# 3385 - Injury negative controls, round 2: Colombia varicella, Hong Kong PCV, Japan MMR (Kohoku) and Taiwan varicella designs on injury outcomes from the approved downloads

**Documented readings (class of each context's primary injury control under its primary cell rule):**

- **Colombia varicella** (COL_VARICELLA_school_autism, pooled): Transport Injuries (INMLCF, Non-Fatal) r 1.36 (0.99-1.93) -> CONSISTENT: no non-specific bias detected at this precision (this does not validate the autism estimate). Autism on its own cells 0.76 (0.51-1.07); autism on the control's cells 0.76 (0.51-1.07).
- **Hong Kong PCV** (HKG_PCV_autism, pooled): Road-Traffic Casualties (Police-Reported) r 0.66 (0.48-0.92) -> BIAS SIGNAL: the design returns an 'effect' on injuries (r < 1 (protective direction)). Autism on its own cells 0.89 (0.66-1.16).
- **Japan MMR (Kohoku)** (JPN_MMR (core v2, Kohoku), pooled (core v2)): Accidents (不慮の事故) r 1.00 (0.41-2.21) -> UNINFORMATIVE -> NOT TESTABLE: no conclusion about bias in this design. Autism on its own cells 0.85 (0.28-2.32).
- **Taiwan varicella** (TWN_VARICELLA_autism, report-only): Accident Deaths (事故傷害) r 0.91 (0.48-2.05) -> UNINFORMATIVE -> NOT TESTABLE: no conclusion about bias in this design. Autism on its own cells 1.07 (0.87-1.44); autism on the control's cells 1.07 (0.86-1.46).

Generated by `07_analysis_code/3385_injury_negative_controls_col_hkg_jpn_twn_round2.py` (every number below is computed by that script and also written to `3385_injury_negative_controls_col_hkg_jpn_twn_round2.csv` / `3385_injury_negative_controls_col_hkg_jpn_twn_round2.json`). Figure: `08_figures/3385_injury_negative_controls_col_hkg_jpn_twn_round2.png` (hollow light markers = UNINFORMATIVE). Fits: `project_results/negative_controls_3330/3385/`. Tests: `07_analysis_code/test_3385_injury_negative_controls_col_hkg_jpn_twn_round2.py`. Status: falsification audit; ecological; the controls are outcome_class 'other' and never enter the autism hierarchy, a gate decision or a pooled estimate.

## Interpretation rule (documented)

Rule file `qa/negative_controls_20260925/3385_rule.json`, SHA-256 `8b49cbb4f7c5582351219ce71ecdc088ad37c96d6afacd85bdcbd162ab62085d` (recorded at writing in `qa/negative_controls_20260925/3385_rule.sha256`); its 'written' field: 2026-09-29 22:30:56 Central Daylight Time (2026-09-30T03:30:56Z), before any 3385 fit: no likelihood, posterior, placebo, DDD or contrast of any outcome (injury or autism) had been computed by this task. project_results/negative_controls_3330/3385/ held only the counts written by the script's --build step (no likelihood evaluated): build.json, col_cells.csv, hkg_cells.csv, jpn_cells.csv, twn_cells.csv. First fit started 2026-09-29T22:31:34-05:00. All 106 fit records (il.fit receipts, Japan importance-sampling receipts and result files) carry this hash, recorded at fit time; fitting and reporting refuse if the file changed or a record carries another hash.

Fit provenance: 99 likelihood fits ran 2026-09-29T22:31:34-05:00 to 2026-09-29T22:40:02-05:00 (logs: `project_results/negative_controls_3330/3385/logs/fit_20260929_223103.log`). Script hash(es) recorded at fit time: `acf5b134d954f2aa...`; current script `918c93e96b3a61d9...` (the script was edited after the fits in reporting and descriptive-accounting code; a fit is reused only while its block input hash, which covers every cell, outcome value and exposure the likelihood reads, and the imported model-code hashes are unchanged). Imported model code (processed/ modules and the round-1 scripts used) equal to the fit-time hashes: True.

> UNINFORMATIVE: the control is not fittable under the rule (fewer than 20 cells with count > 0, or no exposure variation), or ln(high / low) of its 95% CrI > 2.5 x ln(high / low) of the matching autism 95% CrI (autism refitted on the SAME cells for COL and TWN; autism on its own cells for HKG, flagged 'reference on different cells'; for JPN 3341's bound 2.5 x 2.1177), or (JPN only) 3341's thin / ESS criteria.
>
> CONSISTENT: otherwise, if the 95% CrI includes 1 (low <= 1 <= high).
>
> BIAS SIGNAL: otherwise (the 95% CrI excludes 1); direction reported: 'r > 1 (harm direction)' if low > 1, 'r < 1 (protective direction)' if high < 1.
>
> Japan (3341's rule, adopted unchanged): UNINFORMATIVE if (a) not fittable (a cohort count of zero); (b) thin: some cohort cell has fewer than 16 deaths (Poisson SD of the cell's log rate > 0.25, the working likelihood's residual SD); (c) CrI log width > 2.5 x 2.1177 (autism's); (d) pooled ESS < 100 at p18 and again at p20 (3341_rule.json, adopted unchanged).

Expected r = 1 for every injury control. Reported beside the class and never changing it: the future+3 timing placebo (D4: PASS iff the 95% CrI includes 1 under all four kernel settings), the DDD / contrast, the sensitivity cell rules and secondary outcomes, the class under the other kernel settings and priors, prior-domination and the grid resolution (Kish effective grid points; the posteriors are exact 801-point quadrature, no sampling, except Japan, whose importance-sampling ESS is reported).

