pyliferisk geometric valuations discard supplied mortality assumptions

Independent numerical audit of pyliferisk 1.12.0. Geometric annuity and insurance helpers reconstruct the original mortality table and silently reset the supplied mortality scale to 100%. Custom survivor/probability tables may instead fail. For GKM95, mortality 75%, interest 3% and age 60, a zero-growth annual annuity returns 13.616328838004055 instead of 14.93597449181164, approximately 8.8353% lower.

The proposed correction preserves copies of the effective survivor and mortality-probability tables while changing interest. Seven helpers were checked against 630 independent direct cash-flow sums at 60-digit Decimal precision: 498 failures before, including 198 exceptions, and zero after. Four regression tests pass. Substituting the released code causes 82 failing and 42 errored subcases. Existing growth timing and fractional-payment formulas are preserved; the entire library is not claimed correct.

Official PyPI wheel source is byte-identical to upstream master 5c28ca34a30f4e350f604f4b35bf738c816e59bc. Public duplicate screening found no matching report among 14 available issues/PRs and 8 comments, and three scoped keyword searches; this does not establish worldwide priority. Correction: https://github.com/franciscogarate/pyliferisk/pull/17 (proposed, not merged).

Report, source snapshots, patch, Decimal oracle and test evidence: https://github.com/kadyrbekovhamit-cyber/gero-pyliferisk-geometric-mortality-audit

All calculations ran sequentially on CPU using the Python standard library, without GPU. Performance was not benchmarked. Synthetic library-level cases only; no deployed insurer exposure or customer loss is claimed. Research prepared with AI assistance. Original code and tables retain GPL-3.0-or-later licensing.

Archive SHA-256: 643e2016557cb418422277de77d0e20962272b62addbcdf3011c1771154a5d35