital injuries 2000-2023 on the pooled block's frozen 36 cells: all 7 primary controls are prior-dominated (all injuries 1.09 (0.31-3.91), autism on the same cells 1.03 (0.27-3.96); reading UNINFORMATIVE), as the block's own autism estimate is. National WHO deaths give 1.01 (0.50-1.81) (national step only). On the report-only rebuilt panel the 825-cell design gives all injuries 1.12 (0.94-1.32) (CONSISTENT), with other external causes incl. drowning (T75) 0.63 (0.37-0.99) the one bias signal among its secondary rows; its 2024-only cells (3348) left all 7 primary controls prior-dominated (all injuries 0.74 (0.23-3.11)), and 3348's design-B shoulder or upper-arm fracture (S42) 0.22 (0.08-0.65) comes from a design whose harm-injection medians are 0.208, 0.246, 0.264, 0.282 for true r = 1, 2, 3, 5. Nothing held tests the pooled German cells with useful precision.

**Hong Kong PCV (pooled; own data 0.89 (0.66-1.16)).** On the block's own 60 grade x school-year cells the EDB special-needs categories give two correlated harm-direction signals, physical disability 1.78 (1.23-2.90) and physical + visual 1.50 (1.09-2.22) (nested: the second contains the first), while the preferred composite physical + visual + hearing is 1.15 (0.90-1.47) (CONSISTENT), hearing 0.99 (0.72-1.32) and visual 2.01 (0.56-6.06) (UNINFORMATIVE). The rule that classes them was not blind: it was written with the deciding estimate (physical disability) already known, and 3342 labels its reading accordingly rather than as documented. 3342's own null calibration shows that with birth-cohort noise of the observed size the physical-disability value sits at a 12.0%-14.8% two-sided tail and that the design's nominal 95% intervals under-cover (autism's excludes 1 in 13.5% of null simulations). WHO injury deaths are too sparse to test anything (0.89 (0.26-3.74), UNINFORMATIVE). Round 2 (3385) adds the first informative injury control, police-reported road-traffic casualties by age band and year: 0.66 (0.48-0.92) (BIAS SIGNAL, protective direction; 1.15x autism's interval, different cells; future+3 PASS), which is window-dependent (0.98 (0.59-1.63) on 2013-2021, 0.74 (0.57-0.94) without 2020-2022; pedestrians 0.76 (0.52-1.15)). Reading: the design's spurious 'effects' now run in both directions, about 0.4-0.6 on the log scale, which is non-specific noise within the 0.70 context bias SD the hierarchy already assigns to Hong Kong; its autism 0.89 is not interpretable as a vaccine effect on its own.

**England rotavirus (pooled; own data 0.94 (0.86-1.02)).** Round 2 (3380) gives the first informative injury controls: STATS19 police-recorded road casualties on 112 of the block's own age x year cells, 1.00 (0.87-1.16) (CONSISTENT; 1.62x the interval of autism refitted on the same cells, 0.97 (0.89-1.06); future+3 PASS; harm gate PASS), and 1.06 (0.94-1.19) on the extended single-age cells (CONSISTENT); HES admissions on the HES band layout, diagnoses 0.99 (0.57-1.57) and external causes 0.98 (0.53-1.60) (both CONSISTENT, against same-layout autism 0.89 (0.53-1.38)). Of the 13 HES categories (diagnosis and external-cause axes) on those cells, poisoning (T36-T65) 0.41 (0.19-0.85) is a bias signal (protective direction), so England counts as 'bias signal on at least one control' in the any-row tally but not on a reading row. The round-1 WHO England + Wales deaths (0.57 (0.25-1.22), UNINFORMATIVE; own future+3 FAIL, harm gate FAIL; extended cells 0.80 (0.46-1.32)) could not test a context whose autism interval is this narrow; STATS19 can, and finds nothing at about 1.6x autism's width.

**Scotland rotavirus (pooled; own data 1.55 (0.38-5.31)).** The design is null-biased and insensitive by its own harm-injection gate record (`harm_injection_gate.json`, SCT_RV_counts: null_bias true, weak_recovery true; median 1.54 returned at a true r of 1, recovered log share 0.19 at r = 2; verdict PASS), so its own-data 1.55 is close to what it returns when nothing happens (review I1). Round-1 deaths: school-age bands 0.88 (0.24-3.07) (UNINFORMATIVE), extended cells 0.87 (0.32-2.31), the 3347 count-model fit 0.91 (0.25-3.25). Round 2 (3380): PHS unintentional-injury admissions on the block's council-years 0.92 (0.26-3.05) and STATS19 casualties on the identical cells 0.75 (0.22-2.42) (both prior-dominated, UNINFORMATIVE); PHS on the extended cells 1.11 (0.42-2.75) (CONSISTENT). No control constrains the Scottish estimate.

**Northern Ireland rotavirus (pooled; own data 1.01 (0.88-1.16)).** Round-1 WHO deaths 0.86 (0.25-2.76) and 1.46 (0.51-3.51) were too few to read. Round 2 (3380): PSNI road casualties on age bands 0-9 / 10-16 give 1.02 (0.37-2.04) (CONSISTENT; 1.11x the interval of autism re-aggregated to the same bands, 0.65 (0.29-1.32)), extended cells 1.16 (0.48-2.17) (UNINFORMATIVE), killed or seriously injured 2.36 (0.86-4.79) (future+3 FAIL). No bias detected, but the control's interval is 6.2x as wide as the published autism interval, so it can exclude only large non-specific bias.

**Canada rotavirus excl. Quebec (pooled; own data 0.99 (0.32-1.89)).** No injury series fits these cells (the CIHI tables have no province x age cross-tab). Round 2 (3384) used CCDSS asthma incidence, a same-system non-injury control on the identical 166 cells: 0.71 (0.48-1.06) (CONSISTENT; 0.45x autism's interval, autism on the same cells 0.99 (0.32-1.89); future+3 PASS). 10 of 12 kernel x prior settings keep it CONSISTENT; asthma prevalence (a sensitivity outcome) is 1.08 (0.86-1.39) with a failed future+3 placebo. 3384 states before the result that asthma is the least-bad CCDSS choice, not a condition rotavirus vaccination is known not to affect.

**Quebec rotavirus (pooled; own data 1.07 (0.74-1.48)).** No control exists on the block's SISMACQ cells; the CCDSS Quebec 1-19 analogue (report-only, different cells) gives asthma 0.92 (0.25-3.21) (CONSISTENT by 3384; uniform 3330 class UNINFORMATIVE, prior-dominated).

**Czechia MMR (pooled; own data 1.04 (0.32-3.66)).** Round 2 (3381) fitted NRHZS reimbursed-care injury patients on the closest cells (77 districts x ages 5-9 x 2019-2024; the autism cells are ages 4-6 / 7-10): all injuries 1.19 (0.32-4.48), and all 7 primary controls are prior-dominated (reading UNINFORMATIVE). Nothing held tests this context.

