评测量规Evaluation rubric

设计可观察评测量规，不打分排名。 输入：评测目标、任务样本、约束。

Output dimensions and observable criteria; no total-score ranking. Inputs: Evaluation goal, task samples, and constraints.

设计量规。输出 dimensions[]：name、observable_criteria[]、evidence_required、failure_examples。判据必须可观察、可复核；不得给权重总分或排名。若需模型评判，写明位置偏差与人工抽检口径。

占位符用 {字段} 标注，运行前替换为本次输入。

Design a rubric. Output dimensions[] with name, observable_criteria[], evidence_required and failure_examples. Criteria must be observable and checkable; do not assign weighted total scores or rankings. If model judging is used, state position-bias and human spot-check rules.

Replace {field} placeholders with the current inputs before running.

dimensions[]，无总分排行。

dimensions[] without total-score ranking.

• 判据可观察。
• 无权重总分。
• 模型评判有偏差口径。

• Criteria are observable.
• No weighted total score.
• Model judging has bias rules.

• 判据空泛：改成可核对动作。
• 维度重叠：合并或拆清。

• Vague criteria: turn into checkable actions.
• Overlapping dimensions: merge or separate.

提示词效果未实测。机制依据见出处；换模型、版本或输入会改变结果。

This site has not tested the effect of this prompt. See the sources for the mechanism; changing model, version or input can change the result.