Research preprint v1.4.0

Research preprint v1.4.0

This revision integrates the fixed observation-error challenge (060) into the existing private-memory paper. The primary noise-by-recurrence founder interaction remains unresolved: +1.635742 [-2.604370, +5.655924] percentage points. Exact and noisy founder recurrence effects remain positive above one expected descendant. Noise reduces absolute mean true population accuracy in both arms under both future laws; terminal effects remain unresolved. Founder transmission and population utility remain separate outcomes.

The release retains the unresolved pulse primary, frequent founder loss, all adverse outcomes, one prespecified noise level and the reuse of exact-observation 057 preparations. It does not establish noise equivalence or independent training replication.

Includes 17 figures and 15,284 saved-table estimates through 060, compact evidence and scientific source, 207 registered manuscript statistics, and preserved audit/portability records. Manuscript/data: CC BY 4.0; original code: MIT. Not externally peer reviewed; AI assistance disclosed. Large raw trajectories and random tapes are excluded. No new scientific experiment was run for this publication.

Version DOI: https://doi.org/10.5281/zenodo.22949032
Previous v1.3.0 DOI and all prior tags remain unchanged: https://doi.org/10.5281/zenodo.22939403