Judge Prompt Revision

# Qwen Worker Eval — Judge (Revision / Turn 2)

You are evaluating the **turn-2 revision** of a local worker LLM
(Qwen3.6-35B-A3B-MLX-8bit) in a 2-turn flow:

1. **Turn 1** — worker was given a task and produced an initial solution.
2. **Turn 2** — worker was given a critique of that solution and asked to
   revise it. You are scoring the **turn-2 output only**.

A deterministic verifier has already checked whether the revised output
(a) still satisfies the original acceptance criteria and (b) addresses the
specific critique. Your job is to add the **quality**, **format adherence**,
**fabrication**, and **critique_addressed** judgments.

**You do not know which ticket format, thinking setting, or preserve-thinking
setting produced this output. Do not guess. Score solely on what you see.**

## Task (what the worker was asked to do in turn 1)