Research Archives

We identify and characterize a structural isomorphism operating across biological, computational, and economic systems: the topology of local dependency with cumulative global effect. Systems in disparate domains share a common formal pattern in which locally constrained interactions produce emergent global properties without centralized coordination. We introduce the LDT Fingerprint and apply it to eight independent cases across four domains, with all values sourced from published literature.

We propose and formalize a structural isomorphism between ethical principles in AI governance and loop invariants in program verification. We demonstrate that ethical requirements such as non-discrimination, transparency, and accountability can be mapped to state-space constraints formally equivalent to the invariants used in software verification. We formalize this using Linear Temporal Logic (LTL) and model checking.

We present an integrated methodology for autonomous epistemic research comprising three interdependent components: a multi-actor deliberation framework, a sixteen-principle ethical constitution that binds the creator rather than the systems themselves, and a formalized correction methodology in which the primary mechanism of knowledge generation is mutual correction rather than agreement.

When multiple autonomous systems operate under independent constitutions, a structural governance gap emerges. We present a meta-constitutional layer of five principles that operate above all domain constitutions without overriding them: irreversibility protection, uncertainty escalation, full auditability, proportional autonomy, and systemic harm prevention.

We argue that verified negative isomorphisms are first-order epistemic results that delimit the space of possible knowledge. We present three verified negatives produced by an autonomous deliberation system, all achieving Grade A. We analyze their relationship to the broader deliberation corpus: of 307 sessions, 62 produced Grade D verdicts functioning as candidate negatives. We include an analysis of RLHF as a structural category error.

We present empirical evidence from 5,351 cross-domain epistemic correlations across 64 domain pairs, derived from over 360,000 scientific documents, demonstrating that cross-domain fertility follows a highly non-uniform distribution. The most fertile pair produces over 1,400 times more connections than pairs at the tail. We present evidence that fertility and epistemic quality are partially independent, with Materials Science functioning as a structural connector domain.

We present evidence from 307 deliberation sessions and 66 evaluated theses showing that the surprise score assigned at detection does not predict the quality grade assigned after deliberation. The A+B rate is stable across surprise bands (18.4–23.3%). At maximal surprise (1.0), the most common outcome is speculation, not breakthrough. Systems that conflate interestingness with quality produce noise; systems that evaluate quality independently produce knowledge.

We present an alternative to capability-first development in which constitutional governance is implemented before capabilities are deployed. Drawing on 307 deliberation sessions with a 20.2% failure rate demonstrating discriminative power, 3,697 decision narratives referencing constitutional provisions, and zero irreversible decisions without human confirmation, we demonstrate that constitutional governance functions as test-driven development for decision quality.

Of 5,351 detected cross-domain correlations, approximately 1.1% reach near-maturity. We present the maturity funnel with verified data (SEED 32 → SPROUT 5,155 → BRANCH 109 → NEAR-MATURE 55), compare attrition rates to drug discovery, venture capital, and academic peer review, and argue that the maturity gradient is a structural property of epistemic systems, not a deficiency to be optimized away.

We present a systematic identification of epistemic deserts—regions where verified connections between disciplines are absent or near-absent—derived from over 360,000 documents across a 302-subfield taxonomy. Using 5,351 correlations across 64 domain pairs, we identify 14 singleton pairs, classify them into three types, propose four criteria for prioritizing exploration, and argue that the desert map is a research planning instrument.

Six independent epistemic assessments at the highest confidence level converge on a single finding: no formal isomorphism exists between the mathematical structures of artificial intelligence and those of financial markets. We examine why the assumed structural correspondence fails, identify three functional analogies that remain operationally valid under stable conditions, and demonstrate that these analogies break precisely during tail events, liquidity crises, and correlated failures. We discuss implications for algorithmic trading, regulatory design under the EU AI Act, and systemic risk assessment.

The rate of direct, functionally operational transfer of AI techniques between terrestrial and planetary domains approaches zero. Using the case of Earth-based computer vision applied to Jupiter’s magnetosphere reconstruction with Juno mission data, we introduce the concept of epistemic rarity (r=0.99) as a methodological metric—not statistical—for quantifying transfer barriers. We identify three axes of assumption divergence (continuity vs. discretization, determinism vs. uncertainty, bounded vs. unbounded parameter spaces) and argue that these barriers are structural, not technical, raising broader questions about unexamined AI transfer failures across adjacent domains.