On September 25, 2026, four lanes executed 69 delegated trials through the onex delegate on the lab dev lane. A trial passed only when a change-control receipt matched every definition-of-done item on its Linear ticket. Nine trials passed, representing 13 percent. Twenty-three more produced output that passed its check but carried no receipt. Thirty-seven failed their check or produced no usable output. Every accepted answer came from Qwen3.8-27B on the local tier at zero cost. Current capabilities include PR bodies, CI log verdicts, and Pydantic models from a spec. Code review and running a test suite do not work yet, with zero successes out of six and five attempts respectively. Everything else works only with caveats. The biggest single loss is output that was right but never reached a receipt.