## Data acquired (user-approved; public, no login)

| context | file | bytes | sha256 | url / query | retrieved (UTC) |
|---|---|---|---|---|---|
| COL | `raw/colombia/negative_controls/inmlcf_accidental_injuries_6cjn-eevd_metadata.json` | 65,077 | `468d7e7f2f8e11c1848cfb04a81d7e48e64b18ad3b76ad4ec769609a7461e917` | https://www.datos.gov.co/api/views/6cjn-eevd.json | 2026-09-30T03:05:11Z |
| COL | `raw/colombia/negative_controls/inmlcf_accidental_injuries_6cjn-eevd_year_dept_ciclovital.csv` | 63,265 | `843db9d1b86d35bf0d5ae4f9a827379532ce0c0ba2f350613e8db40dc8d20554` | https://www.datos.gov.co/resource/6cjn-eevd.csv?%24select=a_o_del_hecho%2Ccodigo_dane_departamento%2Cciclo_vital%2Ccount%28%2A%29&%24group=a_o_del_hecho%2Ccodigo_dane_departamento%2Cciclo_vital&%24limit=50000 (SoQL {"$select": "a_o_del_hecho,codigo_dane_departamento,ciclo_vital,count(*)", "$group": "a_o_del_hecho,codigo_dane_departamento,ciclo_vital", "$limit": "50000"}) | 2026-09-30T03:05:11Z |
| COL | `raw/colombia/negative_controls/inmlcf_accidental_injuries_6cjn-eevd_year_dept_ciclovital_quinquenal.csv` | 219,710 | `0de121862b8349da7bc0cac449d6f0b01b09da28fcb9d938fbbf378457d4aa97` | https://www.datos.gov.co/resource/6cjn-eevd.csv?%24select=a_o_del_hecho%2Ccodigo_dane_departamento%2Cciclo_vital%2Cgrupo_de_edad_quinquenal%2Ccount%28%2A%29&%24group=a_o_del_hecho%2Ccodigo_dane_departamento%2Cciclo_vital%2Cgrupo_de_edad_quinquenal&%24limit=50000 (SoQL {"$select": "a_o_del_hecho,codigo_dane_departamento,ciclo_vital,grupo_de_edad_quinquenal,count(*)", "$group": "a_o_del_hecho,codigo_dane_departamento,ciclo_vital,grupo_de_edad_quinquenal", "$limit": "50000"}) | 2026-09-30T03:05:11Z |
| COL | `raw/colombia/negative_controls/inmlcf_transport_injuries_ezhf-hscf_metadata.json` | 75,455 | `0091b094575d7393548df0035bb4ff779d95c808b9a0b51ec0f372ee5eb7cf78` | https://www.datos.gov.co/api/views/ezhf-hscf.json | 2026-09-30T03:05:07Z |
| COL | `raw/colombia/negative_controls/inmlcf_transport_injuries_ezhf-hscf_year_dept_ciclovital.csv` | 76,890 | `3bcd5daca178963a484ea310a1e747f98443a5000e68f18b320f9a864127dece` | https://www.datos.gov.co/resource/ezhf-hscf.csv?%24select=a_o_del_hecho%2Ccodigo_dane_departamento%2Cciclo_vital%2Ccount%28%2A%29&%24group=a_o_del_hecho%2Ccodigo_dane_departamento%2Cciclo_vital&%24limit=50000 (SoQL {"$select": "a_o_del_hecho,codigo_dane_departamento,ciclo_vital,count(*)", "$group": "a_o_del_hecho,codigo_dane_departamento,ciclo_vital", "$limit": "50000"}) | 2026-09-30T03:05:09Z |
| COL | `raw/colombia/negative_controls/inmlcf_transport_injuries_ezhf-hscf_year_dept_ciclovital_quinquenal.csv` | 319,034 | `3827ae4a12e11b8259c269ddbf07d673ea1f48b9bc6e30c7dd534255061a8634` | https://www.datos.gov.co/resource/ezhf-hscf.csv?%24select=a_o_del_hecho%2Ccodigo_dane_departamento%2Cciclo_vital%2Cgrupo_de_edad_quinquenal%2Ccount%28%2A%29&%24group=a_o_del_hecho%2Ccodigo_dane_departamento%2Cciclo_vital%2Cgrupo_de_edad_quinquenal&%24limit=50000 (SoQL {"$select": "a_o_del_hecho,codigo_dane_departamento,ciclo_vital,grupo_de_edad_quinquenal,count(*)", "$group": "a_o_del_hecho,codigo_dane_departamento,ciclo_vital,grupo_de_edad_quinquenal", "$limit": "50000"}) | 2026-09-30T03:05:10Z |
| COL | `raw/colombia/negative_controls/PPED-AreaSexoEdadDep-2018-2050_VP.xlsx` | 6,631,115 | `ba42efa193e9d194df2ba4036199bed83c688744892fb6ad56a20a6b406f819f` | https://www.dane.gov.co/files/censo2018/proyecciones-de-poblacion/Departamental/PPED-AreaSexoEdadDep-2018-2050_VP.xlsx | 2026-09-30T03:06:11Z |
| HKG | `raw/hong_kong/negative_controls/td_rta_2019_f1.1.xls` | 32,768 | `6b8b1640926d00514cc248d716e9983a5da570d7d61c3d0932682c7ef0c35fa6` | https://www.td.gov.hk/filemanager/en/content_4994/f1.1.xls | 2026-09-30T03:05:19Z |
| HKG | `raw/hong_kong/negative_controls/td_rta_2024_f1_1.pdf` | 83,373 | `08d9b30831d07aef78f4a1a57296eac2e9d3acb11bfde9821284958dc9f79dad` | https://www.td.gov.hk/filemanager/en/content_5240/f1_1.pdf | 2026-09-30T03:05:15Z |
| HKG | `raw/hong_kong/negative_controls/td_rta_2024_f1_1.xls` | 32,768 | `657ec5f32ef9037434666cc5b7e3ab2d0ac9cc685d64ebb6df175063ea709f4b` | https://www.td.gov.hk/filemanager/en/content_5240/f1_1.xls | 2026-09-30T03:05:13Z |
| HKG | `raw/hong_kong/negative_controls/td_rta_2024_f1_2.xls` | 30,720 | `af8d619c1df3440fcfaf3a4a9b3f18bf696e4d7de8ff0d6291d51744a23e2ec7` | https://www.td.gov.hk/filemanager/en/content_5240/f1_2.xls | 2026-09-30T03:05:16Z |
| HKG | `raw/hong_kong/negative_controls/td_rta_2024_f1_3.xls` | 33,280 | `aec42d6d446792c36d141aeb4b64ed7519b6e27a605bdc2b68ac4a13757e6023` | https://www.td.gov.hk/filemanager/en/content_5240/f1_3.xls | 2026-09-30T03:05:17Z |
| JPN | `raw/japan/negative_controls/yokohama_t021202-siin.csv` | 7,649,295 | `64fe27a9aea61494ca97fc513f8d3a4b653d41b72109a0ff59339896c6b924b2` | https://www.city.yokohama.lg.jp/city-info/yokohamashi/tokei-chosa/portal/tokeisho/02.files/t021202-siin.csv | 2026-09-30T03:06:07Z |
| JPN | `raw/japan/negative_controls/yokohama_t021202-year.csv` | 19,330 | `c209a55bfec2d61bcaa84061c82f42d4a29607a03781ee15a0d18d0c8a85b4fb` | https://www.city.yokohama.lg.jp/city-info/yokohamashi/tokei-chosa/portal/tokeisho/02.files/t021202-year.csv | 2026-09-30T03:05:20Z |
| TWN | `raw/taiwan/negative_controls/mohw_all_cause_deaths_1150831.zip` | 10,029,067 | `e52700c312efe5eb9c05b3a123b1b3579519a37a04ad1534a5dc8f086e898b6e` | https://www.mohw.gov.tw/dl-83417-aa29a17f-289d-4934-a868-fe74b3b14afc.html | 2026-09-30T03:06:14Z |

