Safe measurement plan · State and measurement

Freeze the real surface

Learn why the ranking must measure the path Warp uses, not a convenient shortcut.

Primary local sources: plan, model matrix, per-cell plan, manifest and sanitized inventory.

The idea in plain language

The same model may look faster when the sidecar is skipped. To measure the Warp experience, first identify the active provider. If the local file configuration is active, port 8321 is the primary lane; if the persisted HTTPS entry is active, it becomes primary. Port 8317 remains a paired diagnostic only.

Think of this measurement as a sealed scale: if you swap the scale or move the object midway, the number loses meaning. The analogy breaks because gateways can also transform the request, not merely weigh it.

The freeze repeats catalog, version and hashes and records the active surface. Any ambiguity or drift stops the run. The sanitized inventory proved the same 16 IDs on 8321, 8317 and the redacted HTTPS entry, but the active surface must still be reconfirmed in the benchmark window.

Verifiable excerpt

GET /v1/models → HTTP 200\n16 IDs · hash 9886714a…8286\nprimary = ACTIVE_WARP_PROVIDER_TO_BE_FROZEN

outputs/BENCHMARK-MEASUREMENT-PLAN.md · outputs/benchmark-cell-plan.csv · outputs/planned-run-manifest.json

The path in one picture

Warp → sidecar → CLIProxyAPI → provider → proof gate

Retrieval check

Which lane defines the benchmark headline?