**Brazil maternal dTpa (pooled; own data 0.77 (0.51-1.14)).** Two registers on the identical 108 census cells both return protective 'effects' on injuries: SUS injury admissions 0.51 (0.29-0.83) (BIAS SIGNAL; future+3 FAIL; 6 of 9 components informative, 1 of them (falls 0.38 (0.15-0.86)) bias signals) and, in round 2 (3382), SIM unintentional injury deaths 0.36 (0.15-0.79) (BIAS SIGNAL; future+3 FAIL 0.31 (0.15-0.57)), against autism 0.77 (0.51-1.14); DDD autism / deaths 1.92 (0.97-3.57). Both sources' documented readings: DESIGN BIAS DETECTED (r < 1) (admissions) and DESIGN BIAS DETECTED (r < 1) (deaths). National WHO deaths on the national cohort ramp give 1.08 (0.79-1.40) (CONSISTENT; analogue). The same census design returns injury 'effects' for the report-only products too: admissions for hepatitis A 2.01 (1.15-4.23) and PCV10 + MenC 0.44 (0.28-0.75); deaths for rotavirus 0.37 (0.19-0.86) and PCV10 + MenC 0.26 (0.15-0.54). Reading: the census cross-section design carries non-specific bias, so the maternal dTpa 0.77 is not interpretable as a vaccine effect by itself.

**Colombia varicella (pooled; own data 0.76 (0.51-1.07)).** The school-census disability categories on the same cells are all near 1: hearing 0.98 (0.66-1.49), motor / physical 0.87 (0.67-1.28), visual 0.90 (0.58-1.27) (all CONSISTENT). Round 2 (3385) adds INMLCF non-fatal transport injuries on 500 of the block's 504 cells: 1.36 (0.99-1.93), CONSISTENT by the documented rule but borderline (lower limit 0.991, P(r > 1) 0.973), leaning to harm while autism on the same cells leans protective (0.76 (0.51-1.07); DDD 0.63 (0.39-1.02)). Under the all-cells rule it is 1.39 (1.03-2.02) (BIAS SIGNAL), the transport + accidental composite is 1.33 (1.01-1.96) (BIAS SIGNAL), and the primary is BIAS SIGNAL in 5 of 12 kernel x prior settings. Colombia's autism estimate itself depends on the 2020-2021 lockdown school years: without them (3371) it moves from 0.76 (0.51-1.07) to 1.32 (0.90-1.99), and on those cells the transport control is 1.18 (0.78-1.80) against autism 1.31 (0.89-1.98). Reading: no clean pass; borderline evidence of non-specific harm-direction drift on these cells, and an autism estimate that is not robust to the lockdown years.

**Tanzania schedule (pooled; own data 1.12 (0.54-1.69)).** No injury data exist; albinism, a genetic condition reported through the same school table, gives 1.00 (0.69-1.28) (CONSISTENT), interval 0.55x autism's. It rules out artifacts outside roughly that interval, not an autism lean of 1.12.