Every file has a `.meta.json` sidecar (url, query, retrieval time, sha256, size, terms); the script recomputes each hash and size and refuses to run if one differs. Not obtained: COL_SIVIGILA_370_POISONING (the INS portal shows its request form - e-mail, entity, purpose - before the Microdatos download, the same access as event 300; it was not attempted); Yokohama t021202.xlsx (single year 2024, outside every cohort window; not needed).

## Autism reproduction (the published autism fits, refitted by this script)

| context | check | result | passes |
|---|---|---|---|
| COL | block input sha256 equals the gate record; 12 setting x prior rows; four setting curves; future+3 | max rel diff 3.9e-11; curves 1.6e-09; future+3 2.7e-11; 0.764 (0.509-1.074) | True |
| HKG | frozen 60-cell block refitted vs integration_audit | max rel diff 0.0e+00; curves 0.0e+00; 0.892 (0.657-1.165); published future+3 gate PASS | True |
| JPN | 3341.fit_is mirror on the 9 Kohoku cohorts vs core-v2 us_alone (p18) | posterior fields equal True; ESS equal True; log-likelihood arrays identical True; future+3 equal True; 0.85 (0.28-2.32) | True |
| TWN | 720-cell block and its future+3 twin vs production | curves 3.1e-09 / 1.3e-08; summaries 9.1e-11 / 1.9e-11; 1.071 (0.868-1.442) | True |

## Colombia varicella: COL_VARICELLA_school_autism (pooled); candidate COL_INMLCF_TRANSPORT_AND_ACCIDENTAL_INJURIES

**Design.** Model: COL_VARICELLA_school_autism built by global_adapters_americas.colombia (imported unchanged; its block input sha256 must equal the gate record f22bbfcc...) and fitted by international_likelihoods.fit (linear / spline trend x rho 0, .6, .85 kernels x the block's six varicella exposure scenarios, equal-weight likelihood average, PCV booster / rotavirus / HepA nuisance columns, 801-point grid, location -5.5, noise_sd .25; reference setting, N(0, ln 2) prior). Only y and the population scaling are swapped, through global_block_tools.BlockSet.add on a subset of the block's own 504 region x level x school-year cells (their exposure scenarios, labels and nuisance columns kept). Control counts: INMLCF medico-legal records by event department (DANE code; the block's six smallest departments 88, 91, 94, 95, 97, 99 pooled) x life-cycle band x event year 2019-2024; (00 a 05) -> preescolar, (06 a 11) -> basica primaria, (12 a 17) -> secundaria+media. Primary y = ln(count / DANE PPED 2018-2050 (VP) population of the band's single ages, both sexes, area Total, summed over the region's departments), population scaling = that population. Matching autism: the autism block restricted to the control's cells. DDD: y = ln(autism / enrolment) - ln(control rate) on the control's cells with the block's enrolment scaling (as 3337). Future+3: international_likelihoods.paired_future for the control, the matching autism and the DDD.

**Alignment compromises (every one).**

- Event vs school vs residence geography: INMLCF records the department where the injury event happened (departamento del hecho); the autism numerator is the department of the pupil's school (sede); the exposure is the department's PAI uptake. Children injured outside their department are placed in the wrong cell.
- Life-cycle bands vs school levels: (00-05) vs preescolar (pre-jardin, jardin, transicion; ages ~3-5), (06-11) vs basica primaria (grades 1-5, nominal ages 6-10), (12-17) vs secundaria + media (grades 6-11, nominal ages 11-16). Every cell keeps the block's exposure, the grade-mix exposure of the level's enrolled pupils, not an exposure recomputed for the band's single ages: the (00-05) band includes ages 0-2 (never in preescolar) and the (06-11) and (12-17) bands reach one year older than the levels.
- Event year vs school year: the calendar year of the event is taken as the school year (calendario A runs February-November; calendario B schools run August-June).
- Denominator: the DANE population of the band (all children), not enrolment; the enrolment sensitivity uses the block's own denominator. DANE PPED is the post-COVID update (VP, 30 July 2025).
- Selection: INMLCF medico-legal examinations are the subset of injuries that reach a forensic examination, not all injuries; the 2020-2021 lockdowns cut them sharply. A national year shock is absorbed by the kernel's year effect, a level-specific one is not (excl_2020_2021 sensitivity).
- Three child transport records 2019-2024 with unknown department (code 999) are excluded.

