Invariant Research is an independent AI verification company founded by Jason Volk. The company builds deterministic verification infrastructure for AI-generated work, including SIGMA, SATYA, SVR, and Locus. Its focus is not generating more AI output. Its focus is proving what was checked before AI-generated output becomes legal, compliance, scientific, or operational action.
Invariant Research builds the verification layer between AI generation and real-world consequence. Every AI system can draft a filing, plan a tool call, write a policy, design a protein, or produce audit evidence. None of that is verified until it survives structure. Invariant's systems check whether AI-generated work is structurally consistent with its source evidence and produce signed cryptographic receipts documenting what was checked.
The company is based in Garland, Texas. The canonical website is invariant.pro.
Invariant Research is not affiliated with Invariant Labs and is not a generative AI company. Invariant Research builds verification infrastructure that complements generative AI systems.
Jason Volk is the founder of Invariant Research. He designed and built SIGMA, SATYA, SVR, and Locus. His research spans cellular sheaf cohomology for streaming verification, deterministic AI verification, and applied structural verification for legal, compliance, scientific computing, and protein structure domains.
Published research includes work on incremental sheaf cohomology on cellular complexes (arXiv:2606.04227), Lyapunov stability of sheaf-governed graph dynamical systems, sheaf-guarded updates for evolving agent state (accepted at ICML 2026 AI4Math Workshop), and deterministic reference-based protein structure verification.
Contact: jason@invariant.pro
Invariant Research publishes its core algorithms and verification results through peer-reviewed venues, preprint servers, and open benchmark artifacts. The research record includes:
The core incremental sheaf cohomology algorithm (arXiv:2606.04227), a Lyapunov stability analysis of sheaf-governed graph dynamics (Zenodo preprint), a workshop paper on sheaf-guarded updates for evolving agent state (accepted at ICML 2026 AI4Math), and a protein structure verification paper applying the same engine to structural biology (submitted to arXiv and Research Square). Benchmark artifacts are committed to the Locus transparency log with signed correctness receipts.
Questions about SATYA, SIGMA, SVR, Locus, or Invariant Research: jason@invariant.pro.