## Designs outside the pooled hierarchy

| context | control (source; cells / design) | cells | r (95% interval) | P(r>1) | source verdict | 3330 class | matching autism (basis) | DDD autism / control | future+3 |
|---|---|---|---|---|---|---|---|---|---|
| U.S. NSCH legacy cohorts, PCV 1+ (survey design) | Brain injury / concussion (parent report, ever) (3331; NSCH legacy NIS, PCV 1+) | 120005 | 1.12 (1.02-1.23) | 0.992 | BIAS SIGNAL | BIAS SIGNAL (r > 1 (harm direction)) | 1.10 (0.93-1.32) (same cells) | 0.98 (0.78-1.23) | PASS 1.01 (0.78-1.32) |
| U.S. NSCH modern cohorts, combined 7-series (survey design) | Brain injury / concussion (parent report, ever) (3331; NSCH modern NIS, combined 7-series) | 77105 | 0.85 (0.68-1.06) | 0.074 | CONSISTENT | CONSISTENT | 1.00 (0.84-1.20) (same cells) | 1.18 (0.87-1.59) | FAIL 0.78 (0.64-0.95) |
| U.S. NSCH modern cohorts, PCV 4+ (survey design) | Brain injury / concussion (parent report, ever) (3331; NSCH modern NIS, PCV 4+) | 77105 | 0.81 (0.63-1.04) | 0.051 | CONSISTENT | CONSISTENT | 0.92 (0.75-1.13) (same cells) | 1.14 (0.83-1.56) | FAIL 0.75 (0.60-0.94) |
| U.S. NSCH modern cohorts, rotavirus (survey design) | Brain injury / concussion (parent report, ever) (3331; NSCH modern NIS, rotavirus) | 77105 | 0.90 (0.72-1.12) | 0.179 | CONSISTENT | CONSISTENT | 0.93 (0.79-1.10) (same cells) | 1.03 (0.79-1.34) | PASS 0.98 (0.83-1.17) |
| U.S. SLAITS 2007+2011, PCV7 rollout (survey design) | Brain injury / concussion (parent report, ever) (3331; SLAITS 2007+2011, PCV7 rollout) | 79965 | 1.01 (0.74-1.38) | 0.526 | NOT VALIDATED | CONSISTENT | 0.96 (0.07-14.14) (same cells) | 0.95 (0.80-1.13) | PASS 1.03 (0.82-1.28) |
| U.S. SLAITS 2003/2007 ages 0-5, PCV (survey design) | Injury needing medical attention, past 12 months (3331; SLAITS 2003 ages 0-5 + 2007 ages 2-5, PCV) | 47169 | 1.01 (0.94-1.08) | 0.562 | CONSISTENT | CONSISTENT | 1.11 (0.55-2.26) (same cells) | 1.11 (0.83-1.48) | PASS 1.14 (0.23-5.67) |
| U.S. California CASEMIS jump, PCV7 (age 11) | CASEMIS traumatic brain injury (3331; CA CASEMIS jump, PCV7 (age 11)) | 19 | 1.05 (0.32-3.52) | 0.534 | CONSISTENT | UNINFORMATIVE | 1.03 (0.57-1.85) (same cells) | 0.97 (0.25-3.73) | PASS |
| U.S. California CASEMIS jump, rotavirus 2006 (age 8) | CASEMIS traumatic brain injury (3331; CA CASEMIS jump, rotavirus 2006 (age 8)) | 16 | 1.13 (0.34-3.79) | 0.581 | CONSISTENT | UNINFORMATIVE | 1.22 (0.58-2.55) (same cells) | 1.07 (0.26-4.41) | PASS |
| Brazil rotavirus (census cells) | All Injuries (3332; 27 UF x census bands 0-4/5-9/10-14/15-19, admissions 2021-2023 (census-cohort age)) | 108 | 0.61 (0.35-1.47) | 0.104 | CONSISTENT | CONSISTENT | 0.95 (0.62-1.57) (same cells) | 1.24 (0.74-2.29) | PASS 0.92 (0.52-1.77) |
| Brazil PCV10 + MenC 2010 (census cells) | All Injuries (3332; 27 UF x census bands 0-4/5-9/10-14/15-19, admissions 2021-2023 (census-cohort age)) | 108 | 0.44 (0.28-0.75) | 0.002 | BIAS SIGNAL | BIAS SIGNAL (r < 1 (protective direction)) | 1.19 (0.76-2.05) (same cells) | 2.79 (1.41-6.38) | FAIL 0.48 (0.32-0.76) |
| Brazil PCV10 + MenC 2010 (census cells) | Fractures (3332; 27 UF x census bands 0-4/5-9/10-14/15-19, admissions 2021-2023 (census-cohort age)) | 108 | 0.34 (0.20-0.63) | 0.001 | BIAS SIGNAL | BIAS SIGNAL (r < 1 (protective direction)) | 1.19 (0.76-2.05) (same cells) | 2.83 (1.34-6.86) | FAIL 13.24 (5.88-29.77) |
| Brazil MMRV (census cells) | All Injuries (3332; 27 UF x census bands 0-4/5-9/10-14/15-19, admissions 2021-2023 (census-cohort age)) | 108 | 1.25 (0.80-2.66) | 0.782 | CONSISTENT | CONSISTENT | 0.82 (0.64-1.83) (same cells) | 0.79 (0.56-1.14) | PASS 0.93 (0.72-1.15) |
| Brazil MMRV (census cells) | Head Injury (3332; 27 UF x census bands 0-4/5-9/10-14/15-19, admissions 2021-2023 (census-cohort age)) | 108 | 2.84 (1.02-7.40) | 0.977 | BIAS SIGNAL | BIAS SIGNAL (r > 1 (harm direction)) | 0.82 (0.64-1.83) (same cells) | 0.49 (0.24-0.99) | PASS 1.21 (0.77-3.04) |
| Brazil hepatitis A (census cells) | All Injuries (3332; 27 UF x census bands 0-4/5-9/10-14/15-19, admissions 2021-2023 (census-cohort age)) | 108 | 2.01 (1.15-4.23) | 0.994 | BIAS SIGNAL | BIAS SIGNAL (r > 1 (harm direction)) | 2.17 (1.36-3.98) (same cells) | 1.02 (0.63-1.85) | PASS 1.22 (0.76-2.15) |
| Brazil hepatitis A (census cells) | Poisoning (3332; 27 UF x census bands 0-4/5-9/10-14/15-19, admissions 2021-2023 (census-cohort age)) | 108 | 2.59 (1.06-7.25) | 0.982 | BIAS SIGNAL | BIAS SIGNAL (r > 1 (harm direction)) | 2.17 (1.36-3.98) (same cells) | 0.45 (0.18-1.20) | PASS 2.04 (0.86-5.57) |
| Brazil hepatitis A (census cells) | Road Traffic (3332; 27 UF x census bands 0-4/5-9/10-14/15-19, admissions 2021-2023 (census-cohort age)) | 108 | 0.47 (0.25-0.97) | 0.021 | BIAS SIGNAL | BIAS SIGNAL (r < 1 (protective direction)) | 2.17 (1.36-3.98) (same cells) | 2.92 (1.26-7.92) | PASS 0.61 (0.32-1.29) |
| Norway schedule (primary-care P70-P99) | Injuries ('Skader') (3334; 3872 cells (2017-2024)) | 3872 | 1.25 (0.36-2.21) | 0.685 | CONSISTENT | CONSISTENT | 0.50 (0.18-1.63) (same cells) | 0.38 (0.12-2.59) | PASS 1.17 (0.38-2.17) |
| Estonia hepatitis B (F00-F99) | Injuries (S00-T98) (3334; 272 cells (2000-2016)) | 272 | 1.87 (0.64-3.50) | 0.911 | CONSISTENT | CONSISTENT | 1.08 (0.15-2.99) (same cells) | 0.59 (0.19-1.91) | PASS 1.48 (0.68-3.09) |
| Chile PCV10 (F84 discharges) | All Injuries (3335; national bands 1-9 / 10-19 x 2001-2020 (40 cells)) | 40 | 0.95 (0.54-1.54) | 0.411 | CONSISTENT | CONSISTENT | 0.50 (0.17-1.57) (same cells) | 0.49 (0.16-1.60) | PASS 1.103 (0.573-2.215) |
| Japan JE Hokkaido (MEXT special-support classes) | Hearing + low-vision + physical classes (3340; D1: JE block, prefecture x FY, grade 1 (reference setting)) | 803 | 0.91 (0.62-1.31) | 0.311 | CONSISTENT | CONSISTENT | 1.02 (0.87-1.18) (same cells) | 1.11 (0.76-1.65) | PASS 0.82 (0.53-1.29) |
| Japan MEXT national jumps (reduced forms) | Low-vision + physical classes (3340; D4: national reduced form, JPN_HIB_PCV_2008_2013_MEXT) | 108 | 20 reduced-form rows (paths x trend specs, control controls_lvp): r 0.93-1.08; 20 CONSISTENT, 0 UNINFORMATIVE, 0 up, 0 down |  | CONSISTENT | AGGREGATED (see note) | n/a () | n/a |  |
| Japan MEXT national jumps (reduced forms) | Hearing + low-vision + physical classes (3340; D4: national reduced form, JPN_VARICELLA_ROUTINE_2014_MEXT) | 108 | 10 reduced-form rows (paths x trend specs, control controls3): r 0.90-1.10; 10 CONSISTENT, 0 UNINFORMATIVE, 0 up, 0 down |  | CONSISTENT | AGGREGATED (see note) | n/a () | n/a |  |
| Japan MEXT national jumps (reduced forms) | Hearing + low-vision + physical classes (3340; D4: national reduced form, JPN_HEPB_ROUTINE_2016_MEXT) | 108 | 5 reduced-form rows (paths x trend specs, control controls3): r 0.91-1.06; 4 CONSISTENT, 1 UNINFORMATIVE, 0 up, 0 down |  | CONSISTENT | AGGREGATED (see note) | n/a () | n/a |  |
| Japan MEXT national jumps (reduced forms) | Hearing + low-vision + physical classes (3340; D4: national reduced form, JPN_JE_SUSPENSION_2005_REINSTATEMENT_2010) | 108 | 5 reduced-form rows (paths x trend specs, control controls3): r 0.94-0.96; 4 CONSISTENT, 1 UNINFORMATIVE, 0 up, 0 down |  | CONSISTENT | AGGREGATED (see note) | n/a () | n/a |  |
| Japan MEXT national jumps (reduced forms) |  (3340; D4: national reduced form, JPN_ROTAVIRUS_2020_AND_HPV_SUSPENSION_2013) | 108 | 1 reduced-form rows (paths x trend specs, control ): r nan-nan; 0 CONSISTENT, 0 UNINFORMATIVE, 0 up, 0 down |  | NOT_ESTIMABLE | AGGREGATED (see note) | n/a () | n/a |  |
| Japan MEXT jump-program margins | Hearing + low-vision + physical classes (3340; D5: MARGIN_JPN_JE (JE-only FE, excl. Hokkaido & cohorts<=2012)) | 3243 | 0.83 (0.46-1.51) |  | UNINFORMATIVE | UNINFORMATIVE | 0.96 (0.81-1.15) (same cells) | n/a | A-shift k=3 log r +0.290 (point; RMS k=2..6 0.297) |
| Japan MEXT jump-program margins | Hearing + low-vision + physical classes (3340; D5: MARGIN_JPN_VZV (VZV-only NS (J2 coding: c<=2009 0, 2010-13 dropped))) | 3918 | 1.00 (0.85-1.16) |  | CONSISTENT | CONSISTENT | 1.08 (0.96-1.23) (same cells) | n/a | A-shift k=3 log r +0.010 (point; RMS k=2..6 0.037) |
