Code and derived data for "Environmental variance and the evolution of aging: A bet-hedging model for the segregation-to-repair transition"

Code and derived numerical results for the manuscript "Environmental variance and the evolution of aging: A bet-hedging model for the segregation-to-repair transition" (Tanigawa, submitted to the Journal of Theoretical Biology). The paper treats the choice between segregating macromolecular damage at division and investing in somatic repair as a bet-hedging problem under environmental uncertainty, with long-run geometric-mean fitness as the criterion. It establishes when a transition between the two strategies exists, where it falls, and what fixes its order. Contents. 23 analysis scripts (22 in Python 3.12, one in R for the phylogenetic correction) plus 2 inherited figure scripts, so that no figure in the paper lacks its generating code; 37 derived result files (CSV/JSON) underlying every quantitative statement; the machine-checked lemmas (Lean 4.32.2 with Mathlib at revision 905b95818e); and a verification harness of 322 automated checks that regenerates every headline number and checks it against the manuscript source. 54 invariants and property tests (pytest + Hypothesis) run from the deposit alone. Third-party inputs. Two are redistributed here under their own licences so that the deposit is self-contained: the AnAge lifespan table (Human Ageing Genomic Resources, CC BY 3.0 Unported) and the node-dated mammal phylogeny of Upham et al. (2019), Dryad 10.5061/dryad.tb03d03 (CC0 1.0). One analysis — the domain-count comparison of Supplementary §S5.4, which is reported as inconclusive and supports no claim — needs seven proteomes that the README identifies exactly but does not redistribute, because of their size and distribution terms. Not included. The manuscript text and the published figures are not in this deposit (journal copyright). Every figure is regenerable from the deposited data using the included scripts. Licensing. Code is released under the MIT License; derived data under CC BY 4.0. The two bundled third-party files retain their own licences as stated above. Changes in v1.0.1. The model's time scales are now stated explicitly: time is measured in environmental episodes of exogenous duration, long relative to the relaxation of damage for every admissible strategy, and results are given in that quasi-static limit. Nothing in the computation changed — this states what the equations already described. The AR(1) sensitivity analysis and its script were removed, because in that limit the growth rate depends on the marginal distribution of the environment alone. One reported number was corrected: the transition for a cost quadratic in asymmetry is 0.534, not 0.481; the order-parameter curves, the Landau coefficients, β = 1/2, the critical variance 1.52 and the linear-cost transitions are unchanged. Four defects that prevented anyone but the author from running the verification harness were fixed. Changes in v1.0.2. Documentation and checking only. No analysis, no figure and no reported number changed; the code that produces the results is byte-identical to v1.0.1. Supplementary §S3 previously named the Lean toolchain and the Mathlib revision but not the Python and R stacks, which appeared only in the README. The versions are now stated in the paper, and the paper says what they are: values read from the interpreters and binaries that actually executed, not from a package listing. The verification harness gained a check that reads the versions from the running interpreter and confirms that each appears in the Supplementary Material, so the two cannot drift apart (315 checks in v1.0.1, 322 here). A full cell-by-cell audit of every table found one input value of the convexity decomposition of §S8.4 that was not registered as a parameter of the sweep, although its five siblings were. It is registered now, so every numeric cell of every table in the paper is either regenerated by the harness or listed with its provenance.