At HiFi it is important to add the "right errors" with the right structure while keeping the reproduction precision as high as possible.
Only if HiFi means: High Preference. HiFi means High Fidelity though. It means the least possible error and in practice it means errors below audible thresholds.
Again... you need to differentiate between transducers and the electronic path when you make generalizations like you wrote.
Hiraga found that a steadily declining THD pattern was preferable
Preferable does not mean higher accuracy, more true than the recorded signal. Adding something that isn't there in the actual recording to make it sound 'better' should be done in the mixing/mastering stage. Why would adding a
certain type of harmonic distortion in a specific amount (and thus also IM distortion) make any recording suddenly sound more real ?
On top of that... that preference for added distortion (in a specific manner) was VERY measurable wasn't it ? It was not something that was not measurable yet increased 'fidelity'.
This is how Matti Otala found the concept of TIM
Matti did measure a specific type of distortion didn't he ? It was heard and measured and found. It was a degradation of sound quality and not an improvement by adding slew rate distortion (see the video you like to refer to).
TIM is not a problem anymore as it is known and designs can be made (fast output stages + low pass on input) so that this audible error is no loger bothering us. What's it have to do with 'things that cannot be measured' and why would adding a certain amount of it ever 'improve' sound quality ?
Why would adding crossover distortion 'add' to perceived fidelity ?
Differences can be measured on pretty much anything more or less meaningful or meaningless.
Finally something you are correct about. Let's change the word 'meaningful' to 'audible' because if it is not audible it is not meaningful.
The measurement typically makes no sense before listening to what has caused the measurement difference.
Unless you try to design something that has the least possible errors or the designer knows about audibility of the specific measured error(s) or when you try to 'emulate' a certain HD 'signature' to mimic specific types of devices.
Hifi and its development is just as much about improving, assessing and choosing which object or topology among two or more is the most well-sounding and true-to-nature in sound, where the tested objects are not defective, but only different.
No HiFi is about highest fidelity and not about preference. The fact that some folks prefer certain amounts and types of distortion does not make it higher fidelity or 'more true to life' but is about preference and those types of 'added distortion' are all VERY, VERY measurable and design-able by the grace of measurements which will tell in which ballpark the 'added' distortion of the original signal should be to please individuals that claim to prefer this distortion. But only if they KNOW that type of 'pleasant' and 'sound quality/signature' enhancing changes are added.
What you mean by 'improvements that are (currently acc. to some) not measurable need to be proven first. Which they can't when tested 'blind'.
Things like silver instead of copper wiring that, when 'tested' sighted, are clearly there but disappear the moment you don't know have been used.
This is something very different than closing the eyes while still knowing it is there. That is not 'blind' testing despite sight not being used.
Obviously you have 'tons' of evidence in your design efforts that show you things YOU do not measure yet sound different (but only when you know). You seek explanations by 'measurements' and continue to try to tell the 'scientific' community to validate this but refuse to play by their rules.
You must prove without a shadow of a doubt (means statistical) with rigorous test methods (not knowing what is playing BUT knowing what to listen FOR) to impartial folks the 'improvements that do not show up in any measurements DO exist. You failed so far and that's why you hang on to the 'knowing' part. That makes it easy.