**Cells and losses.**

- transport: 342,796 records in the aggregate; child bands 2019-2024 13,748, of which 3 with unknown department (code 999, excluded); 'Sin información' band 2019-2024 (all ages) 0; placed on the 504 cells 13,745.
- accidental: 27,621 records in the aggregate; child bands 2019-2024 2,802, of which 0 with unknown department (code 999, excluded); 'Sin información' band 2019-2024 (all ages) 0; placed on the 504 cells 2,802.
- DANE PPED: 33 departments, single ages of both sexes sum to the total in every row (max abs diff 0).
- transport__observed: 500 of 504 cells used (4 dropped; 4 zero-count cells on the 504); dropped: COL_18 preescolar 2020, COL_70 basica_primaria 2020, COL_86 preescolar 2021, COL_86 basica_primaria 2022.
- transport__complete_series: 480 of 504 cells used (24 dropped; 4 zero-count cells on the 504); dropped: COL_18 preescolar 2019, COL_18 preescolar 2020, COL_18 preescolar 2021, COL_18 preescolar 2022, COL_18 preescolar 2023, COL_18 preescolar 2024, COL_70 basica_primaria 2019, COL_70 basica_primaria 2020, COL_70 basica_primaria 2021, COL_70 basica_primaria 2022, COL_70 basica_primaria 2023, COL_70 basica_primaria 2024, COL_86 preescolar 2019, COL_86 preescolar 2020, COL_86 preescolar 2021, COL_86 preescolar 2022, COL_86 preescolar 2023, COL_86 preescolar 2024, COL_86 basica_primaria 2019, COL_86 basica_primaria 2020, COL_86 basica_primaria 2021, COL_86 basica_primaria 2022, COL_86 basica_primaria 2023, COL_86 basica_primaria 2024.
- transport__all_cells_cc05: 504 of 504 cells used (0 dropped; 4 zero-count cells on the 504).
- transport__observed__enrolment: 500 of 504 cells used (4 dropped; 4 zero-count cells on the 504); dropped: COL_18 preescolar 2020, COL_70 basica_primaria 2020, COL_86 preescolar 2021, COL_86 basica_primaria 2022.
- transport__observed__excl_2020_2021: 335 of 504 cells used (169 dropped; 4 zero-count cells on the 504).
- accidental__observed: 416 of 504 cells used (88 dropped; 88 zero-count cells on the 504).
- transport_plus_accidental__observed: 501 of 504 cells used (3 dropped; 3 zero-count cells on the 504); dropped: COL_18 preescolar 2020, COL_70 basica_primaria 2020, COL_86 preescolar 2021.

**Results.**

Autism reference: Autism (School Census) on its 504 cells 0.76 (0.51-1.07); future+3 0.99 (0.29-3.77) PASS.

| control | role | cell rule | denominator | cells used / total | count used | r (95% CrI) | P(r>1) | autism, same cells | width / autism | class | future+3 (reference) | future+3 | DDD autism / control | DDD P(r>1) | Kish grid pts |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Transport Injuries (INMLCF, Non-Fatal) | primary | observed | DANE band population | 500 / 504 | 13,745 | 1.36 (0.99-1.93) | 0.973 | 0.76 (0.51-1.07) | 0.90 | CONSISTENT | 1.07 (0.35-3.28) | PASS | 0.63 (0.39-1.02) | 0.029 | 36 |
| Transport Injuries (INMLCF, Non-Fatal) | sensitivity | complete_series | DANE band population | 480 / 504 | 13,664 | 1.37 (0.99-1.94) | 0.973 | 0.78 (0.53-1.10) | 0.91 | CONSISTENT | 1.09 (0.35-3.37) | PASS | 0.64 (0.39-1.05) | 0.036 | 36 |
| Transport Injuries (INMLCF, Non-Fatal) | sensitivity | all_cells_cc05 | DANE band population | 504 / 504 | 13,745 | 1.39 (1.03-2.02) | 0.982 | 0.76 (0.51-1.07) | 0.90 | BIAS SIGNAL (r > 1 (harm direction)) | 1.11 (0.37-3.27) | PASS | 0.62 (0.39-1.00) | 0.025 | 36 |
| Transport Injuries (INMLCF, Non-Fatal) | sensitivity | observed | enrolment (block) | 500 / 504 | 13,745 | 1.49 (0.90-2.30) | 0.950 | 0.76 (0.51-1.07) | 1.27 | CONSISTENT | 0.98 (0.31-3.12) | PASS | 0.60 (0.37-1.06) | 0.036 | 53 |
| Transport Injuries (INMLCF, Non-Fatal) | sensitivity | observed | DANE band population | 335 / 504 | 10,959 | 1.18 (0.78-1.80) | 0.801 | 1.31 (0.89-1.98) | 1.05 | CONSISTENT | 0.80 (0.25-3.20) | PASS | 1.11 (0.62-2.18) | 0.634 | 47 |
| Accidental Injuries (INMLCF, Non-Fatal) | secondary | observed | DANE band population | 416 / 504 | 2,802 | 1.56 (0.93-3.97) | 0.954 | 0.82 (0.55-1.17) | 1.93 | CONSISTENT | 1.31 (0.40-3.82) | PASS | 0.58 (0.31-1.05) | 0.035 | 77 |
| Transport + Accidental Injuries | secondary | observed | DANE band population | 501 / 504 | 16,547 | 1.33 (1.01-1.96) | 0.979 | 0.76 (0.51-1.07) | 0.89 | BIAS SIGNAL (r > 1 (harm direction)) | 0.92 (0.34-2.72) | PASS | 0.63 (0.40-0.99) | 0.022 | 34 |

Class under the other kernel settings and priors (descriptive; never changes it): transport__observed: reference|reference CONSISTENT, reference|wide CONSISTENT, reference|harm_centered BIAS SIGNAL, linear_only|reference BIAS SIGNAL, linear_only|wide BIAS SIGNAL, linear_only|harm_centered BIAS SIGNAL, spline_only|reference CONSISTENT, spline_only|wide CONSISTENT, spline_only|harm_centered CONSISTENT, rho_0_85|reference CONSISTENT, rho_0_85|wide CONSISTENT, rho_0_85|harm_centered BIAS SIGNAL; transport__complete_series: reference|reference CONSISTENT, reference|wide CONSISTENT, reference|harm_centered BIAS SIGNAL, linear_only|reference BIAS SIGNAL, linear_only|wide BIAS SIGNAL, linear_only|harm_centered BIAS SIGNAL, spline_only|reference CONSISTENT, spline_only|wide CONSISTENT, spline_only|harm_centered CONSISTENT, rho_0_85|reference CONSISTENT, rho_0_85|wide CONSISTENT, rho_0_85|harm_centered BIAS SIGNAL; transport__all_cells_cc05: reference|reference BIAS SIGNAL, reference|wide BIAS SIGNAL, reference|harm_centered BIAS SIGNAL, linear_only|reference BIAS SIGNAL, linear_only|wide BIAS SIGNAL, linear_only|harm_centered BIAS SIGNAL, spline_only|reference CONSISTENT, spline_only|wide CONSISTENT, spline_only|harm_centered CONSISTENT, rho_0_85|reference BIAS SIGNAL, rho_0_85|wide BIAS SIGNAL, rho_0_85|harm_centered BIAS SIGNAL; transport__observed__enrolment: reference|reference CONSISTENT, reference|wide CONSISTENT, reference|harm_centered CONSISTENT, linear_only|reference BIAS SIGNAL, linear_only|wide BIAS SIGNAL, linear_only|harm_centered BIAS SIGNAL, spline_only|reference CONSISTENT, spline_only|wide CONSISTENT, spline_only|harm_centered CONSISTENT, rho_0_85|reference CONSISTENT, rho_0_85|wide CONSISTENT, rho_0_85|harm_centered CONSISTENT; transport__observed__excl_2020_2021: stable (CONSISTENT); accidental__observed: reference|reference CONSISTENT, reference|wide CONSISTENT, reference|harm_centered CONSISTENT, linear_only|reference CONSISTENT, linear_only|wide CONSISTENT, linear_only|harm_centered CONSISTENT, spline_only|reference CONSISTENT, spline_only|wide BIAS SIGNAL, spline_only|harm_centered BIAS SIGNAL, rho_0_85|reference CONSISTENT, rho_0_85|wide CONSISTENT, rho_0_85|harm_centered CONSISTENT; transport_plus_accidental__observed: reference|reference BIAS SIGNAL, reference|wide BIAS SIGNAL, reference|harm_centered BIAS SIGNAL, linear_only|reference BIAS SIGNAL, linear_only|wide BIAS SIGNAL, linear_only|harm_centered BIAS SIGNAL, spline_only|reference CONSISTENT, spline_only|wide CONSISTENT, spline_only|harm_centered CONSISTENT, rho_0_85|reference CONSISTENT, rho_0_85|wide CONSISTENT, rho_0_85|harm_centered CONSISTENT.

