Separate local identities and durable journals. Verification policy stays under the reviewer’s control.
Inside the engineering PROBLEM / APPROACH / DECISIONS
THE PROBLEM
A result alone cannot explain which task ran, who produced it, or whether an independent verifier can reproduce the claim after a restart.
THE APPROACH
Three local processes persist separate identities and journals. Signed tasks, evidence, and two verifier receipts form a canonical package that can be checked and replayed.
THE DECISIONS
Keep reviewer pins independent of the producer. Bind evidence to a specific operation. Preserve durable state across recovery. Reject tampering with executable checks.
CURRENT SCOPE
Fixed membership, trusted verifier code, and local transport. A passing task-bound check is not a protocol safety theorem or a production RChain finality claim.
Security / AIResearch workspace
Omni
Test how robust an investigation is when its evidence is challenged, removed, or replayed.
Public-source collection, competing hypotheses, evidence stress tests, and portable run receipts. Integrity verifies binding and consistency; it does not establish the truth of a source.
integrity mechanisms demonstrate binding/consistency, not truth
quantum-adjacent artifacts do not imply quantum advantage
RY/CX circuits, exact-reference comparisons, fresh random starts, and persistent search memory. A simulator for small circuits; hardware execution and quantum advantage are not claimed.
A Rust control plane and C++ route share an operation contract. Persisted Ed25519 identities sign execution receipts. Current admitted operations are echo, ASCII uppercase, and sha256.
Quorum and view-change logic are studied alongside persistence and selected formal invariants. Research artifacts are not presented as a production consensus guarantee.
formal artifacts must be scoped precisely
README-level mathematical claims are not promoted to registry facts without artifact-level verification
Uniform allocation remains the baseline after exploratory adaptive methods failed to establish a general advantage. Negative and neutral results are retained.
failed or neutral experiments remain recorded
experimental wins are not generalized beyond their measured conditions
I’m interested in the space between a system that runs and a system we can reason about.
My work combines practical software engineering with research across AI, distributed systems, quantum computing, formal methods, and security. I build the implementation, test its failure modes, and keep the limits of the result visible.
Some projects are usable tools. Others are experiments with unresolved questions. Both belong here, with enough context to tell them apart.
Private research index 9 names
Selected private project identities already published in this portfolio. Source code and internal records remain private.