Decide when error budget burn warrants a freeze

An agent watches each service's error budget burn rate: how fast the service is spending its allowed errors, where 1 means exactly on budget. A short window catches spikes and a long window catches sustained burn, and the agent must not cry wolf on one noisy reading.

Write burn_rate_decision(samples, fast_window, slow_window, fast_threshold, slow_threshold): the rule your agent follows on each check.

• samples (list of float): burn-rate readings, oldest first, newest last.
• fast_window, slow_window (int): how many of the newest readings each window averages.
• fast_threshold, slow_threshold (float): the average each window must reach to count as hot.
• Return "recommend_freeze", "page_owner", "no_action" or "insufficient_data".

Rules, in order:

1. If either window is below 1, or slow_window is smaller than fast_window: "insufficient_data".
2. If there are fewer readings than slow_window: "insufficient_data".
3. Average the last fast_window readings and the last slow_window readings. A window is hot when its average is at least its threshold.
4. Both windows hot: "recommend_freeze".
5. Only the fast window hot: "page_owner".
6. Otherwise, including a hot slow window with a cool fast one: "no_action".

Then, in a sentence or two: should the agent freeze releases on its own, or only recommend a freeze? Take a side.