**Reading.**

- CONSISTENT: no non-specific bias detected at this precision (this does not validate the autism estimate). Primary: Transport Injuries (INMLCF, Non-Fatal) (transport__observed) r 1.36 (0.99-1.93); class CONSISTENT (CrI includes 1).
- 95% CrI excluding 1 (beside the reading): transport__all_cells_cc05 1.39 (1.03-2.02) (r > 1 (harm direction); class BIAS SIGNAL); transport_plus_accidental__observed 1.33 (1.01-1.96) (r > 1 (harm direction); class BIAS SIGNAL).

Descriptive notes (computed from the fits; outside the rule and added after the results were read; they never change a class or the reading):

- The primary control leans to the harm side (median 1.36, P(r > 1) 0.973) while autism on the same cells leans protective (0.76); the DDD (autism relative to transport injuries) is 0.63 (0.39-1.02), P(r > 1) 0.029. The primary's lower limit is 0.991.
- 2 of the 6 other Colombian control fits have a 95% CrI excluding 1, all in the harm direction: transport__all_cells_cc05 1.39 (1.03-2.02); transport_plus_accidental__observed 1.33 (1.01-1.96). The primary's class under the other settings and priors: reference|harm_centered BIAS SIGNAL, linear_only|reference BIAS SIGNAL, linear_only|wide BIAS SIGNAL, linear_only|harm_centered BIAS SIGNAL, rho_0_85|harm_centered BIAS SIGNAL (all others CONSISTENT).
- Dropping school years 2020-2021 moves both fits: the control to 1.18 (0.78-1.80) and autism on the same 335 cells to 1.31 (0.89-1.98) (from 0.76 (0.51-1.07) on all 504), so the Colombian autism estimate itself depends on the lockdown years.
- Future+3 placebo: every Colombian control, matching-autism and DDD fit passes.

## Hong Kong PCV: HKG_PCV_autism (pooled); candidate AA-HKG-02

**Design.** Model: international_likelihoods.fit, the fit of the published HKG_PCV_autism estimate (linear / spline trend x rho 0, .6, .85 kernels with population scaling x the block's three exposure scenarios, equal-weight average, 801-point grid, N(0, ln 2) prior); the frozen autism block (project_results/integration_audit/inputs.json) is refitted and must reproduce the published summaries. Control cells: Transport Department road-traffic casualties (killed + seriously + slightly injured; Figure 1.1 by age group: 2019 edition for 2009-2013, 2024 edition for 2014-2024) at WHO bands 5-9 / 10-14 / 15-19 (kernel ages 7, 12, 17; the autism ages) x calendar year; region 'Hong Kong'; y = ln(casualties / WHO MDB population of the band) (discovery_global_mortality population table, held), population scaling = that population; cohort = year - age. Exposure: the HKG_PCV_autism scenarios (routine 908 cohort map, 20% prior private receipt, catch-up) as functions of birth cohort, mapped by Lexis halves over the band's single ages (3343's HkgPath, imported read-only; the published gate provider's rules). A cell enters iff count > 0; a control is fittable iff >= 20 such cells and the exposure varies. Future+3: the same path at cohort c + 3 on the identical cells. No autism count exists on these cells, so no shared-cell DDD: the independent-posterior contrast ln r_autism - ln r_control (3343.contrast, grid convolution) is reported instead.

**Alignment compromises (every one).**

- Different cells: autism cells are grades P1-S6 x school years 2015/16-2019/20 (EDB special educational needs); the control cells are age bands x calendar years 2009-2024. The width reference is autism on its own 60 cells ('reference on different cells'); no shared-cell DDD exists.
- Exposure: mapped from birth cohort by Lexis halves averaged over the band's single ages (a death or casualty at completed age a in year t -> 0.5 f(t - a - 1) + 0.5 f(t - a)), not the grade-cohort map of the autism cells.
- Outcome system: police-reported road-traffic casualties of any road user by accident location in Hong Kong (territory-wide; residence not recorded), not deaths or hospital records.
- Population: WHO MDB mid-year populations; TD's own rates imply populations up to 3.0% different (2009-2019). The 0-4 band has no WHO population (no age 0), so it is not used.
- Editions: 2009-2013 from the 2019 edition, 2014-2024 from the 2024 edition (equal in all 24 overlapping cells).
- 2020-2022 COVID school suspensions cut child casualties; the single-region kernel has one year effect shared by all bands, so band-specific pandemic shocks remain (excl_2020_2022 sensitivity).
- Pedestrian casualties: only 2014-2024 (the 2019 edition's Figure 1.3 was not a listed file).

