Distinct criteria
Each criterion rewards a different observable dimension: no double-counting the same behavior.
A rubric exists to make grading consistent and expectations visible. It fails at both when its level descriptions are adjectives in a trench coat: "good organization", "very good organization": because no two graders draw the line between adjectives in the same place, and no learner can act on them.
The fix is behavioral language. "Ideas follow a stated structure, and each section links to the next" describes something a grader can observe and a learner can do; "well organized" describes a feeling. The generator is instructed to write every cell at that standard, and to make adjacent levels differ in observable ways: what is present, what is missing, how often.
Each criterion must measure a separate behavior. If communication and presentation style both reward clear speech, the rubric scores one behavior twice. The generator creates distinct criteria, and the weight column shows the priority of each one.
Treat the generated rubric as a draft. Compare the top level with work that meets your standard, then make each description precise. The generator provides distinct criteria, separate levels, and explicit weights.