I remember an add, back in the Intel 386 days. It said “turn your 386 into a 486”. Reading between the lines, it seems what it did was replace the trig calculation software library so it would do 4 digit calculations instead of 8-digit (or maybe just added a trig lookup table). In the day, without math coprocessor chips, this saved a lot of calculation time and significantly sped up some video games. After all, the VGA resolution of the day maxed out at 800x600 pixels, so 4 digits of accuracy would still be overkill in most games. (Which is what the ad was aimed at)
The point the OP is aiming at, I assume, is what good were the extra digits, how often were they used? Considering they typically only used discrete angles, you are working with 2 digits of accuracy for the angle. Did anyone ever interpolate to get fractional angles? Presumably that is reasonably accurate when not approaching asymptotes. 4 digits is 1 in 10,000; that’s 1 inch in 833 feet, or about 1/4 inch in 200 feet. In asymptotic situations, a different strategy may be necessary.
I ran across a program on a numerical controller where the programmer used a linear scale for a result that actually depended on square root. Considering the process did not need much accuracy, the range of error for the anticipated range of input was close to acceptable…)
(Also I recall reading once that someone discovered an error a few digits into an entry in one of the earlier 1800’s trig tables, which apparently was repeated by many other publications. In the early days, such tables were complied by hand.)
I took a course in computational mathematics, where we delved into how errors propagate in mathematics. Most complex math is done in binary floating point, including conversion back and forth to decimal. So the accuracy of what a program will produce is limited by the type of floating point number representation used. 64 bit, for example (if deep distant memory serves) allows for 1 byte of exponent (+/-127) and 7 bytes of mantissa. Using the rule of thumb 10^3 = 2^10, your best accuracy with 56 bits is about 16 significant digits.
However, order of operation is important in math, particularly with iterative calculations - errors can blow up on repetition. One example cited was the early use of computers to calculate the wing spar needed for a new design of jet airliner in the 1960’s. The answer came out - 11 feet thick. Time to re-examine the algorithm.
For trig functions, the most efficient algorithms have been determined long ago to minimize calculations to get to high accuracy.