research ledger
Notes from the pipeline
Incident write-ups, benchmark methodology, and data notes — some written, some generated by the verification pipeline itself. Every claim traces to a ledger with a named runner and run.
what OVM is and how it works
The tool, the five-minute watcher, the Mac mini verification lane, and the gate that keeps broken releases out of the registry.
the canary that catches what --version can't
Codex 0.144.0 passed every traditional check and couldn't execute a command. The incident and the three-layer canary built from it.
how the performance index is calculated
Median-normalized launch, spawn, and response metrics averaged into one lower-is-faster score — formula, rationale, caveats.