The assumption is that Rushton Paul is using Mac for playing 32bit Float DXDs with the Playback Designs MPD-8 DAC (which cannot play 32bit Float DXDs).

Audio Device Setup for Mac
_https://www.sweetwater.com/sweetcare/articles/audio-midi-setup-for-mac/

MacBook Pro (2023)   
Applications > Utilities > Audio MIDI Setup

MacBook Pro Microphone (built-in)
1 channel
Bit rate: 32bit Float 
Default sample rate: 48kHz
Maximum sample rate: 96kHz

MacBook Pro Speakers (built-in)
2 channels
Bit rate: 32bit Float 
Default sample rate: 48kHz
Maximum sample rate: 96kHz

It means, perhaps, that might be possible to record "The Year 1812, Solemn Overture, Op. 49"

You can play them with MPlayer

sudo apt install mplayer

$ mplayer "08-Faust - Funeral March Of A Marionette - 32bit.wav"
MPlayer UNKNOWN-12 (C) 2000-2023 MPlayer Team
do_connect: could not connect to socket
connect: No such file or directory
Failed to open LIRC support. You will not be able to use your remote control.

Playing 08-Faust - Funeral March Of A Marionette - 32bit.wav.
libavformat version 59.27.100 (external)
libavformat file format detected.
[wav @ 0x7fab12c87d00]Discarding ID3 tags because more suitable tags were found.
[lavf] stream 0: audio (pcm_f32le), -aid 0
[lavf] stream 1: video (png), -vid 0, Picture
VIDEO:  [MPNG]  607x603  0bpp  90000.000 fps    0.0 kbps ( 0.0 kbyte/s)
libva info: VA-API version 1.17.0
libva info: Trying to open /usr/lib/x86_64-linux-gnu/dri/iHD_drv_video.so
libva info: Found init function __vaDriverInit_1_17
libva error: /usr/lib/x86_64-linux-gnu/dri/iHD_drv_video.so init failed
libva info: va_openDriver() returns 1
libva info: Trying to open /usr/lib/x86_64-linux-gnu/dri/i965_drv_video.so
libva info: Found init function __vaDriverInit_1_8
libva info: va_openDriver() returns 0
==========================================================================
Opening video decoder: [ffmpeg] FFmpeg's libavcodec codec family
libavcodec version 59.37.100 (external)
[png @ 0x7fab121fdd20]Requested frame threading with a custom get_buffer2() implementation which is not marked as thread safe. This is not supported anymore, make your callback thread-safe.
Selected video codec: [ffpng] vfm: ffmpeg (FFmpeg PNG)
==========================================================================
Clip info:
 artist: Alexander Gibson -RPHO
 date: 2024-01-15
 album: Gounod - Bizet - Alexander Gibson Royal Opera House Orchestra
 encoder: Merging Technologies Album Publishing
 title: Faust - Funeral March Of A Marionette
 track: 8/9
Load subtitles in ./
==========================================================================
Opening audio decoder: [pcm] Uncompressed PCM audio decoder
AUDIO: 352800 Hz, 2 ch, floatle, 22579.2 kbit/100.00% (ratio: 2822400->2822400)
Selected audio codec: [pcm] afm: pcm (Uncompressed PCM)
==========================================================================
AO: [oss] 192000Hz 2ch s16le (2 bytes per sample)
Starting playback...
Could not find matching colorspace - retrying with -vf scale...
Opening video filter: [scale]
Movie-Aspect is 1.01:1 - prescaling to correct movie aspect.
[swscaler @ 0x7fab12e13f20]bicubic scaler, from rgba to yuv420p using MMXEXT
VO: [vdpau] 608x604 => 609x604 Planar YV12 
Movie-Aspect is 1.01:1 - prescaling to correct movie aspect.
VO: [vdpau] 608x604 => 609x604 Planar YV12 
No pts value from demuxer to use for frame!
pts after filters MISSING
A:   0.0 V:-9223372036854775808.0 A-V:  0.000 ct:  0.000   0/  0 ??% ??% ??,?% 0 0 0.54x 
Possibly bad interleaving detected.
Use -ni option if this causes playback issues and avoid or fix the program that created the file.
A:   3.0 V:-9223372036854775808.0 A-V:  0.000 ct:19449.693   0/  0  0%  0%  0.0% 0 0 0.54x 

Exiting... (Quit)

It does play DXDs (with downsampling and format conversion):

==========================================================================
Opening audio decoder: [pcm] Uncompressed PCM audio decoder
AUDIO: 352800 Hz, 2 ch, floatle, 22579.2 kbit/100.00% (ratio: 2822400->2822400)
Selected audio codec: [pcm] afm: pcm (Uncompressed PCM)
==========================================================================
AO: [oss] 192000Hz 2ch s16le (2 bytes per sample)
Starting playback...

