You can't use madVR with mpv.io which is a deal breaker for many (better scaling algorithms, better dithering, display calibration with 3DLUT, HDR conversion/processing and so on).
mpv supports display calibration and HDR conversion as well. You can load a 3DLUT for display calibration if you want, but for mpv, you would more likely load an ICC profile and have it generate the correct LUT automatically. You can even just let it pick up the ICC profile configured in the operating system. As for HDR, there are a number of built-in tone mapping algorithms: https://mpv.io/manual/master/#options-hdr-tone-mapping
I disagree on a couple of points though. Jinc may in fact produce noticeable ringing (while NGU doesn't), and neither produce aliasing so asserting the opposite seems strange. The difference in sharpness is also quite obvious. On top of that mpv's anti-ringing filter is very basic in comparison.
HDR processing also seems quite limited and requires a lot of hand-tuning compared to madVR.
If it works for you, great, but it would be disingenuous to say that both programs are at a feature parity.
Context for others: NGU is a proprietary algorithm created by the author of madVR.
I'll admit I haven't watched any HDR content in mpv yet, but it supports all the standard tone-mapping algorithms (hable, reinhard, etc.) and apparently the default algorithm (mobius) was chosen for its colour accuracy.
I'm not saying madVR doesn't have its strengths. Error-diffusion dithering is strictly more accurate than what mpv does, and for people who like super-resolution upscalers, madVR tends to have a larger selection and faster implementations. Still, it would be disingenuous to say that madVR has a larger feature-set than mpv. They have different feature-sets. mpv's convolution-based upscalers are much more tunable and its colour management is more advanced, since it can use an ICC profile to auto-generate a LUT for any input gamut, rather than relying on an external CMS for this.