Environmental recurrence selects a costly memory-use policy across two operator bundles without a resolved one-bit population benefit

This research preprint and reproducibility package reports three prospectively specified abstract computational studies of a supplied one-record private cache with a fixed one-proposal opportunity cost. Across the original 32-individual model and an independently implemented 32-locus torus/tournament operator bundle, lag-two recurrence selectively enriched the cache-use policy allele. Study 002 further showed that directional cache information amplified selection relative to a recurrence-gated, parent-distance-matched noninformative control. Population-accuracy effects remained bounded below the prespecified positive one-bit scale or unresolved, supporting a selection/performance separation at each study's declared scale. The package retains adverse and unresolved outcomes, original execution failures, and the Study 003 audit-header recovery: a derived audit tree normalized exactly 12 non-payload header bytes while preserving every payload byte, after which the independent audit passed with zero mismatches. The cache architecture was supplied; retention and retrieval were coupled; one HALF law aligned to delay and fixed costs/horizons were tested. No universal-replication, component-attribution, gradual-search, biological, equilibrium, invasion, general memory-origin, or priority claim is made. This preprint has not undergone external peer review. Manuscript, figures, and data are CC BY 4.0; original code is MIT. Repository and verified release: https://github.com/jackchenx3/recurrence-selects-costly-memory-use/releases/tag/v1.2.0