The inattentional gap in task conditioned AI models that omit otherwise reportable safety critical signals (reproducibility deposit)

Code, per-item frozen results, prompts, and documentation for the manuscript "The inattentional gap in task conditioned AI models that omit otherwise reportable safety critical signals" (Kwan Soo Shin et al., PolymathMinds Lab, Asan, Republic of Korea). Every reported statistic reproduces offline from the frozen per-item JSON files with the included analysis scripts; API rerun scripts are provided for extension. Stimulus images (source and composited) are third-party-derived and are withheld from public redistribution because per-file rights records could not be documented (SOURCE_PROVENANCE_STATEMENT.md); SHA-256 manifests anchor the frozen set, and the images are available from the corresponding author for editorial review and research verification. v1.2 (2026-08-01): rights closure (stimulus images withheld; provenance statement added), matched signal-detection analysis script and frozen results aligned with Table 3, critic rerun defaults aligned with the frozen same-critic results, metadata (title, affiliation) aligned with the manuscript. v1.1 (2026-08-01): personal-data remediation, signal-absent / specified-task / critic-specificity experiments, stimulus integrity audit. v1.0 (2026-06-24): original deposit (now restricted).