Qualify lower-effort Grok managed writable lanes without weakening acceptance #688

The current managed writable Grok boundary recognizes only Grok46Xhigh and Grok47Xhigh. TypedRuntime::from_launch requires the exact xhigh effort; advertised lower efforts and an otherwise valid role_models entry do not supply writable authority. See the typed runtime contract.

The owner requested better token efficiency after excessive Grok4.7 usage. A narrowly scoped ordinary artifact task completed through native Grok4.6/low, but that receipt does not qualify MACO-contained writable execution. Meanwhile the existing mixed-provider test plan requested high, which cannot satisfy the typed contract. Prompts, model selection, and the effectful adapter need explicit agreement about the admitted effort lane; silently upgrading effort or treating catalog membership as effectful authority is inappropriate.

This is a technical capability-qualification gap, separate from the production economics pilot in #631, mixed executable bindings in #665, and 4.7 ACP confinement in #616. It does not authorize lowering review gates or additional paid comparison probes.

Acceptance:

• Qualify at least one useful lower-effort standard Grok worker lane using a substantive MACO-contained task and the selected account, exact model/effort, structured output, command journal, candidate path boundary and independent acceptance evidence.
• Bind requested and observed model/effort without silent xhigh upgrades or ambiguous aliases. Unknown or mismatched identities and unsupported model/protocol/effort combinations remain fail-closed.
• Preserve opaque launch proof for the rendered cwd/output/permission/native-delegation contract. Consider separating process containment proof from reasoning-effort identity rather than adding effort assumptions to unrelated safety checks; no specific refactor is required.
• Keep strong supervisor, independent review, acceptance and publication gates unchanged. A lower-effort worker cannot grant itself writable capability, accept its result or select a weaker reviewer.
• Cover negative cases for missing identity/usage, wrong schema, widened paths, native delegation, unsupported effort and protocol mismatch. Real selected-account proof is required before promoting an additional writable contract.
• Update operator documentation so configuration and rejection messages state which effort/protocol lanes are actually admitted. Keep cost/quality and Build quota-saving claims unqualified until Qualify a production model-mix default with real-provider evidence #631 has comparable measured evidence.

Current capacity limitation: a necessary Grok4.6 artifact request on 2026-09-28 returned HTTP402 Grok Build usage balance exhausted. No lower-effort paid comparison or alternative-account retry was launched after that failure. Capacity or a native leaf receipt is not qualification evidence.