01 · The idea
What stayed with us
A compact call-only model combines top-five tool retrieval, constrained decoding, confidence escalation, bounded context, quantization, and a small packaged Mac artifact.
02 · The local translation
What we did with it
A 94-case local smoke reached 34.0 percent exact tool selection. Oracle catalog restriction reached 38.3 percent overall but regressed Pace and retained false actions.
03 · The boundary
Where the comparison stops
Catalog restriction is a latency lever, not a capability rescue. Package size and narrow call formatting do not establish grounded tool competence.