Interesting take on judging rolls. This is meant for full scale, but the concept is the same. I found that I was under-estimating a lot.

https://jimbourke.com/index.php?page=rolltrainer

Very interesting.....

that is cool.

that is some really cool software. 20 tries at 20 different rolling elements and the results are below. one thing that is interesting is the number of times the plane does not enter the roll wings level and you to incorporate that into the downgrade.

25% perfect (no or insignificant error)
25% excellent (small margin of error)
30% good (within the ballpark)
20% awful - (you sure you were watching the right plane?)
some of which I attribute to learning how the software works.

Being off at the start is often missed. I know I missed it the first few times. You also have to take perspective into consideration. Level at altitude will look different than down low. I know the full scale guys always talk about “level” being dependent on where you are in the box. It’s not supposed to matter (level is level), but most judges don’t look at it that way.