**Cells and losses.**

- TD Figure 1.1: 2019 edition 2009-2013, 2024 edition 2014-2024; the overlap years 2014-2019 agree in 24 band-year cells: True (max abs diff 0). Figure 1.2's 0-19 sub-totals equal the band sums: all users True, pedestrians True.
- Population: WHO MDB (discovery table). The TD 2019 edition's rates per 1,000 imply the population TD used: equal to WHO's (within 0.1%) in 2009-2016 and -0.7% to +3.0% in the other years (the TD file does not say which population estimates it used; max |difference| 3.0% over 44 cells). WHO's are used throughout.
- road__bands_2009_2024: 48 of 48 cells (no zero cells; casualties per cell 194 or more).
- road__bands_2013_2021: 27 of 27 cells (no zero cells; casualties per cell 207 or more).
- road__bands_excl_2020_2022: 39 of 39 cells (no zero cells; casualties per cell 302 or more).
- pedestrian__bands_2014_2024: 33 of 33 cells (no zero cells; casualties per cell 49 or more).

**Results.**

Autism reference: Autism (EDB SEN, P1-S6) on its 60 cells 0.89 (0.66-1.16); future+3 1.27 (0.86-1.92) PASS.

| control | role | cells | cells used / total | count used | r (95% CrI) | P(r>1) | width / autism | class | future+3 (reference) | future+3 | autism relative to control | P(>1) | Kish / ESS |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Road-Traffic Casualties (Police-Reported) | primary | 5-9/10-14/15-19 x 2009-2024 | 48 / 48 | 22,323.0 | 0.66 (0.48-0.92) | 0.008 | 1.15 | BIAS SIGNAL (r < 1 (protective direction)) | 0.88 (0.56-1.64) | PASS | 1.34 (0.86-2.05) | 0.908 | 38 |
| Road-Traffic Casualties (Police-Reported) | sensitivity | 5-9/10-14/15-19 x 2013-2021 | 27 / 27 | 12,088.0 | 0.98 (0.59-1.63) | 0.471 | 1.78 | CONSISTENT | 1.03 (0.54-1.96) | PASS | 0.90 (0.50-1.61) | 0.362 | 59 |
| Road-Traffic Casualties (Police-Reported) | sensitivity | 5-9/10-14/15-19 x 2009-2024 excl. 2020-2022 | 39 / 39 | 19,111.0 | 0.74 (0.57-0.94) | 0.008 | 0.88 | BIAS SIGNAL (r < 1 (protective direction)) | 0.86 (0.61-1.20) | PASS | 1.21 (0.82-1.75) | 0.837 | 29 |
| Pedestrian Casualties | secondary | 5-9/10-14/15-19 x 2014-2024 | 33 / 33 | 3,366.0 | 0.76 (0.52-1.15) | 0.080 | 1.40 | CONSISTENT | 0.95 (0.62-1.73) | PASS | 1.17 (0.70-1.87) | 0.741 | 45 |

Class under the other kernel settings and priors (descriptive; never changes it): road__bands_2009_2024: stable (BIAS SIGNAL); road__bands_2013_2021: stable (CONSISTENT); road__bands_excl_2020_2022: stable (BIAS SIGNAL); pedestrian__bands_2014_2024: stable (CONSISTENT).

**Reading.**

- BIAS SIGNAL: the design returns an 'effect' on injuries (r < 1 (protective direction)). Primary: Road-Traffic Casualties (Police-Reported) (road__bands_2009_2024) r 0.66 (0.48-0.92); class BIAS SIGNAL (r < 1 (protective direction)) (CrI excludes 1).
- 95% CrI excluding 1 (beside the reading): road__bands_2009_2024 0.66 (0.48-0.92) (r < 1 (protective direction); class BIAS SIGNAL); road__bands_excl_2020_2022 0.74 (0.57-0.94) (r < 1 (protective direction); class BIAS SIGNAL).

Descriptive notes (computed from the fits; outside the rule and added after the results were read; they never change a class or the reading):

- The protective-direction signal is on the same side of 1 as autism's 0.89; the independent-posterior contrast (autism relative to road casualties) is 1.34 (0.86-2.05), P(> 1) 0.908. The class is stable under every kernel setting and prior; the future+3 placebo passes (0.88 (0.56-1.64)). Without 2020-2022: 0.74 (0.57-0.94); on 2013-2021 (3343's band window): 0.98 (0.59-1.63); pedestrian casualties 2014-2024: 0.76 (0.52-1.15).
- Round 1 (3343) could not test this design with WHO MDB injury deaths (UNINFORMATIVE); the road-casualty control's CrI log width is 1.15x autism's, so it is the first informative injury control for Hong Kong.

## Japan MMR (Kohoku): JPN_MMR (core v2, Kohoku) (pooled (core v2)); candidate AA-JPN-06

**Design.** Model: the core-v2 Japan block used unchanged (rpl.japan_kernel linear / quadratic / spline x rho 0 / .5 / .85; CohortLikelihood location -5.5, shape 2.5, noise_sd .25; rpl.prepare_shifts logit-normal coverage jitter; run_us_joint_likelihood.proposal_v2 eta4, seeds 23091 / 23092, p18 (p20 if the pooled p18 ESS < 100); N(0, ln 2) priors; the Japan-only posterior of us_alone), run by a line-for-line mirror of 3341.fit_is whose autism refit must reproduce the published own-data posterior bit for bit. Cells: the autism block's nine birth cohorts 1988-1996; outcome = cohort-cumulative Yokohama City deaths at completed ages 15-19 (t021202-siin.csv; a death in year t in the 15-19 group is shared among cohorts t-20 .. t-15 with weights 0.1, 0.2, 0.2, 0.2, 0.2, 0.1); denominator = the cohort's national live births (WHO MDB, 3341.live_births); exposure = the Kohoku MMR coverage path of the same cohort, unchanged. Future+3: 3341.placebo (global_placebo_family JPN_MMR family on the swapped records). DDD: 3341's log-ratio device ddd_records (Kohoku autism rate relative to the centred Yokohama injury rate, Kohoku denominators).

**Alignment compromises (every one).**

- Year coverage verified before use: t021202-year.csv (all causes) covers 1998-2024 and t021202-siin.csv (by cause) 2000-2024, so the discovery record's intended cells (Yokohama deaths at ages 0-9 in 1989-2003 for the 1988-1996 cohorts) are not available by cause. The only 5-year age group whose Lexis window (years c+15 to c+20) lies inside 2000-2024 for all nine cohorts is 15-19: the control is deaths at completed ages 15-19, not at the autism ages (cumulative incidence to age 7).
- Cohort attribution from 5-year age groups assumes deaths uniform over the five single ages (weights 0.1, 0.2, 0.2, 0.2, 0.2, 0.1).
- Geography: Yokohama city (18 wards, deaths by residence), which contains Kohoku ward; not Kohoku itself.
- Denominator: national live births of the cohort (Yokohama births are not held); cohort-cumulative city deaths at 15-19 also depend on migration into and out of Yokohama between birth and ages 15-19.
- 不慮の事故 (V01-X59) includes forces of nature (X30-X39), which the round-1 WHO composites excluded.

