Safe measurement plan · State and measurement

Read the matrix without promoting a hypothesis

Separate 53 strict cells, 9 conditional cells, unsupported scenarios and the image route.

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

The idea in plain language

Each cell answers one narrow question: can this model receive this input size? ✓ means eligible under a known limit; △ requires a preflight; — exceeds the window; N/A means the text comparison does not apply. The 62 cells are candidates, not 62 authorizations.

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 plan covers 256, 128k, 256k, 500k and ~1M. DeepSeek does not enter 256k; GPT-5.5, GPT-5.6 and Kimi 2.7 do not enter 500k/1M. Qwen uses ~981k, not a literal 1M. Gemini 3.6 has five conditional cells and null targets until the alias is proven.

Verifiable excerpt

strict_eligible_cells = 53\nconditional_cells = 9\ncandidate_cells_after_all_preflights = 62

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

May a △ cell run immediately?