Which is more accurate: digital or vinyl?

64 bits? Bit depth describes the dynamic range, 16 bits=96 dB, 20 bits=120 dB, 24 bit=144 dB. The later could record a jet taking off and a pin drop without changing the gain. That's more than enough for the most dynamic music with headroom to spare.

"Bit depth describes the dynamic range"

That is true for any particular application or format spec, like CD redbook, but not in general.

More bits can mean more accuracy and/or more dynamic range. So if the specified maximum dynamic range stays the same, more bits will mean finer resolution per sample in terms of representing amplitude more accurately.

How does a finer definition of amplitude equate to resolution? Would smoother amplitude levels be audible?

Assuming fixed dynamic range, resolution results from a combo of the # bits used to quantify each digital signal amplitude value captured and how frequently (the frequency with which) the signal is quantified.

So a finer definition of amplitude contributes to higher POTENTIAL or maximum real resolution assuming the maximum dynamic range stays the same.

How much resolution is actually achieved is a different story that depends on how well executed the implementation is but more bits used will enable more resolution, all else held constant.

Alternately, increasing the dynamic range and quantifying using the same # of bits will result in lower potential or maximum resolution because the same # of possible digital values must now be used to represent a larger variation in amplitude.

So yes, theoretically smoother amplitude levels can occur. Whether these are audible or not still depends on other factors and makes things more challenging to model accurately.

"sample" is really not an accurate term to apply in the digital signal processing scenarios because a sample is a statistical concept as well and in stats means a small but representative subset of a population.

Whereas with digital audio and video the intent is to quantify the entire signal as completely as possible, not 'sample" it, which is different.

Maybe this is part of why there are so many misconceptions about digital audio?

16 bits specifies 65,536 levels, approximately 0.00146 decibel per level. I've read conflicting accounts. What's the smallest level change the human ear is capable of discerning?