I wouldn’t be surprised if it occasionally isn’t completely accurate.
I’ve just added an experimental feature that lets you use a larger AI model to generate these context aware translations. Check it out here: Experimental option to use full gpt-4o model for AI features