Btw I’m not sitting through the 20min video you posted. But fast-forwarding through it, it looks like his conclusion is the same: FSR implemented in-game is always going to look better than RSR, because it is placed in the right stage of rendering (the when I talked about in my message). Neither OpenXR Toolkit nor RSR are placed where optimal, therefore the resulting quality is inferior (see 6 minutes in the video). This is also what my diagram showed above (with quality being inferior for implementation outside the game engine).
Regarding the rumor you mentioned, I doubt it’s accurate. The most recent development I’ve heard of from NVIDIA is that this week at their annual GTC conference, they are announcing a software called the Streamline SDK, which will greatly facilitate engine integration of DLSS. Snippet from their slide deck (available publicly if you register for GTC):
This Streamline SDK will enable game developers to spend minimal amount of effort integrating the “SL interposer”, which is solving the exact problems of the when and the how I mentioned in my message. Basically giving developers and easy way to answer this when and how questions to enable things like DLSS.