| Japan MEXT jump-program margins | Low-vision + physical classes (3340; D5: MARGIN_JPN_HIBPCV (HIBPCV-only NS, subsidy scenario S2 (national path shifted 2 coh...) | 4985 | 1.01 (0.92-1.11) |  | CONSISTENT | CONSISTENT | 1.03 (0.97-1.10) (same cells) | n/a | A-shift k=3 log r +0.015 (point; RMS k=2..6 0.032) |
| Japan MEXT jump-program margins | Low-vision + physical classes (3340; D5: MARGIN_JPN_HEPB (pair HIBPCV x HBV, NS, upper, L=0, additive rB)) | 4985 | 0.99 (0.87-1.13) |  | CONSISTENT | CONSISTENT | 0.96 (0.90-1.02) (same cells) | n/a | placebo_rB/rA shifts (see JSON) |
| Japan MEXT pair interactions | Sensory / motor classes (3340; D6: pair interaction thetas) |  | 12 descriptive rows in the source; not classifiable by rule |  | NOT_CLASSIFIABLE | AGGREGATED (see note) | n/a () | n/a |  |
| Korea schedule (NISE special education) | NISE visual + hearing + physical (3344; 17 provinces x grades E4-H3 x 6 years (918 cells)) | 917 | 0.90 (0.56-1.50) | 0.270 | CONSISTENT | CONSISTENT | 0.53 (0.26-0.80) (same cells) | 0.46 (0.20-0.95) | PASS 1.20 (0.85-2.42) |
| Korea schedule (NISE special education) | All External Causes Excl. Self-Harm (3345; matched: national bands 10-14 / 15-19 x 2018-20, 2022-23 (primary)) | 10 | 0.94 (0.25-3.50) | 0.461 | UNINFORMATIVE | UNINFORMATIVE | 0.97 (0.26-3.65) (same cells) | 1.08 (0.28-4.21) | PASS 0.98 (0.28-3.55) |
| Taiwan varicella (MOE special education) | Special-education visual + hearing + limb + CP (3346; 21 counties x {grades 1-9, senior high} x 2008-2025) | 720 | 0.89 (0.73-1.04) | 0.070 | CONSISTENT | CONSISTENT | 1.07 (0.87-1.44) (same cells) | 1.22 (0.91-2.06) | FAIL 0.85 (0.68-0.99) |
| Chile PCV10 (F84 discharges) | Unintentional injuries (3347; identical cells) | 40 | 0.79 (0.40-1.49) | 0.217 | CONSISTENT | CONSISTENT | 0.50 (0.17-1.57) (same cells) | 0.58 (0.18-1.86) | PASS 0.65 (0.34-1.15) |
| U.S. PCV7 2000 national jump (benchmark) | Unintentional injuries (3347; ages 1-4 + 5-9 Lexis window) | 30 | 0.99 (0.66-1.34) | 0.478 | CONSISTENT | CONSISTENT | 1.17 (0.97-1.39) (published (other cells)) | n/a | PASS 1.02 (0.82-1.31) |
| England PCV 2006 national jump (benchmark) | Unintentional injuries (3347; ages 1-4 + 5-9 Lexis window) | 82 | 0.91 (0.64-1.27) | 0.287 | UNINFORMATIVE | UNINFORMATIVE | 0.94 (0.86-1.03) (published (other cells)) | n/a | PASS 0.95 (0.69-1.30) |
| England MenB 2015 national jump (benchmark) | Unintentional injuries (3347; ages 1-4 + 5-9 Lexis window) | 85 | 1.21 (0.78-1.84) | 0.809 | CONSISTENT | CONSISTENT | 0.86 (0.55-1.35) (published (other cells)) | n/a | PASS 0.96 (0.59-1.45) |
| England hexavalent HepB 2017 national jump (benchmark) | Unintentional injuries (3347; ages 1-4 + 5-9 Lexis window) | 75 | 2.21 (1.17-3.94) | 0.992 | BIAS SIGNAL | BIAS SIGNAL (r > 1 (harm direction)) | 1.13 (0.82-1.55) (published (other cells)) | n/a | PASS 1.04 (0.67-1.60) |
| Japan HepB 2016 national jump (benchmark) | Unintentional injuries (3347; ages 1-4 + 5-9 Lexis window) | 72 | 1.04 (0.73-1.48) | 0.595 | CONSISTENT | CONSISTENT | n/a (n/a) | n/a | PASS 1.05 (0.78-1.43) |
| Japan rotavirus 2020 national jump (benchmark) | Unintentional injuries (3347; ages 1-4 + 5-9 Lexis window) | 55 | 0.87 (0.48-1.53) | 0.309 | CONSISTENT | CONSISTENT | n/a (n/a) | n/a | PASS 1.15 (0.72-1.85) |
| Denmark PCV 2007 national jump (benchmark) | Unintentional injuries (3347; ages 1-4 + 5-9 Lexis window) | 82 | not fitted: 12 of 82 cells with zero deaths |  | UNINFORMATIVE | NOT FITTED | n/a (n/a) | n/a | NOT TESTABLE (12 of 82 cells with zero deaths (Gaussian contract refuses zeros)) n/a |
| New Zealand PCV 2008 national jump (benchmark) | Unintentional injuries (3347; ages 1-4 + 5-9 Lexis window) | 80 | not fitted: 2 of 80 cells with zero deaths |  | UNINFORMATIVE | NOT FITTED | n/a (n/a) | n/a | NOT TESTABLE (2 of 80 cells with zero deaths (Gaussian contract refuses zeros)) n/a |
| Germany PCV rebuilt panel (2024 cells) | All Injuries (S00-T98 Excl. T78, T80-T88) (3348; design A: merged Laender x bands 0-4/5-9/10-14, 2024 only (33 cells)) | 33 | 0.74 (0.23-3.11) | 0.331 | UNINFORMATIVE | UNINFORMATIVE | 0.87 (0.22-3.50) (same cells) | 0.91 (0.24-3.62) | PASS 0.874 (0.235-3.518) |
| Germany PCV rebuilt 16-Land panel (2024 cells) | Shoulder or Upper-Arm Fracture (S42) (3348; design B: 16 Laender, +0.5 continuity (sensitivity) x bands 0-4/5-9/10-14, 2024 only (4...) | 48 | 0.22 (0.08-0.65) | 0.008 | BIAS SIGNAL | BIAS SIGNAL (r < 1 (protective direction)) | 0.83 (0.21-3.36) (same cells) | n/a | PASS 0.340 (0.113-1.712) |
| England MMR, UTLA design (injury admissions ages 0-4) | Injury admissions aged 0-4 (Fingertips 90832) (3380; C6 UTLA Admissions 0-4, cell set S (2016-2022; ages 0-4)) | 1020 | 1.29 (0.38-4.49) | 0.657 | UNINFORMATIVE | UNINFORMATIVE | 0.53 (0.36-0.96) (published (other cells)) | 0.42 (0.11-1.57) | PASS 0.95 (0.31-3.49) |
| England MMR, UTLA design (injury admissions ages 0-4) | Injury admissions aged 0-4 (Fingertips 90832) (3380; C6 UTLA Admissions 0-4, cell set E (2016-2022; ages 0-4)) | 1020 | not fitted: E collapses to S: the m24 labels 2008-2010 filled from Fingertips 30309 cannot be used because MMR1@5y and DTaP@24m |  | SEE S (collapsed) | NOT FITTED | 0.53 (0.36-0.96) (published (other cells)) | n/a | not run (outcome not fitted) |
| Germany PCV rebuilt panel (2024 cells) | All Injuries (S00-T98 Excl. T78, T80-T88) (3381; DEU_B (report): DEU_PCV_rebuilt, 825 cells) | 825 | 1.12 (0.94-1.32) | 0.898 | CONSISTENT | CONSISTENT | 0.49 (0.21-1.08) (same cells) | 0.47 (0.21-1.03) | PASS 1.07 (0.91-1.25) |
| Germany PCV rebuilt panel (2024 cells) | Other External Causes Incl. Drowning (T75) (3381; DEU_B (report): DEU_PCV_rebuilt, 825 cells) | 825 | 0.63 (0.37-0.99) | 0.022 | BIAS SIGNAL | BIAS SIGNAL (r < 1 (protective direction)) | 0.49 (0.21-1.08) (same cells) | n/a | PASS 1.26 (0.77-1.90) |
| Brazil rotavirus (census cells) | All Unintentional Injuries (V01-X59) (3382; 27 UF x census bands 0-4/5-9/10-14/15-19, SIM deaths 2021-2023 (age at death)) | 108 | 0.37 (0.19-0.86) | 0.013 | BIAS SIGNAL | BIAS SIGNAL (r < 1 (protective direction)) | 0.95 (0.62-1.57) (same cells) | 1.91 (0.86-4.75) | FAIL 0.26 (0.13-0.54) |
| Brazil PCV10 + MenC 2010 (census cells) | All Unintentional Injuries (V01-X59) (3382; 27 UF x census bands 0-4/5-9/10-14/15-19, SIM deaths 2021-2023 (age at death)) | 108 | 0.26 (0.15-0.54) | 0.001 | BIAS SIGNAL | BIAS SIGNAL (r < 1 (protective direction)) | 1.19 (0.76-2.05) (same cells) | 3.04 (1.29-8.00) | FAIL 0.28 (0.18-0.46) |
| Brazil PCV10 + MenC 2010 (census cells) | Drowning (W65-W74) (3382; 27 UF x census bands 0-4/5-9/10-14/15-19, SIM deaths 2021-2023 (age at death)) | 108 | 0.18 (0.10-0.43) | 0.001 | BIAS SIGNAL | BIAS SIGNAL (r < 1 (protective direction)) | 1.19 (0.76-2.05) (same cells) | 2.30 (0.91-6.42) | PASS 0.42 (0.22-1.11) |
| Brazil MMRV (census cells) | All Unintentional Injuries (V01-X59) (3382; 27 UF x census bands 0-4/5-9/10-14/15-19, SIM deaths 2021-2023 (age at death)) | 108 | 1.22 (0.61-1.96) | 0.759 | CONSISTENT | CONSISTENT | 0.82 (0.64-1.83) (same cells) | 0.66 (0.37-2.26) | PASS 0.71 (0.37-1.16) |
| Brazil MMRV (census cells) | Road Traffic (V01-V89) (3382; 27 UF x census bands 0-4/5-9/10-14/15-19, SIM deaths 2021-2023 (age at death)) | 107 | 2.32 (1.47-3.46) | 0.999 | UNINFORMATIVE | BIAS SIGNAL (r > 1 (harm direction)) | 0.82 (0.63-1.84) (same cells) | 0.36 (0.18-0.67) | PASS 1.18 (0.78-1.91) |
| Brazil hepatitis A (census cells) | All Unintentional Injuries (V01-X59) (3382; 27 UF x census bands 0-4/5-9/10-14/15-19, SIM deaths 2021-2023 (age at death)) | 108 | 0.75 (0.38-1.74) | 0.231 | CONSISTENT | CONSISTENT | 2.17 (1.36-3.98) (same cells) | 2.24 (1.00-6.01) | FAIL 0.47 (0.26-0.89) |
| Brazil hepatitis A (census cells) | Drowning (W65-W74) (3382; 27 UF x census bands 0-4/5-9/10-14/15-19, SIM deaths 2021-2023 (age at death)) | 108 | 0.34 (0.14-0.81) | 0.009 | BIAS SIGNAL | BIAS SIGNAL (r < 1 (protective direction)) | 2.17 (1.36-3.98) (same cells) | 3.03 (1.23-8.39) | FAIL 0.30 (0.14-0.67) |
| Taiwan varicella (MOE special education) | Accident Deaths (事故傷害) (3385; county x {grades 1-9, senior high} x school year, observed cells) | 628 | 0.91 (0.48-2.05) | 0.388 | UNINFORMATIVE | UNINFORMATIVE | 1.07 (0.86-1.46) (same cells) | 1.08 (0.56-2.65) | PASS 0.73 (0.38-1.51) |

