funny my experience on this has been a bit variable with hardware…
so with computers using usb its very true, as they do not prioritise USB midi data (due to protocol they use) whereas audio is prioritised (different ‘usb class’)
but once you moved to embedded hardware, that bet is off the table.
sequencers won’t be running at ‘audio rate’, and so have ‘sample accuracy’ they will be running at the resolution of the events … so Hapaxx is 192 ppqn.
so the audio and midi clock is going to have that accuracy.
(I think this is why some old gear has a reputation of being very tight, even over midi)
a midi serial link can be fast enough to support it…
I guess one ‘issue’ is , with midi you are dependent on both the sender (e.g. hapax) and the receiver (in my case hermod)… if both are fast, you are ok… (*)
whereas with CV, you just get the signal and use it.
so yeah, its a bit mixed…
also something else to consider … I recently did a test on a computer (Mac) of midi latency vs cv latency… and generally midi is less latent (due to no audio buffer) BUT is more susceptible to jitter.
but susceptible is key here… on a well setup system, I don’t get that much jitter either.
so I don’t really know
whilst my heart say, cv clock is best… in practice I usually end up using midi
(*) that said, I’ll point out that hermod and many other sequencers basically use external clocks to determine tempo, so its often more down to how fast they react to clock changes than accuracy of clock
(**) I should say the above is ‘simplified’… as its going a bit off topic… and there are so many variables , eg. things like the ADC converters used by modules/sequencers etc… and we don’t really know ‘polling rates’ , irq priorities, and a whole host of other variables.