◇ Could this help me?
Yes for the technique, with the framing corrected. Supplying explicit success criteria and having a separate grading pass check the work against them is a real and well-documented practice, and it is genuinely better than accepting a confident first draft, because it forces you to say what good means before seeing the output rather than reacting to whatever arrives. Two corrections to the pitch: this is not undocumented or unnoticed, it is a standard evaluation pattern that appears in provider documentation and has been widely discussed, and more importantly a grader built on the same model shares the author's blind spots, so a loop that runs until it passes can converge on work that satisfies the rubric while missing what you actually wanted. That makes it a strong quality filter rather than a substitute for reading the result, which is the part the walk-away framing gets wrong. General practice.
Before you start this task, help me write a grading rubric for it. Ask me what done actually looks like, and turn my answers into specific, checkable criteria rather than vague qualities, so that the rubric is capable of failing work rather than passing anything. Then do the task. Then grade your own output against each criterion separately, quoting the part of the work that satisfies it or explaining why it does not, and revise anything that fails. When it passes, tell me explicitly which criteria you are least confident about and where the rubric might be satisfied while missing what I actually wanted, so I know where to look myself.