CLIProxyAPI · six candidate lanes · visual course

Measure the proxy. Isolate the harness.

Eight lessons for reviewing CLIProxyAPI as the general layer. This revision executed zero inference; preflight, canaries, and high volume remain blocked.

16 IDs6 text lanes53 strict + 9 conditional / lane0 inference · BLOCKED

Boundary and attribution

Lesson 1Start with CLIProxyAPISeparate proxy capability from what the sidecar, harness, and UI add to the path.Lesson 2Qualify six lanesA registered endpoint does not prove every model works through every protocol or harness.

Comparable measurement

Lesson 3Separate network from experienceTTFT, TTFE, TPS, and latency compare only when they start and end at the same boundary.Lesson 4Deliver the context you claimThe nominal target changes when the cap, envelope, and output reserve share one window.

Functional behavior

Lesson 5Test tools and thinking by contractA valid tool call and a visible reasoning block answer different questions.Lesson 6Measure each lane's real continuationResending, remote state, local resume, and compaction are different forms of memory.

Control and decision

Lesson 7Preflight errors, price, and failoverNo deliberate monetary ceiling still leaves interpretability, safety, and identity gates.Lesson 8Stop at the current gateThis revision executed no inference; preflight, canaries, and high volume remain blocked.