BIOISO presents a biologically-inspired framework for computational entities that adapt their own algorithmic structure across generations. A BIOISO entity carries a. loom specification as its "genome, " promotes surviving mutations into the next compiled binary through a meiosis mechanism, and bounds the scope of structural change within a generation through a telomere lifecycle counter. The paper argues for a five-tier ceiling hierarchy of optimization complexity (T1–T5) — fixed-rule dispatch, stochastic search, hyper-heuristic selection, surrogate-model optimization, and structural self-modification via meiosis — with each tier adding exactly one primitive the tier below cannot express. The T5 primitive is the first mechanism that operates on the space of algorithms rather than the space of parameter values, operator selections, or surrogate model weights. The paper validates the T5 structural primitive empirically in two controlled experiments. The COCO/BBOB benchmark suite (30 trials × 4 functions × 2 conditions, DIM=10) shows T5 producing a 10× median normalised-fitness reduction on the f2 ill-conditioned ellipsoid against a Halton-T4 baseline, with the benchmark's scope limitation (the T4 stage is Halton-approximated rather than GP-UCB) made explicit. The AEGIS delta-neutral DeFi experiment (10 trials × 5 market regimes × 2 conditions) demonstrates inter-generational meiosis: at StrongBull regime boundaries T5 produces a per-epoch Sharpe advantage of +0. 517 (10/10 correct topology switches), but the net 5-epoch cumulative Sharpe of −0. 024 — a small loss, not a gain — is reported honestly, driven by parameter re-convergence cost (−0. 339) at the return-Ranging boundary plus a MildBear false-positive (−0. 285). Per (StrongBull + return-Ranging) cycle the additive contribution is +0. 178 Sharpe; compounding cumulative advantage is projected over multi-cycle backtests but is not demonstrated within the 5-epoch window. The paper also explicitly acknowledges that the AEGIS validation is closed-loop under an analytical Sharpe model, not open-loop against live market execution. The framework seeds ten domains in §5: one T5 domain empirically validated (aegisdeltaₙeutral, §4. 6), seven T5 domains theoretically motivated (§5. 1–§5. 7, covering AMR coevolution, HFT flash crash, JIT compilation, drug resistance, ICS zero-day defense, quantum error mitigation, climate intervention), and two calibration domains (§5. 9 biosphere T4, §5. 10 oceancirculation T3) included to demonstrate the framework does not default to T5 universally. This is preprint v1. The title names the research program; the operational model and initial empirical evidence are what this iteration delivers. The formal autopoietic isomorphism with Maturana–Varela autopoiesis — testing whether BIOISO's (G, T, M, Ω) structure satisfies the criteria of organizational closure, self-production of components, boundary maintenance, and substrate-independence — is the subject of a further companion paper currently in preparation with a biological-systems researcher; this paper does not undertake that demonstration. The full scope of what this paper claims vs. defers is documented explicitly in §1. 1 (Scope of claims) and §3. 5 (Relation to autopoiesis). The companion essay "Onwards: The Formal Tradition Was Waiting for Its Executor" (submitted to ACM SIGPLAN Onward! 2026) carries the broader theoretical framing, including the Nous/Logos formulation, the GS six-tier obligation cascade (T1 Development through T6 Meta-telos with the harness as a cross-cutting capability and the judgment layer as irreducible human work), the biological isomorphism argument as structural analogy, and the directed-formal-autopoiesis category. Repository (full implementation, test suite, experiment evidence, lineage): https: //github. com/jghiringhelli/loom Paper source: docs/publish/bioiso-paper. md in the repository above. Companion theoretical framework: Ghiringhelli, J. C. (2026). Generative Specification: A Pragmatic Programming Paradigm for the Stateless Reader. Zenodo. https: //doi. org/10. 5281/zenodo. 19637142
Juan Ghiringhelli (Thu,) studied this question.