**Chile PCV10 (report-only).** Identical-cell DEIS injury discharges 0.95 (0.54-1.54) (CONSISTENT) and WHO injury deaths 0.79 (0.40-1.49) (CONSISTENT) against F84 0.50 (0.17-1.57); 0 of 16 fitted discharge outcomes are bias signals. No bias detected, which does not rescue a block that fails plasmode.

**Korea schedule (report-only).** NISE sensory + physical 0.90 (0.56-1.50) (CONSISTENT) against autism 0.53 (0.26-0.80); but intellectual disability (0.80 (0.53-1.00), BIAS SIGNAL) and all non-autism categories (0.81 (0.56-0.96), BIAS SIGNAL) share autism's protective lean and fail future+3, so the Korean autism estimate looks like shared neurodevelopmental-label dynamics. WHO deaths: matched cells 0.94 (0.25-3.50) (UNINFORMATIVE); the ages 1-19 analogue 0.98 (0.44-2.69) (CONSISTENT).

**Taiwan varicella (report-only).** Sensory + physical 0.89 (0.73-1.04) (CONSISTENT) against autism 1.07 (0.87-1.44); the future+3 placebo fails downward for the composite (0.85 (0.68-0.99)) and for physical (0.77 (0.57-0.94)), while autism's fails upward, so the null at the actual cutoff does not certify the design. Round 2 (3385): MOHW accident deaths on the same cells 0.91 (0.48-2.05) (UNINFORMATIVE; 2.7x autism's interval).

**Japan JE Hokkaido and MEXT jumps (report-only / excluded).** D1 hearing + low-vision + physical classes 0.91 (0.62-1.31) (CONSISTENT; interval 2.48x autism's 1.02 (0.87-1.18)); SDID 1.06 (0.22-1.87) (UNINFORMATIVE); national reduced forms: JPN_HIB_PCV_2008_2013_MEXT CONSISTENT; JPN_VARICELLA_ROUTINE_2014_MEXT CONSISTENT; JPN_HEPB_ROUTINE_2016_MEXT CONSISTENT; JPN_JE_SUSPENSION_2005_REINSTATEMENT_2010 CONSISTENT; JPN_ROTAVIRUS_2020_AND_HPV_SUSPENSION_2013 NOT_ESTIMABLE.

