What an image has to be for an attention model to mean anything on it.

Mirrors tbp/monty/frameworks/models/salience/strategies/vocus2/vocus2.py::OperatingLimits, a Protocol in Monty. A sigma is a distance in pixels, so the same sigma means a broad blur on a small image and a tight one on a large image; limits are how "these numbers were only ever tried on images this size" is enforced instead of remembered.

Monty answers a ValueError or None; a behaviour cannot answer an exception, so this answers :ok or {:error, message} and the caller decides whether to raise (which is what Monty does unless the run suppresses runtime errors).