The Evidence Layer for Scientific AI

Physics & Climate

The Wavefunction: Mathematics or Physical Reality?

Quantum theory clearly uses the wavefunction to predict experimental probabilities, while its physical ontology remains interpretation-dependent.

Dark Energy and Accelerated Expansion

Accelerated expansion is strongly measured, but the physical mechanism remains unknown and current evidence still favors ΛCDM.

Why Temperature Inversions Suppress Rising Air

An inversion places warmer air above colder surface air, creating stable buoyancy conditions rather than a warm layer trapped beneath denser air.

Biology & Medicine

Urolithin A and Human Longevity

Cellular and animal findings make a healthy-aging effect biologically plausible, but human lifespan extension remains unproven without long-term randomized evidence.

Molecular Mechanism of AFB1 Genotoxicity

AFB1 becomes genotoxic after metabolic activation to an epoxide that forms persistent DNA lesions, with mutation risk shaped by repair and replication context.

Phenotypic Clusters in Small Fiber Neuropathy

Distinct clusters may reflect distal axon injury, sensory-neuron dysfunction, or altered pain processing, and loss and gain phenotypes can coexist.

Oligodendroglia and Multiple Sclerosis Risk Variants

Multimodal preprint evidence links selected MS loci to oligodendroglial regulation and phenotypes, but endogenous allele causality and in-vivo relevance remain unresolved.

MACP2, Reactive Oxygen Species, and Plant Autophagy

Arabidopsis genetics and imaging support MACP2 involvement in starvation autophagy and redox control, while a direct pore-and-peroxide transport mechanism is unproven.

Broiler intestinal transporter illustration

Nutrient Transporters in Broiler Chickens

Transporter transcripts offer useful biological clues, but expression alone does not establish protein abundance, nutrient flux, or improved growth.

Policosanol, Red Yeast Rice, and Astaxanthin

A small single-sex rabbit study reported reduced vascular lipid staining, but surrogate outcomes and sponsor-linked design limit translation to human cardiovascular benefit.

Methods, AI & Technology

10x Visium Compared with Xenium

Visium offers broad transcript discovery from spatial spots, whereas Xenium offers targeted single-cell and subcellular localization, making direct matrix comparisons potentially misleading.

Dynamic Reweighting, Diffusion, and Kramers–Moyal Estimates

Reweighted conditional moments can estimate target diffusion only when weights model observation or selection processes rather than merely reflecting sampling dynamics.

Neural Classification for Barrier-on-Chip Monitoring

ECIS plus neural classification shows promise as a chip-specific maturity gate, but small datasets, domain shift, and limited uncertainty calibration restrict broad automation claims.

Cross-Subreservoir Entanglement for EEG Forecasting

A single entangling gate is a plausible hypothesis, but its benefit cannot be estimated until held-out performance is compared against quadratic readout augmentation.

Persistent Manifold Learning of Protein Properties

The mathematical framework is distinctive, but current empirical gains do not yet show whether persistence itself outperforms ordinary multiscale geometry and sequence embeddings.

Large Language Models for Malaria Drug Discovery

TxGemma-9B performs strongly across global ranking while LlaSMol-Mistral concentrates hits near the top, and retrospective benchmarks do not establish prospective drug value.

Ribosome Collision Screening from rRNA Fragments

dricARF offers a potentially useful discovery screen from standard Ribo-seq, but RNase and library effects mean it is not yet a direct quantitative collision assay.

BioFirewall for Design-Stage Biosecurity

The prototype makes traceability and escalation testable in software, while computational feasibility does not yet demonstrate complete hazard detection or deployed biosecurity effectiveness.