I really appreciate you running these tests. I’m actually fairly impressed at the results 
This is not totally surprising as our pipeline is technically 32-bit floating point. A back of the envelope computation shows that should be more than enough to pass 24-bit integer data untouched, but I think it would miss a few lower bits on 32-bit integer data.
The good news here (if you’ll allow me to be glass half full) is that 32-bit audio files are (in my own personal opinion) pretty rare (I can’t say I’ve seen any in the wild), and if I might go a step further, completely unnecessary for human ears.
This is extremely good news (not that I respect or like MQA if we’re being honest), but I was hoping this would be the case.
We’ll eventually do this, it just turns out to be a bit more complicated than the opportunistic version.
Again, I really appreciate all the testing you’ve done. You have a standing invite to our beta Slack channel if you want access, just drop me a DM.
