Good and bad rubric, side by side
This task is made up for practice. It is not from a real task.
The made-up task
A bad rubric, and what is wrong with each row
A good rubric
Why it works: each row checks one thing; every user limit has its own critical; the penalties catch two different tempting mistakes and do not mirror any critical; there is no penalty for installing flask-cors, because new_1 already covers it (that would be an inverse pair); penalties need no image because the grader judges them from the answer alone; the bonus stays inside the user's question.
Outcome Revised. Rationale: "Split c1 into the diagnosis (c1) and the fix (c2) and removed the flask-cors requirement because the user cannot install packages; added new_1 and new_2 for the user's two limits; rewrote p1 as a separate wrong fix; removed c3, c5 and c6 as vague or unrequested; narrowed the bonus to the request."
Outcome Revised. Rationale: "Improved the rubric and fixed some criteria."
Why it works: a reviewer can see every change and the reason, including deletions.
Why it fails: it says nothing a reviewer can check.