**Norway schedule and Estonia hepatitis B ('other'-class, report-only).** Primary-care injuries 1.25 (0.36-2.21) against P70-P99 0.50 (0.18-1.63); Estonian injury initial cases 1.87 (0.64-3.50) against F00-F99 1.08 (0.15-2.99); both CONSISTENT. 3334's post hoc Tallinn analysis shows the Estonian injury lean rests on one area's 2004 reporting step.

**Germany PCV rebuilt and England MMR UTLA (report-only, round 2).** The rebuilt German panel on 825 cells: all injuries 1.12 (0.94-1.32) (CONSISTENT) against F84 0.49 (0.21-1.08) on the same cells. England UTLA injury admissions at ages 0-4 on the MMR UTLA design: 1.29 (0.38-4.49) (UNINFORMATIVE).

**U.S. survey designs (outside the hierarchy).** The one survey design where autism leans harmful, NSCH legacy PCV, shows the same lean on brain injury (1.12 (1.02-1.23) vs autism 1.10 (0.93-1.32)); the SLAITS 2007+2011 design is NOT VALIDATED (its autism refit did not reproduce).

**Benchmarks (excluded national jumps).** 6 fitted; England hexavalent HepB 2017 returns injury deaths 2.21 (1.17-3.94) (BIAS SIGNAL, harm direction), a reminder that single-country national steps pick up period shocks (its exposed cells include 2020-2021).

## Empirical-calibration preview (not a replacement for the hierarchy)

Method, fixed in the rule before anything below was computed: each negative-control estimate enters as m = ln(median r) with s = (ln high - ln low) / 3.92; the systematic-error distribution is m_i ~ N(mu, tau^2 + s_i^2) fitted by maximum likelihood (95% profile-likelihood intervals); an autism estimate b (SD s_b from its interval) is calibrated to exp(b - mu) with SD sqrt(s_b^2 + tau^2 + se(mu)^2). This is the negative-control-only calibration of Schuemie et al., which assumes the systematic error is exchangeable across contexts and the same at every true effect. S7, S7b and S8 were specified in the round-2 amendment before they were computed.

| set | n | exp(mu) (95% profile) | tau (95% profile) | se(mu) | mean / SD of log r (unweighted) | members |
|---|---|---|---|---|---|---|
| C_INJ | 5 | 0.96 (0.51-1.87) | 0.48 (0.09-1.26) | 0.27 | -0.00 / 0.67 | US_PCV / 3331 tbi 1.65 (0.77-4.29); US_RV2006 / 3331 tbi 2.37 (1.09-4.58); BRA_MDTPA / 3332 all_injury 0.51 (0.29-0.83); ENG_RV / 3333 all_unintentional 0.57 (0.25-1.22); JPN_MMR / 3341 inj_unintentional 0.86 (0.46-1.61) |
| S1_no_us_rv2006 | 4 | 0.74 (0.45-1.33) | 0.21 (0.00-0.99) | 0.20 | -0.22 / 0.53 | C_INJ without U.S. RV2006 |
| S2_uk_cell_set_E | 7 | 1.02 (0.67-1.63) | 0.37 (0.00-0.90) | 0.20 | 0.08 / 0.52 | US_PCV / 3331 tbi 1.65 (0.77-4.29); US_RV2006 / 3331 tbi 2.37 (1.09-4.58); BRA_MDTPA / 3332 all_injury 0.51 (0.29-0.83); JPN_MMR / 3341 inj_unintentional 0.86 (0.46-1.61); ENG_RV / 3333 all_unintentional 0.80 (0.46-1.32); SCT_RV / 3333 all_unintentional 0.87 (0.32-2.31); NIR_RV / 3333 all_unintentional 1.46 (0.51-3.51) |
| S3_with_noninjury | 9 | 0.99 (0.75-1.33) | 0.23 (0.00-0.64) | 0.12 | 0.04 / 0.49 | US_PCV / 3331 tbi 1.65 (0.77-4.29); US_RV2006 / 3331 tbi 2.37 (1.09-4.58); BRA_MDTPA / 3332 all_injury 0.51 (0.29-0.83); ENG_RV / 3333 all_unintentional 0.57 (0.25-1.22); JPN_MMR / 3341 inj_unintentional 0.86 (0.46-1.61); US_RotaShield / 3326 hearing 1.41 (0.48-3.16); HKG_PCV / 3342 physical_sensory 1.15 (0.90-1.47); COL_VARICELLA / 3338 motor_physical 0.87 (0.67-1.28); TZA_SCHEDULE / 3339 albinism 1.00 (0.69-1.28) |
| S4_all_designs | 27 | 0.98 (0.88-1.10) | 0.15 (0.00-0.35) | 0.05 | 0.01 / 0.42 | 27 members (see calibration_preview.json) |
| S4b_all_designs_same_scale | 22 | 1.01 (0.84-1.21) | 0.29 (0.08-0.51) | 0.09 | 0.03 / 0.47 | 22 members (see calibration_preview.json) |
| S5_depriored | 5 | 1.00 (0.42-2.49) | 0.71 (0.26-1.76) | 0.37 | 0.04 / 0.94 | C_INJ members with the N(0, ln 2) prior removed |
| S7_add_round2_informative_injury | 12 | 0.93 (0.72-1.19) | 0.29 (0.00-0.62) | 0.11 | -0.09 / 0.52 | C_INJ plus ENG_RV / 3380 primary (C1 HES Diagnosis, cell set S) 0.99 (0.57-1.57); ENG_RV / 3380 primary (C3 STATS19 England, cell set S) 1.00 (0.87-1.16); NIR_RV / 3380 primary (C7 PSNI Casualties, cell set S) 1.02 (0.37-2.04); SWE_PCV / 3381 INJ_ALL (SWE_A) 1.06 (0.98-1.13); BRA_MDTPA / 3382 unintentional (27 UF x census bands 0-4/5-9/10-14/15-19, SIM deaths 2021-2023) 0.36 (0.15-0.79); COL_VARICELLA / 3385 transport (region x level x school year 2019-2024, observed cells) 1.36 (0.99-1.93); HKG_PCV / 3385 road (WHO bands 5-9/10-14/15-19 x calendar year) 0.66 (0.48-0.92) |
| S7b_add_round2_all_eligible_injury | 14 | 0.98 (0.76-1.26) | 0.32 (0.06-0.65) | 0.12 | -0.01 / 0.55 | S7 plus US_PCV / 3383 M1_unintentional (M1: Unintentional injury) 2.59 (1.15-5.76); JPN_MMR / 3385 accidents (nine Kohoku birth cohorts 1988-1996, Yokohama deaths at ages 15-19) 1.00 (0.41-2.21) |
| S8_drop_failed_own_checks | 3 | 1.44 (0.72-3.07) | 0.28 (0.00-1.27) | 0.27 | 0.41 / 0.52 | US_PCV / 3331 tbi 1.65 (0.77-4.29); US_RV2006 / 3331 tbi 2.37 (1.09-4.58); JPN_MMR / 3341 inj_unintentional 0.86 (0.46-1.61) |

S4 follows the rule literally and so mixes the NSCH/SLAITS prevalence ratios per +10 pp (a different scale with small Wald SEs) with individual-scale r; S4b, added after the rule, keeps only the individual-scale rows. C_INJ contains two members from one U.S. fit (PCV and RV2006 TBI), which are not independent; so does S8.

