About the institute
ANULUM Institute is an independent research and technology practice in Marbach SG, Switzerland, run by Miroslav Šotek. It works on fusion reactor systems and their control, quantum phase dynamics, stochastic and neuromorphic computing, and the coordination and verification of AI agents, and it publishes the software and the papers that come out of that work under open licences.
Operator
- Operator
- Miroslav Šotek · ORCID 0009-0009-3560-0851
- Address
- Obergasse 8, 9437 Marbach SG, Switzerland
- Legal form
- Sole proprietorship; commercial-register entry pending, a GmbH is planned (see the Impressum)
- Contact
- protoscience@anulum.li · all channels
- Code
- github.com/anulum
Research groups
The portfolio registry lists 39 projects, 33 of them public, in five research groups.
- Agentic Coordination, Assurance and Continuity Systems · coordination, verification and continuity infrastructure for AI agents and their outputs
- SC Neuromorphic Computing Systems · stochastic and spiking computation, from simulation to hardware-oriented compilation
- SCPN Quantum Computing Systems · synchronised phase dynamics on quantum hardware
- SCPN Reactor Systems · the solver laboratory, a shared kernel library and the device-family cores of the fusion programme
- SCPN Systems Integration and Control · control engines and the coherence-control compiler that connect the physics to actuation
Method
Every project carries a capability inventory that names what is implemented and an evidence maturity that says how far it has been shown to work; the site reads those files rather than restating them, and a checker compares the two daily. A device core at computational-prototype maturity is called that. Solvers run behind fail-closed gates: a capability that cannot be validated refuses rather than approximates. Claims in manuscripts are classified as measured, derived, projected, literature-supported or a limitation, and each is bound to a command, dataset, commit or primary source before a PDF is accepted; negative and null results stay in.
Publishing
Work is published first as a Zenodo record with a DataCite DOI, then submitted to a journal with the preprint disclosed. The publications hub lists every record with its licence and status, and the review drafts that exist in repositories are marked as drafts. Software is released under AGPL-3.0-or-later or the licence its repository states, with a commercial licence available.
What this site does not claim
No reactor page describes, approximates or validates a real machine unless a filed source prints the number. No preprint is called a peer-reviewed article. No readiness, yield, gain or breakeven statement is made by the device cores. Where a figure is an upper bound, a reconstruction or a value the source does not give back, the page says so in those words.
Working with the institute
Consulting covers AI training, custom software, security, in-house AI deployment and GPU clusters. Peer review, experimental collaboration and patronage go through the contact channels.