**Cells and losses.**

- Yokohama t021202-year.csv (all causes) covers 1998-2024; the cause-specific t021202-siin.csv covers 2000-2024. JPN_MMR needs ages 0-9 in 1989-2003 (the discovery record): not available by cause. No cohort is lost at ages 15-19 (windows c+15 to c+20 = 2003-2016).
- Accidents (不慮の事故): cohort-cumulative deaths 1988-1996 = 12.3, 12, 12.1, 10.9, 9.2, 8.1, 7, 6.2, 6.6.
- External Causes Except Suicide: cohort-cumulative deaths 1988-1996 = 15.6, 15.5, 15.8, 14, 11.9, 10.8, 9.5, 8.5, 8.8.
- Transport Accidents (交通事故): cohort-cumulative deaths 1988-1996 = 10, 9.8, 9.5, 8.2, 6.3, 5, 4.4, 4.2, 4.5.

**Results.**

Autism reference: Autism (Kohoku, All ASD) on its 9 cells 0.85 (0.28-2.32); future+3 0.89 (0.25-2.68) PASS.

| control | role | cells | cells used / total | count used | r (95% CrI) | P(r>1) | width / autism | class | future+3 (reference) | future+3 | DDD autism / control | P(>1) | Kish / ESS |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Accidents (不慮の事故) | primary | cohorts 1988-1996, ages 15-19 | 9 / 9 | 84.4 | 1.00 (0.41-2.21) | 0.500 | 0.79 | UNINFORMATIVE | 0.68 (0.24-1.57) | PASS | 0.84 (0.27-2.35) | 0.374 | ESS 227,649 |
| External Causes Except Suicide | secondary | cohorts 1988-1996, ages 15-19 | 9 / 9 | 110.4 | 0.98 (0.41-2.17) | 0.480 | 0.79 | UNINFORMATIVE | 0.65 (0.23-1.49) | PASS | 0.85 (0.27-2.33) | 0.375 | ESS 225,872 |
| Transport Accidents (交通事故) | secondary | cohorts 1988-1996, ages 15-19 | 9 / 9 | 61.9 | 1.20 (0.42-2.81) | 0.650 | 0.90 | UNINFORMATIVE | 0.66 (0.22-1.63) | PASS | 0.80 (0.26-2.27) | 0.336 | ESS 222,398 |

**Reading.**

- UNINFORMATIVE -> NOT TESTABLE: no conclusion about bias in this design. Primary: Accidents (不慮の事故) (accidents__ages15_19) r 1.00 (0.41-2.21); class UNINFORMATIVE (b: thin (minimum cohort count 6.2 < 16)).
- No control interval (primary, secondary or sensitivity) excludes 1.

Descriptive notes (computed from the fits; outside the rule and added after the results were read; they never change a class or the reading):

- Every Yokohama outcome is thin by criterion (b) (minimum cohort deaths: Accidents (不慮の事故) 6.2, External Causes Except Suicide 8.5, Transport Accidents (交通事故) 4.2; < 16), as the build counts showed before the rule; the posteriors (primary 1.00 (0.41-2.21), CrI log width 0.79x autism's) are reported but carry no reading. Every future+3 placebo passes.

## Taiwan varicella: TWN_VARICELLA_autism (report-only); candidate AA-TWN-02

**Design.** Model: TWN_VARICELLA_autism (report-only) reused unchanged: 3346.base_panel (taiwan_panel + taiwan_enrolment; 720 cells = 21 harmonized counties x {grades 1-9, senior high} x school years 2008-2025, offshore senior-high series dropped) and 3346.make_block (taiwan_exposures: six varicella scenarios; taiwan_extra: Hib / PCV / HepA / senior-high x post-2014 nuisance columns; BlockSet.add), fitted by international_likelihoods.fit (as global_fit._fit_one). Control: MOHW 全死因 underlying-cause deaths, both sexes, township of residence -> county (period-specific code lists of 欄位說明.xls; NFKC, 台 -> 臺, aao.TWN_MERGE), age codes 07 + 08 (5-14) on grades 1-9 and 09 (15-19) on senior high, calendar year t on school year t (starting August t). y = ln(deaths / students), population scaling = students (the block's). Matching autism: the autism block restricted to the control's cells. DDD: y = ln(autism / deaths) on those cells. Future+3: taiwan_exposures(shift 3), unshifted nuisance columns, for control, matching autism and DDD.

**Alignment compromises (every one).**

- Level vs age: grades 1-9 of school year t (born 2 Sept t-15 .. 1 Sept t-6) vs deaths at completed ages 5-14 in calendar year t (born t-15 .. t-5); senior high (born 2 Sept t-18 .. 1 Sept t-15) vs deaths at 15-19 (born t-20 .. t-15), older and wider. Every cell keeps the block's exposure (the grade birth window), not an exposure recomputed for the death ages.
- School year vs calendar year: school year t (August t - July t+1) takes calendar-year-t deaths (calendar_next sensitivity: t+1).
- Geography: deaths by township of residence aggregated to the block's 21 harmonized counties (2010 mergers and Kinmen + Lienchiang pooled); the autism numerator is by school county.
- Denominator: enrolled students of the level (the block's), not the resident population by age.
- 事故傷害 includes forces of nature (X30-X39): Typhoon Morakot (August 2009) gives 高雄市 126 accident deaths at ages 5-19 in 2009 against 39-47 in 2008 and 2010-2015 (excl_kaohsiung_2009 sensitivity).
- The offshore senior-high series (Penghu, Kinmen + Lienchiang) are not in the block; their 40 accident deaths at 15-19 in 2008-2025 are excluded.
- Deaths are rare at grades 1-9: zero cells are dropped by the primary rule (outcome-dependent selection), bounded by the series and +0.5 sensitivities.