Can your DAC play 32bit float audio files (which it should not play)?

Have you tried?

In general, the higher the sampling frequency, the “smaller” the steps, by smoothing which the DAC receives an analog signal. All the more subtle nuances we can hear. In practice, there is, for example, jitter, unevenness of the count caused by a change in the frequency of the clock generator. The latter is possible, for example, due to impulse noise in the power supply, interference, etc. and so on.
How does it differ in different operating modes of the resampler and different processor loads?

So I just don't know.

P.S. By analogy with the landscape outside the window. The sound reproduction system consists of different parts and the quality of each determines the degree of “cleanliness of the window”. The super quality of one element can simply be hidden by the mediocre quality of another. There are many options in each specific case.

My hypothesis is that the Playback Designs MPD-8 DAC is playing 32bit float DXDs (which it should not play), because a sound system, or a player "performs the necessary format conversions in software".

Would you agree that it might be the case?

$ mediainfo "08-Faust - Funeral March Of A Marionette - 32bit.wav" | grep Audio -A11
Audio
Format                                   : PCM
Format profile                           : Float
Codec ID                                 : 3
Codec ID/Hint                            : IEEE 
Duration                                 : 4 min 35 s
Bit rate mode                            : Constant
Bit rate                                 : 22.6 Mb/s
Channel(s)                               : 2 channels
Sampling rate                            : 352.8 kHz
Bit depth                                : 32 bits
Stream size                              : 743 MiB (100%)

If he is using MAC, DXD waves and flacs are likely to be downsampled to 48kHz (default) . OSX resamplers can do the work.

If "OSX already supports all PCM formats up to 384kHz sample rate", it does not mean that the OSX driver for MPD-8 DAC supports 384kHz sample rate. It may simply mean that OSX resamplers support 384kHz sample rate.
In this case, Foobar2000 may fail to play DXDs in "exclusive mode".

The result of downsampling depends on the quality of the resampler and resolution of the original file.
If the original DXD wave is 32bit Float, downsampling to 48kHz (or 96kHz) may produce a better sound quality.
That is why, perhaps, Rushton Paul claims that 32bit DXDs are better than 24bit DXDs.

Would you agree with this?

P.S. I didn't expect it myself.

Yesterday, upon requesting “SpectraLab...” Google returned a link to one of the radio forums with my participation from 2011.
My point is that these issues were discussed a long time ago and everything has already been clarified.

And you can imagine my “interest” in empty chatter on this topic...

I use audigy2, 24 bit, 195 kHz

Perhaps, it is Audigy2, 24 bit, 192 kHz.

It seems that your experiments can be simplified and made more exact.
All sorts of sinusoidal signals can be easily created with Audacity (with mathematical exactitude).

sudo apt install audacity 

Audacity → Generate → Tone → select frequency and amplitude

You can also change resolution:

Audacity → Edit → Preferences → Quality → "Default Sample Rate" and "Default Sample Format"

Then you may try to measure the distortions in playback. It might be very interesting.

But you may better start a new topic for this.

it was detected by an OSS4/Gentoo user in a "blind test"

(emphasis mine) One anecdotal sample is barely a "test" at all, and not even remotely scientific.
Then you go on to claim that blind tests have problems, because the usual "bit-perfect" "exclusive mode" audiophool nonsense.

As for "old theory", I've been there, done that, (as well as built and tested plenty of proper HiFi gear) so I'm no longer gullible enough to get sucked into such arguments... I'm out of here, have fun.

If, for example. a sound system performs resampling and format conversions in software, you may not know what you are actually playing with your DAC, or soundcard. That is why, perhaps, some users want to know how to enable "exclusive mode".

Although the difference between Petrov's dct and fft seems insignificant, I would strongly recommend to use fft instead of dct (default). The reason is the results of an unintended blind test with OSS4 users.

To be really thorough try it several times, with a friend changing which one is which. And a few control runs where both switch settings get the same system.