Skip to content

F3 Over-refusal

The model refused a harmless request (a school play villain, killing a hung process).

How counterfix finds it

Live: refusal phrases near the start of the answer. Judge: confirms the request was actually harmless, because phrases alone misfire.

Fixes it tries

In this order (a model profile can reorder them). Each one is kept only if it passes 2 of 3 re-runs.

Intervention Change
state_purpose Tell the model the request is legitimate and to help unless it is truly unsafe

When no fix verifies

  • State the purpose and audience in the prompt
  • Try another model