Predictive Self-Maintenance Paper 3: Tier A Prospective Maintenance Action-Value Reproducibility Package — Dataset A and AJF009

This record contains the reproducibility materials supporting the Tier A empirical analyses in the manuscript Predictive Information and Cellular Maintenance: Prospective Action Value in Single-Cell Lineages. Tier A tests whether measured pre-action biological information has prospective maintenance action value to an externally specified action policy. It is evidence relevant to Predictive Self-Maintenance (PSM), but is distinct from endogenous cellular use of future-relevant information. Stronger endogenous PSM requires evidence that the living system itself accesses and causally uses future-relevant information. Version 3.0 expands the archive to include two complementary Tier A analyses. Dataset A. The original reproducibility package contains the canonical 560-lineage × six-action analysis table, lineage and action summaries, independently reproduced scalar-state held-out results, frozen audit records for the richer transcriptomic-state analysis, figure source data, programmatic figures, executable scalar-state reproduction code, provenance information and checksums. The scalar-state analysis reproduced action-only 5/560 versus state-informed 11/560 detections, ΔV = 0.01071429, paired p = 0.17956543, with 95% lineage-bootstrap CI [−0.00178571, 0.02500000]. The frozen positive-evidence gate was not passed. The richer transcriptomic-state analysis likewise did not pass its frozen positive-evidence gate. AJF009 sequential-treatment validation. Version 3.0 adds a frozen follow-on analysis containing 10,851 lineage-history blocks from 10,122 unique lineages and 32,553 lineage-action observations. Five-fold grouped held-out validation produced an aggregate history-aware prospective action-value increment of +0.26897, with positive increments in all five folds. The history × subsequent-action increment was +0.07063. A prespecified 1,999-permutation test produced zero exceedances, corresponding to an empirical one-sided p ≈ 0.0005 using the +1 correction. The history-informed external policy produced an aggregate held-out gain of +0.185065 and selected a different subsequent action for 79.31% of evaluated histories. The AJF009 supplement also records sensitivity analyses using RPM-normalized abundance, removal of six spike-in controls and expansion to the cDNA-supported lineage population. These analyses preserved positive predictive and interaction increments. AJF009 provides substantially stronger Tier A evidence than Dataset A because its sequential design supplies richer history–action–outcome support. It does not establish endogenous cellular action selection: the subsequent actions were experimentally imposed and the policy analysis is externally specified. The AJF009 supplement transparently records a forensic limitation. The original generated Step 1H and Step 1I package binaries were not recovered from the surviving project archive. Their recorded SHA-256 hashes survive in the frozen project record. The present supplement therefore preserves the surviving frozen method specification, numerical evidence, robustness results, provenance and executable consistency checks without fabricating unrecovered permutation vectors or out-of-fold prediction tables. The underlying biological data are third-party public data and are not redistributed in this archive. Source-study and repository identifiers are provided in the reproducibility materials.