Accuracy according to need
The question of accuracy is tough, but there are general guidelines:
Default accuracy is recommended for all but the most complicated parts.
Absolute accuracy can solve problems with merges parts of differing size.
Pro/E's concept of absolute accuracy is fairly simple. Given an absolute value of .001, Pro/E can only measure an edge of .001 unit. (What unit? Feet, actually. Create a test part to check!). Interestingly enough, Pro/E will scale the absolute accuracy if the units are changed!!!
Based on that concept, the shortest edge Pro/E could measure with relative accuracy at default (.0012) is .0012 feet * model size. For a 1 foot cube, that would be .0012 x sqrt(3) or .002 ft (.025 in).
Is that good enough for the intended purpose of the model? For example, a cube shaped casting twelve inches on a side will have a model size of ~21 in. A 90deg .03 round will have an arc length of .047 in - no problem
A .01 round would have an arc length of .015 - that should be a problem. Why didn't that generate a geom check? Well, according to PTC, there's a fudge factor that's somewhere around 3 - so much for empirical analysis.
Do these calculations on sample parts. Check for geom checks.
Certainly, absolute accuracy should not need to be any tighter than smallest edge (ft), unless of course, the minimum accuracy value is higher. Converting to relative accuracy is simply dividing absolute accuracy by model size (ft).