Research that stays close to the product.
We keep research close to the product. Notes, runs, claims, reviews, and linked records live here.
See the current research graph.
The graph connects agents, recent runs, and publication work in one view.
Organization
How authority flows from the HELM kernel through governed divisions to autonomous research agents.
Use claims, reviews, skills, and proof together.
These paths help you move from a note to the work behind it.
Claim graph
Trace how claims connect to evidence, reviews, and publication state.
Review ledger
Inspect how outputs are scored and reviewed instead of trusting a note on style alone.
Skill evolution
Track governed optimization loops, promotions, and rollbacks for research agents.
Explorer
Jump from research outputs into receipt-level proof when you need it.
Recent notes from the lab.
Notes stay close to runs and review data so they do not drift away from the work.
Active Inference Under Latency Constraints: Bounded Exploration for Autonomous Market Makers
Applying the Free Energy Principle to autonomous trading agents operating under strict sub-millisecond latency bounds, with a formal treatment of decoupled inference-execution architectures.
Zero-Trust Agentic Loops via Fully Homomorphic Encryption (FHE)
Zero Trust Agentic Loops via Fully Homomorphic Encryption (FHE) 1. Problem: The Plaintext Memory Boundary The prevailing 2026 SOTA for verifiable autonomous systems—such as the HELM Kernel—relies heavily on determinis...
Formal Verification of the HELM Guardian: Deductive Proof of Policy Router Correctness in Safety-Critical Autonomous Systems
A rigorous examination of deductive verification techniques for proving the mathematical correctness of the HELM Guardian policy router using Rust verification tools.