CORRECTION 6 - COMPREHENSIVE REV12 REANALYSIS

Supersedes the active analysis and release claims of rev11. Historical files
are retained as records of earlier beliefs and are not current instructions.

Confirmed scoring corrections: 22 category changes, including 15 core and
7 robustness trials. Four prior joint successes fail script fidelity (three
core, one robustness). Detailed old/new fields and exact raw strings are in
results/full/scoring_changes.json; original raw JSON files are unchanged.

Core non-Western neutral categories A/B/C/D/E become 112/749/2/213/424.
Core non-Western preserve categories A/B/C/D/E become 93/898/4/200/305.
Joint successes become 41/1500 neutral and 33/1500 preserve.
Arabic-Indic neutral best per-model joint score becomes 30/100, not 31/100.

H1 is restored to the Arabic-Indic primary population; the all-script pooled
analysis is relabelled exploratory. H2 now fits the specified random-intercept
structure and standardizes predictions; the raw contrast is supplementary.
Script fidelity now has its digit-emitting denominator. H3 includes paired
tests and valid Holm correction. The former ICC/design effect are withdrawn.

The bootstrap ambiguity is documented, with two different targets and matched
generation/statistic definitions. Basic and percentile results are exposed
without selecting a method by the claim it yields. Fit convergence, fixed
initialization, diagnostics and full draws are recorded. No promise that
earlier point estimates or verdicts remain unchanged is made.

The source generator/runner paths are repaired in the new copy. New runs
require a model commit hash and isolated output destination. Missing original
model revisions, font hashes and rendered stimulus files cannot be recovered
from the supplied archive and are disclosed as historical limitations.

The manuscript's original LaTeX project was not supplied. A replacement text
draft and release report accompany the analysis; they are not asserted to be
an official ACL-formatted submission. Source integration, final rendering,
author declarations and independent statistical approval remain pending.
