OptionalaccuracyThe model's own accuracy in the merit's units, below which a predicted decrease is not one and the solve settles without trying the step (the termination of an SQP on its subproblem's accuracy, Nocedal and Wright chapter 18); never settles so when absent.
OptionalfailedWhether the subproblem's solver could not solve the model, after its own recovery: no step is proposed, and the solve stops failed (Reason), as TrajOpt's does where its convex solver fails (sco's OPT_FAILED); a step otherwise.
Whether the step reaches the region's boundary, where a schedule that grows only there (ISchedule.boundary) grows.
The step's largest coordinate, in the region's units.
OptionalmetWhether the step meets the linearised constraints to the caller's tolerance, the steering's linear feasibility (IOptions.steer).
The model's decrease of the merit from the iterate to the step, m(0) - m(p).
The iterate reached, projected onto the caller's hard set; may be lazy and select its proposal-local correction metadata. Steering reads only diagnostics; the outer loop reads the chosen trial before its evaluation and correction.
OptionalviolationThe violation of the linearised constraints at the step, in the units the penalty weighs, the model's own; the steering reads it (IOptions.steer).
The convex model's step within a region.