Beside the hierarchy: the pooled model gives each context an assumed bias SD of 0.35 or 0.70 on the log scale (3320 'Bias SD') plus a shared bias SD of 0.15. The injury controls' tau is 0.48 (95% profile 0.09-1.26) under C_INJ and 0.29 (0.00-0.62) under S7; the C_INJ interval contains 0.35 and 0.70, the S7 interval 0.35, so the controls can neither confirm nor reject the assumed context values. The systematic-error mean's standard error, se(mu), is 0.27 under C_INJ and 0.11 under S7 against the shared-bias SD of 0.15 the hierarchy assumes (review I3).

Calibrated own-data autism estimates (C_INJ null; sensitivities beside; the calibrated interval is exp(b - mu) +/- 1.96 SD):

| context | status | own-data autism | calibrated (C_INJ) | P(r>1) | width x | S6 leave own out | S7 (round 2 added) | S7 P(r>1) | S7 width x | S7 leave own out | S7b | S8 | S1 | S3 | S4b | S5 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| U.S. PCV (Core) | hierarchy | 1.17 (0.97-1.39) | 1.21 (0.40-3.63) | 0.63 | 6.1 | 1.35 (0.44-4.21) | 1.25 (0.67-2.35) | 0.76 | 3.5 | 1.29 (0.68-2.42) | 1.19 (0.59-2.39) | 0.81 (0.37-1.76) | 1.58 (0.86-2.89) | 1.18 (0.69-2.01) | 1.16 (0.63-2.14) | 1.17 (0.24-5.65) |
| U.S. RotaShield (Core) | hierarchy | 0.87 (0.36-1.83) | 0.91 (0.23-3.51) | 0.44 | 1.7 | - | 0.94 (0.34-2.58) | 0.45 | 1.2 | - | 0.89 (0.31-2.56) | 0.60 (0.20-1.84) | 1.18 (0.44-3.20) | 0.88 (0.34-2.29) | 0.87 (0.32-2.36) | 0.87 (0.15-5.10) |
| U.S. Rotavirus 2006 (Core) | hierarchy | 0.95 (0.76-1.19) | 0.99 (0.33-2.99) | 0.49 | 4.9 | 1.29 (0.69-2.40) | 1.02 (0.54-1.95) | 0.53 | 2.8 | 1.07 (0.62-1.84) | 0.97 (0.48-1.98) | 0.66 (0.30-1.46) | 1.29 (0.69-2.40) | 0.96 (0.55-1.67) | 0.95 (0.50-1.78) | 0.95 (0.20-4.64) |
| Japan MMR (Core) | hierarchy | 0.85 (0.28-2.32) | 0.88 (0.19-4.02) | 0.44 | 1.4 | 0.85 (0.16-4.54) | 0.91 (0.27-3.09) | 0.44 | 1.2 | 0.91 (0.26-3.17) | 0.87 (0.25-3.04) | 0.59 (0.16-2.17) | 1.15 (0.34-3.84) | 0.86 (0.27-2.77) | 0.84 (0.25-2.83) | 0.85 (0.13-5.64) |
| Sweden PCV | hierarchy | 1.01 (0.79-1.56) | 1.05 (0.34-3.27) | 0.53 | 3.3 | - | 1.09 (0.54-2.17) | 0.59 | 2.0 | 1.11 (0.51-2.43) | 1.03 (0.49-2.19) | 0.70 (0.30-1.61) | 1.37 (0.70-2.67) | 1.02 (0.56-1.88) | 1.00 (0.51-1.98) | 1.01 (0.20-5.03) |
| Germany PCV | hierarchy | 1.03 (0.27-3.96) | 1.07 (0.19-6.01) | 0.53 | 1.3 | - | 1.11 (0.25-4.83) | 0.55 | 1.1 | - | 1.05 (0.23-4.73) | 0.71 (0.15-3.34) | 1.40 (0.32-6.02) | 1.04 (0.25-4.37) | 1.02 (0.24-4.44) | 1.03 (0.13-8.12) |
| Hong Kong PCV | hierarchy | 0.89 (0.66-1.16) | 0.93 (0.30-2.84) | 0.45 | 3.9 | - | 0.96 (0.49-1.87) | 0.45 | 2.3 | 0.86 (0.64-1.15) | 0.91 (0.44-1.89) | 0.62 (0.27-1.39) | 1.21 (0.64-2.30) | 0.90 (0.51-1.61) | 0.89 (0.46-1.71) | 0.89 (0.18-4.39) |
| England Rotavirus | hierarchy | 0.94 (0.86-1.02) | 0.98 (0.33-2.91) | 0.49 | 13.0 | 0.87 (0.27-2.80) | 1.01 (0.55-1.87) | 0.52 | 7.3 | 1.00 (0.44-2.27) | 0.96 (0.49-1.90) | 0.65 (0.30-1.40) | 1.28 (0.71-2.29) | 0.95 (0.57-1.59) | 0.94 (0.52-1.70) | 0.95 (0.20-4.54) |
| Scotland Rotavirus | hierarchy | 1.55 (0.38-5.31) | 1.61 (0.29-8.88) | 0.71 | 1.3 | - | 1.67 (0.39-7.11) | 0.75 | 1.1 | - | 1.58 (0.36-6.96) | 1.07 (0.23-4.92) | 2.10 (0.50-8.86) | 1.57 (0.38-6.43) | 1.54 (0.36-6.53) | 1.55 (0.20-12.04) |
| Northern Ireland Rotavirus | hierarchy | 1.01 (0.88-1.16) | 1.05 (0.35-3.14) | 0.54 | 7.9 | - | 1.09 (0.59-2.02) | 0.61 | 4.5 | 1.09 (0.57-2.12) | 1.03 (0.52-2.05) | 0.70 (0.32-1.52) | 1.37 (0.76-2.48) | 1.02 (0.61-1.73) | 1.01 (0.55-1.84) | 1.01 (0.21-4.89) |
| Canada Rotavirus (Excl. Quebec) | hierarchy | 0.99 (0.32-1.89) | 1.03 (0.25-4.18) | 0.52 | 1.6 | - | 1.06 (0.36-3.11) | 0.55 | 1.2 | - | 1.01 (0.33-3.08) | 0.69 (0.21-2.21) | 1.34 (0.47-3.87) | 1.00 (0.36-2.78) | 0.98 (0.34-2.86) | 0.99 (0.16-6.01) |
| Quebec Rotavirus | hierarchy | 1.07 (0.74-1.48) | 1.11 (0.36-3.47) | 0.57 | 3.3 | - | 1.15 (0.57-2.31) | 0.65 | 2.0 | - | 1.09 (0.51-2.33) | 0.74 (0.32-1.71) | 1.45 (0.74-2.84) | 1.08 (0.59-2.00) | 1.06 (0.54-2.11) | 1.07 (0.22-5.33) |
| Czechia MMR | hierarchy | 1.04 (0.32-3.66) | 1.08 (0.21-5.52) | 0.54 | 1.3 | - | 1.12 (0.29-4.36) | 0.56 | 1.1 | - | 1.06 (0.26-4.27) | 0.72 (0.17-3.03) | 1.41 (0.37-5.43) | 1.05 (0.28-3.93) | 1.03 (0.27-4.00) | 1.04 (0.14-7.58) |
| Brazil Maternal DTaP | hierarchy | 0.77 (0.51-1.14) | 0.80 (0.25-2.53) | 0.35 | 2.9 | 0.65 (0.24-1.82) | 0.82 (0.40-1.70) | 0.30 | 1.8 | 0.74 (0.49-1.11) | 0.78 (0.36-1.71) | 0.53 (0.23-1.25) | 1.04 (0.51-2.10) | 0.77 (0.41-1.48) | 0.76 (0.37-1.55) | 0.77 (0.15-3.87) |
| Colombia Varicella | hierarchy | 0.76 (0.51-1.07) | 0.79 (0.25-2.50) | 0.35 | 3.1 | - | 0.82 (0.40-1.67) | 0.29 | 1.9 | 0.86 (0.44-1.70) | 0.78 (0.36-1.69) | 0.53 (0.23-1.23) | 1.04 (0.52-2.06) | 0.77 (0.41-1.45) | 0.76 (0.38-1.53) | 0.77 (0.15-3.83) |
| Tanzania Schedule | hierarchy | 1.12 (0.54-1.69) | 1.16 (0.34-3.95) | 0.60 | 2.2 | - | 1.20 (0.52-2.75) | 0.67 | 1.5 | - | 1.14 (0.47-2.75) | 0.77 (0.30-2.00) | 1.51 (0.67-3.40) | 1.13 (0.53-2.41) | 1.11 (0.49-2.52) | 1.12 (0.21-5.93) |
| Chile PCV10 (F84 discharges) | report-only (reference) | 0.50 (0.17-1.57) | 0.52 (0.11-2.47) | 0.20 | 1.4 | - | 0.54 (0.15-1.92) | 0.17 | 1.1 | - | 0.51 (0.14-1.89) | 0.35 (0.09-1.34) | 0.68 (0.19-2.39) | 0.50 (0.15-1.73) | 0.50 (0.14-1.76) | 0.50 (0.07-3.43) |
| Korea schedule (NISE special education) | report-only (reference) | 0.53 (0.26-0.80) | 0.55 (0.16-1.88) | 0.17 | 2.2 | - | 0.57 (0.25-1.31) | 0.09 | 1.5 | - | 0.54 (0.23-1.31) | 0.37 (0.14-0.95) | 0.72 (0.32-1.61) | 0.54 (0.25-1.15) | 0.53 (0.23-1.19) | 0.53 (0.10-2.82) |
| Taiwan varicella (MOE special education) | report-only (reference) | 1.07 (0.87-1.44) | 1.11 (0.37-3.39) | 0.58 | 4.4 | - | 1.15 (0.60-2.22) | 0.66 | 2.6 | - | 1.09 (0.53-2.25) | 0.74 (0.33-1.65) | 1.45 (0.77-2.73) | 1.08 (0.62-1.91) | 1.06 (0.56-2.02) | 1.07 (0.22-5.25) |
| Japan JE Hokkaido (MEXT special-support classes) | report-only (reference) | 1.02 (0.87-1.18) | 1.06 (0.35-3.16) | 0.54 | 7.2 | - | 1.09 (0.59-2.04) | 0.61 | 4.1 | - | 1.04 (0.52-2.07) | 0.71 (0.33-1.53) | 1.38 (0.76-2.50) | 1.03 (0.61-1.74) | 1.01 (0.55-1.86) | 1.02 (0.21-4.92) |

