claude
beb093aebb
Two commands replace work that 66 sessions of transcripts show being redone by hand. `make t` replaces the CGO preamble, pasted 391 times across past sessions and documented in CLAUDE.md as the way to do it. It also sets MAVEN_ONNX_LIB, which that recipe did not: the four TestONNX* measurements self-skip without it and the run still prints "ok", so every targeted eval done the old way reported the hash ratchet while reading as a real embedder score. -race keeps it honest against `make test`, -count=1 keeps a stale cache from passing as a result. `make audit` replaces the inventory sweep. The four longest sessions spent 93 greps rebuilding it before their first edit. Runs in 0.75s. Its stub search is narrower than the sweeps were, on purpose. "not wired" is this repo's word for a nil dependency and matched ~30 comments describing working code; "placeholder" names real identifiers and matched 16 more; internal/ipc/unimplemented.go is the deliberate Unimplemented*Server pattern, not 60 gaps. A gap report that reports the architecture back at you is one nobody reads twice.