Stress-test adjacent-level boundaries with synthetic cases
Packaged know-how that tells an agent how to do a job well.
You might say…
“The boundary cases I didn't find at design time always showed up on a real client — and then I had to decide on the fly under pressure.”
What it does
Generates synthetic borderline subjects for a drafted rubric, scores each against the anchors, and surfaces cases where adjacent-level boundaries are ambiguous — recommending anchor-text fixes.
Trigger: Use after a first-draft rubric exists, to harden inter-rater consistency before packaging.
I/O: First-draft rubric with anchor descriptions → synthetic borderline cases with scores, identified ambiguous boundaries, and recommended anchor-text fixes
Recognise the problem?
The primitives are the commodity part. The fastest next step is a conversation about composing them into something that works for you.
Start a conversation