Reading of the preview. The systematic-error mean exp(mu) ranges 0.74-1.44 across the sets (C_INJ 0.96, S7 0.93), so calibration moves the autism medians little; what it changes is width. Adding the round-2 controls more than doubles the set (n = 12) and narrows both the tau interval and se(mu); tau itself is 0.48 under C_INJ and 0.29 under S7. Under S7 the intervals that widen most are England Rotavirus (7.3x), Northern Ireland Rotavirus (4.5x), U.S. PCV (Core) (3.5x), U.S. Rotavirus 2006 (Core) (2.8x); calibrated intervals span 0.25 to 7.11 (C_INJ: 0.19 to 8.88). The sets have 3-27 members, several of them national or different-cell analogues and two from one U.S. fit; S8, which keeps only members that pass their own checks, rests on 3 (two of them one fit). The preview sizes the problem; it does not replace the hierarchy.

## Gaps and the download request

140 discovery candidates were screened for round 1; 65 were held and 18 infeasible. The approved request `qa/negative_controls_20260925/DOWNLOAD_REQUEST.md` (53 entries: 11 HIGH, 12 MEDIUM, 30 LOW) is left as approved (not regenerated); round 2 (3380-3385) fitted what was downloaded from it. Contexts whose reading rows are still all uninformative or missing: U.S. RotaShield (Core), Quebec Rotavirus, Czechia MMR.

## Caveats

- Each source used its own documented rule; the classes are not identical across sources (3326 uses SIMILAR / NULL-CENTRED / BELOW with a 2x width limit, most others use CONSISTENT / BIAS SIGNAL / UNINFORMATIVE with a 2.5x limit, a prior-dominance flag and, in 3380-3385 and 3341, a thinness criterion). The uniform 3330 class exists only to compare rows; it has no thinness criterion, so it can read a thin row as CONSISTENT or BIAS SIGNAL where the source says UNINFORMATIVE (Sweden S32 / S42; Japan Yokohama).
- Several rules were written with some estimates known (notably 3342, whose deciding estimate was known, and 3326 / 3331 for TBI). 3330 and its round-2 amendment were written after every per-row estimate was known; only the calibration sets (and their expected members) were fixed before they were computed.
- Many controls are not on the autism block's cells: WHO deaths are national and band-aggregated (analogues for Sweden, Germany and Brazil), the U.S. death cells and Hong Kong road casualties are different cells on the same exposure path, survey designs are on a prevalence-ratio-per-10-pp scale, and several controls are administrative categories (IDEA TBI) or non-injury conditions (disability categories; Canadian asthma).
- An informative control that is CONSISTENT does not validate an autism estimate: it only rules out non-specific bias larger than its own interval. Several controls also fail their own future+3 placebo (England WHO school-age deaths, Brazil admissions and deaths, the U.S. death cells, Taiwan sensory + physical), which is itself evidence of timing or trend confounding in those cells.
- The any-row tally counts a context as showing a bias signal when any one of its rows does; with many categories (England HES, Swedish fracture codes) some intervals exclude 1 by chance, which is why the reading-row tally is shown beside it.
- Injury controls measure flows (admissions, deaths, casualties) where autism is often a prevalence stock; kernels absorb level differences, but a control can still differ from autism in its sensitivity to trends.
- The calibration preview treats posterior medians and CrIs under a common N(0, ln 2) prior as approximate Gaussian estimates; prior-dominated controls are excluded and the de-priored sensitivity (S5) shows how much the prior shrinkage matters. Members are not independent where they share a fit (U.S. PCV and RV2006 TBI).

Inputs (sha256 at run time) are recorded in `project_results/negative_controls_3330/3330/calibration_preview.json`.