Hi Tim,
You are not bothering anyone but there has been lately very similar questions and it hints at the fact there is not enough documentation on that topic or at least, easily accessible documentation.
The core of the issue lies in the fact that a random internet image has already been almost certainly rendered, i.e. it has a Look applied to make it pleasing or faithful to the original scene on the authoring display, and encoded for Display presentation. We say that the image is in a Display-Referred (or Output-Referred) state.
With that in mind, ACES is designed to process images that are in a Scene-Referred state, thus, if you push an image from the web that has been already rendered through ACES, you will basically render your image two times which is unsuitable.
The Utility - sRGB - Texture colourspace will only decode the image by applying the sRGB EOTF but it will not undo any rendering. There is no way of knowing how a random image was rendered.
Assuming that you did manage to un-render/de-render your Display-Referred image and thus brought it back to the Scene-Referred state, pushing it through the sRGB Output Device Transform would still not make it look like the original one because the ACES rendering is obviously different to that of your image.
When you do that, nothing happens in the display chain: The image is not decoded to linear and it is not re-encoded for display, it is basically a No-Op.
The ideal workflow would be to no use random internet images but only properly acquired Scene-Referred images. If you do really have to use random internet images, then you need to de-render them, I would suggest reading that thread as there are few things of direct interest for you: Linear-sRGB not behaving as expected
Cheers,
Thomas