Bold questions across the sciences
Honest answers, or none. Occam Research is the independent practice of Oleg Dolgikh — chasing decidable questions across disciplines and holding every claim to evidence. Peer-reviewed work, open data, and, when a claim deserves a fight, a public duel.
The capstone of the CRN program: a falsifiable account of how moderate disorder can sharpen selective signal routing on real connectomes — tested, published, and peer-reviewed. The interactive companion lives below.
Current research
Transport and computation
CRN studies how topology, disorder and readout shape transport on biological connectomes. The Frontiers article is peer-reviewed; supporting preprints and datasets are identified separately.
Journal articleStructural safety of learning
When can local physical learning preserve the conducting structure required by its own task? VCS gives sufficient certificates within a stated resistor-network model.
VCS preprint v0.2.1Learning and learnability
Recent work gives sufficient certificates for joint voltage training and preservation of learnability. Assumptions, admissible targets and limitations are stated explicitly.
SCI-006B preprint v0.3How the work goes
Scout
Range across fields for questions sharp enough to be decided — where an independent researcher can still land a real result.
Test
Attack the question with computation, real data, and explicit null models. No claim survives without evidence — and the nulls have to fail honestly.
Duel
Put the strongest version of a claim against its strongest objection, in the open, to a verdict. One of several formats — see Duellum Veritatis.
Duellum Veritatis
CRN — the founding trophy
Coherent Resonant Netting is, as an active program, closed. But it is the trophy that forged everything here: it set the method — falsifiable claims, real connectome data, honest null models — and it still shapes what I scout and how I fight. Peer-reviewed in Frontiers in Computational Neuroscience; below, the interactive companion across organisms.
The CRN hypothesis, in brief
Biological agents face an energy–information bottleneck: inference needs rapid exploration of large hypothesis spaces, yet high-gain spiking is metabolically expensive. CRN proposes a two-regime decision architecture. Stage-I (netting) — low-cost wave-like filtering via GKSL/Lindblad dynamics, with tunable dephasing κ and disorder ε on the structural connectome, concentrating probability on target hypotheses. Stage-II (fixation) — expensive spiking commitment that broadcasts the winner. By filtering before firing, the system cuts costly O(N) broadcast events to O(1). The testable signature is Disorder-Enhanced Selectivity: moderate disorder improves target selectivity on real connectomes and depends on native topology — degree-preserving rewiring destroys it, and classical random walks cannot reproduce it.
Historical CRN scorecard
This is the March 2026 working assessment, preserved for provenance. Its scores are internal assessments, not probabilities, independent peer-review ratings, or the current status of every research claim. Consult the dated publications for their stated findings and limitations.
Publications & Data
Recent preprints: Structural Safety · Joint Training and Learnability Preservation
Defensive Publications
Host-agnostic pre-delivery gate for conversational agents: after a backend computes a candidate response but before delivery, CRTC revalidates the durable inbound queue, suppresses stale answers, persists supersede/merge state, and redispatches a bounded merged turn.
Defensive disclosure archived on Zenodo. Release archive SHA256: effbd3af8be61fc78dfb073722b19a2eb955fe0984c0e202612e8b465c162309.
Open Resources
The researcher
Oleg Dolgikh
Oleg Dolgikh is an independent researcher working on computation, transport, and learning in networks. His research spans computational neuroscience, open-system models of biological connectomes, and mathematical guarantees for local physical learning. Through Occam Research, he shares research with explicit assumptions, reproducible materials, and clear limits on each claim.