**Cells and losses.**

- MOHW zip: ROC 97-114 (2008-2025) files; cause codes {'97|accidents': '38', '97|assault': '40', '109|accidents': '39', '109|assault': '41'}; age codes {'國中小': ['07', '08'], '高級中等學校': ['09']}; township deaths not mappable to a county: 0; death counties outside the block: []; offshore senior-high deaths dropped with their series (the block drops Penghu and Kinmen+Lienchiang senior high): {'accidents': 40.0, 'assault': 0.0}.
- accidents__observed: 628 of 720 cells used (92 dropped: 92 zero-count cells, by level {'國中小': 87, '高級中等學校': 5}).
- accidents__series: 432 of 720 cells used (288 dropped: 92 zero-count cells, by level {'國中小': 87, '高級中等學校': 5}).
- accidents__all_cells_cc05: 720 of 720 cells used (0 dropped: 92 zero-count cells, by level {'國中小': 87, '高級中等學校': 5}).
- accidents__observed__calendar_next: 593 of 720 cells used (127 dropped: 87 zero-count cells, by level {'國中小': 83, '高級中等學校': 4}, 40 cells without a following calendar year).
- accidents__observed__excl_kaohsiung_2009: 626 of 720 cells used (94 dropped: 92 zero-count cells, by level {'國中小': 87, '高級中等學校': 5}).
- accidents_plus_assault__observed: 637 of 720 cells used (83 dropped: 83 zero-count cells, by level {'國中小': 78, '高級中等學校': 5}).

**Results.**

Autism reference: Autism (Special Education) on its 720 cells 1.07 (0.87-1.44); future+3 1.40 (1.12-2.20) FAIL.

| control | role | cell rule | denominator | cells used / total | count used | r (95% CrI) | P(r>1) | autism, same cells | width / autism | class | future+3 (reference) | future+3 | DDD autism / control | DDD P(r>1) | Kish grid pts |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Accident Deaths (事故傷害) | primary | observed | students (block) | 628 / 720 | 5,676 | 0.91 (0.48-2.05) | 0.388 | 1.07 (0.86-1.46) | 2.74 | UNINFORMATIVE | 0.73 (0.38-1.51) | PASS | 1.08 (0.56-2.65) | 0.588 | 80 |
| Accident Deaths (事故傷害) | sensitivity | series | students (block) | 432 / 720 | 5,124 | 1.32 (0.67-3.40) | 0.786 | 0.90 (0.68-1.16) | 3.03 | UNINFORMATIVE | 0.89 (0.45-2.08) | PASS | 0.59 (0.30-1.23) | 0.069 | 91 |
| Accident Deaths (事故傷害) | sensitivity | all_cells_cc05 | students (block) | 720 / 720 | 5,676 | 1.26 (0.67-3.15) | 0.761 | 1.07 (0.87-1.44) | 3.06 | UNINFORMATIVE | 0.94 (0.50-2.11) | PASS | 0.78 (0.41-1.65) | 0.222 | 85 |
| Accident Deaths (事故傷害) | sensitivity | observed (school year Y <- calendar Y+1) | students (block) | 593 / 720 | 5,232 | 0.93 (0.51-2.00) | 0.410 | 1.06 (0.84-1.44) | 2.54 | UNINFORMATIVE | 0.58 (0.30-1.05) | PASS | 1.30 (0.59-3.51) | 0.737 | 74 |
| Accident Deaths (事故傷害) | sensitivity | observed | students (block) | 626 / 720 | 5,550 | 0.96 (0.51-2.18) | 0.445 | 1.07 (0.86-1.47) | 2.74 | UNINFORMATIVE | 0.75 (0.40-1.52) | PASS | 1.04 (0.54-2.49) | 0.544 | 80 |
| Accident + Assault Deaths | secondary | observed | students (block) | 637 / 720 | 5,924 | 0.96 (0.51-2.18) | 0.446 | 1.07 (0.86-1.45) | 2.78 | UNINFORMATIVE | 0.82 (0.43-1.73) | PASS | 1.04 (0.54-2.48) | 0.542 | 80 |

Class under the other kernel settings and priors (descriptive; never changes it): accidents__observed: stable (UNINFORMATIVE); accidents__series: stable (UNINFORMATIVE); accidents__all_cells_cc05: stable (UNINFORMATIVE); accidents__observed__calendar_next: stable (UNINFORMATIVE); accidents__observed__excl_kaohsiung_2009: stable (UNINFORMATIVE); accidents_plus_assault__observed: stable (UNINFORMATIVE).

**Reading.**

- UNINFORMATIVE -> NOT TESTABLE: no conclusion about bias in this design. Primary: Accident Deaths (事故傷害) (accidents__observed) r 0.91 (0.48-2.05); class UNINFORMATIVE (CrI log width 1.452 > 2.5 x 0.531).
- No control interval (primary, secondary or sensitivity) excludes 1.

Descriptive notes (computed from the fits; outside the rule and added after the results were read; they never change a class or the reading):

- Every Taiwan control is UNINFORMATIVE by the width criterion (CrI log width 2.54-3.06x autism's on the same cells); none excludes 1. The autism block's own future+3 fails (1.40 (1.12-2.20)). On the control cells the future+3 placebo fails for the control itself in 0 of 6 fits, for autism on the same cells in 6 of 6, and for the DDD in 1 of 6 (accidents__observed__calendar_next).
- Correction to the rule text (known_at_writing and the TWN alignment compromises): 高雄市 had 126 accident deaths at ages 5-19 in 2009 against 38-47 in 2008 and 2010-2015 (2015: 38), not 39-47; the rule file is hash-locked and unchanged, and the sensitivity it defines (the two 高雄市 2009 cells dropped) is unaffected.

## Caveats common to all four contexts

- Injury controls test the exposure path and model on outcomes vaccination cannot cause; a CONSISTENT class is not evidence that the autism estimate is unbiased, and UNINFORMATIVE means the control cannot detect bias of the size that matters.
- Every model is the matching autism block's model, unchanged, including its Gaussian log-rate working likelihood with population-scaled residual variance and fixed noise 0.25 (or the Japan CohortLikelihood); it describes large counts better than small ones (Japan's cohort cells, Taiwan's grade 1-9 cells, small Colombian preescolar cells).
- The DDDs share every cell and exposure with the separate fits; they restate the autism-versus-control comparison and are not independent evidence.
- Nothing here changes an autism estimate, a gate decision, blocks.csv, gate_decisions.json or the pooled hierarchy.