Answers are grounded in this paper and cite their sources.
Manuscript, code and confirmatory evidence for a fixed-carrier value decomposition comparing fixed inheritance, explicit ecological cues, finite experienced history and reward-trained recurrent history. In one fixed three-state carrier, a conditional policy using ecological descriptors captured 99.19% of the per-condition oracle advantage over fixed inheritance, and 97.96% on four combinations withheld during optimisation. Removing true descriptors but retaining an investigator-specified finite history state recovered 65.51% of the explicit-cue advantage; a 12-unit recurrent policy trained only through downstream lineage value recovered 72.06% on the full evaluation and 62.02% on withheld combinations. Resetting recurrent state every generation removed 104.48% of the learned advantage. The deposit contains the complete surviving run directories for confirmatory stages 3A through 6, including preregistrations, scripts, row-level evidence, model states and checksums, together with the manuscript, supplement, corrected evidence ledger and Monte Carlo uncertainty analysis. These are task-local computational value decompositions, not claims to have originated the general theory of phenotypic memory or cue integration.
No takes yet. Share an insight, caveat, or question.
Joel Pearcey (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: