Human-AI Collaborative
Scientific Architecture

A formalization of the M·A·H·A (Mindfulness, Authenticity, Health, Action) framework as an integrative socio-biological model. This paper proposes that contemporary metabolic, attentional, and relational pathologies may be productively understood not as isolated public health failures, but as coupled downstream consequences of a single, structurally extractive economic architecture.

A Monte Carlo forecast (N = 100,000) of where, when, and by which instrument the hypothesized Planet Nine would most plausibly be detected — propagating a weighted three-model parameter ensemble through Kepler orbit simulation, applying existing survey masks, and modeling LSST, Subaru/HSC, and infrared detection through 2036. The planet is treated as an unproven hypothesis; the analysis quantifies detectability and a null-detection posterior rather than asserting existence.

A replication of the extreme trans-Neptunian object orbital-clustering test on the fully characterized CFEPS/OSSOS sample, using an independent Python selection-function pipeline benchmarked against the official Fortran SurveySimulator. The reconstructed sample shows no statistically significant clustering — robust to implementation choice and directionally consistent with Napier et al. (2021). Includes a composition-agnostic hypothesis matrix, a candidate vetting protocol, and an empirical self-audit of the agentic AI instrument.

A study implementing a grammar-constrained symbolic regression framework to systematically discover optimal local plasticity rules governing the readout weights of an empirical neural population. The framework systematically discovered homeostatic Hebbian rules that out-perform prior hand-crafted baselines in tracking representational drift.

A leakage-audited test of whether a neural network can predict the Betti numbers of candidate G2 manifolds from real Kreuzer-Skarke-derived Calabi-Yau topology and a Z2 involution encoding, benchmarked against a linear baseline. The network wins on the non-trivial target and loses on the linear-dominated one — the mixed result is reported in full. Includes a self-audit of a prior version whose circular result and synthetic data were removed in revision. The smooth G2 resolution is assumed, not constructed.

A verified map of the contested de Sitter problem in the string/M-theory swampland program, structured around seven open problems and their competing camps — KKLT control, the Large Volume Scenario, the de Sitter Swampland Conjecture, the Dark Dimension, quintessence, Dine-Seiberg, and Trans-Planckian Censorship. Each camp's strongest arguments are represented without adjudication, and citations carry provenance tags marking which identifiers were independently resolved. A synthesis and orientation tool, not original research.

A reproducible active point-mass N-body pipeline and numerical convergence framework for studying highly inclined retrograde perturbers, built on REBOUND's ias15 solver rather than phase-averaged secular approximations, with a coupled infrared detectability parameterization. The 20,000-year proof-of-concept run is explicitly an integrator-stability and boundary-validation phase — roughly 1.55 perturber revolutions, far short of the secular shepherding timescale — and reaches no shepherding conclusion by construction.