An evaluation scaffold for automated research workflows: scenario design, trace capture, failure review, and reproducible comparisons between agents.
Agentic tooling can look impressive while hiding the hard questions: observed state, changed files, failures, and reproducibility status.
The harness treats automated work as an experiment. It records scenarios, tool calls, outputs, failures, and review records so comparisons are inspectable rather than anecdotal.
Evidence trails expose quality, regressions, and failure modes across runs.
A testbed for studying size, fragmentation, resolver behavior, and protocol boundaries when DNSSEC meets post-quantum cryptographic material.
Lean-oriented formalization work for reasoning about stack behavior, invariants, and the gap between executable systems and proof-carrying claims.
A measurement-oriented harness for tracking post-quantum TLS behavior, handshake shape, certificate chains, and interoperability risks across real deployments.