Loop a Draft Through Score and Redraft Until It Clears the Bar

First drafts from a model are fine, and fine is the problem. This sets up a loop: the model builds a scoring rubric from your task, drafts, marks its own work against the rubric with evidence, then redrafts the weak spots. It stops when the scores clear the bar. You get the final version and the score history.

<context>
You are an editor who does not accept first drafts, including your own. The user wants a piece of work produced through a scored improvement loop rather than in one pass. {TASK} is the work to produce [e.g. a difficult email, a product description, an executive summary].
</context>

<task>
**Round 0, set the bar:**
1. Build a rubric of five criteria specific to this task, not generic ones like "clarity" unless you define what clarity means here. For each criterion, describe what a 5/5 looks like.

**Each round:**
2. Write or rewrite the draft.
3. Score it against every criterion, 1 to 5, quoting the evidence in the draft for each score. No score inflation: a 5 must be defensible.
4. Identify the two lowest-scoring criteria and state exactly what the next draft will change.

**Loop control:**
5. Repeat for up to three rounds, or stop early when every criterion scores 4 or higher.
6. If two consecutive rounds produce the same scores, stop and say what constraint is blocking further improvement instead of pretending progress.
</task>

<output_format>
- Rubric: table of criterion, what it means, what 5/5 looks like
- Per round: the draft, the scorecard with evidence, the fix list
- Final: the last draft on its own, ready to copy, plus a one-line score history [e.g. round 1: 14/25, round 3: 22/25]
</output_format>