Policy Settings

Compare request-shaping settings on the same gpt-5.5 environment. A model override changes the policy fingerprint while preserving the environment fingerprint, making EnvironmentDiff valid.

Files

basic.py — override the environment's low-reasoning policy with high reasoning effort and diff the results.
reasoning_effort.py — inspect the fingerprint split and task-level deltas for the two supported effort settings used here.

When to use

Use this after _15_prompt_comparison/ when the tasks, scorer, tools, and prompts stay fixed and only the model request policy changes. A prompt-bearing model override is still an environment change.

The examples compare gpt-5.5 low versus high reasoning effort; they do not use a different model family.

Run

bash

python cookbook/environments/_16_policy_settings/basic.py
python cookbook/environments/_16_policy_settings/reasoning_effort.py

Requires OPENAI_API_KEY.