What it measures

Evaluates nuanced contextual non-compliance (unsafe requests, false premises, unanswerable queries, unsupported modalities) vs. over-refusal.

Reading the measurement

Separate satisfying instructions from factual accuracy, human preference and safe behavior. Inspect strict all-constraint success as well as average criterion credit. Judge choice and adversarial prompt selection can materially change the result.

Published evaluation

The source cell reports without / with a system prompt. Compliance is undesirable on noncompliance categories but desirable on the contrast set; direction follows that distinction.

Where this fits in an evaluation suite

Format compliance, preference, truthfulness, and safety require separate measurements. Keep desirable writing style from masking factual errors.

Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.

A reproducible comparison

The source cell reports without / with a system prompt. Compliance is undesirable on noncompliance categories but desirable on the contrast set; direction follows that distinction.

Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.

PRIMARY EVIDENCE & FURTHER READING

  1. The Art of Saying No: Contextual Noncompliance in Language Models — published result table ↗

    Brahman, Faeze, Kumar, Sachin, Balachandran, Vidhisha, Dasigi, Pradeep, Pyatkin, Valentina, Ravichander, Abhilasha et al. · 2024

    All 14 authors

    Brahman, Faeze, Kumar, Sachin, Balachandran, Vidhisha, Dasigi, Pradeep, Pyatkin, Valentina, Ravichander, Abhilasha, Wiegreffe, Sarah, Dziri, Nouha, Chandu, Khyathi, Hessel, Jack, Tsvetkov, Yulia, Smith, Noah A., Choi, Yejin, Hajishirzi, Hannaneh

Reviewed 2026-10-27. Explanation and interpretation by RLTasks. Benchmark ownership remains with its original authors.