Does anyone know the reason behind this ? Why aren't the 5 attempts kept to calculate the average?
Back in the day (
way back, like 1990s), people used to measure their averages by timing twelve solves, dropping the highest/lowest and averaging the rest. This is pretty easy to do with a standard calculator or even by hand, since you're just dividing by ten (i.e. moving the decimal point one position). If you look at Jessica Fridrich's website,
she mentions considering only fullstep solves, a quaint notion by today's standards. I guess at some point she decided that getting lucky wasn't a crime and switched to what we now call ao12; see also
this post by Chris Hardwick on the history of speedcubing.
This was already the standard for determining an average at home in the late 1990s / early 2000s, so when the WCA came around (circa 2003?), I
suppose they decided they wanted to adopt something similar, but actually doing 12 solves in a competition would take too long, so they went for just 5 solves (averaging the middle 3) instead of 12 solves. Just like the ao12, the ao5 is robust against one outlier: getting one especially lucky solve (e.g. LL skip) or one especially unlucky solve (e.g. corner twist DNF) will still affect the ao5, but not nearly as much as compared to using the mean. For the events with short solves (pretty much everything other than 4-7, mega, bigbld, mbld with 3+ cubes) and current popular solving methods, skips play a huge role in getting good single results, but they're not fundamentally linked to a cuber's skill and are just a matter of luck; using a trimmed average reduces the influence of luck. (I'm not counting "forced skips", e.g. using VLS to "skip" OLL, as true skips here.) The same remark applies to all the reasons people get DNFs in solves (corner twist, missing a slice move, etc.).
(It's also important to note that it's not fundamentally that important,
within one competition, whether mo5 or ao5 is used. The probabilities of getting especially lucky/unlucky singles aren't that high. However, if you want to track records
across competitions, trimmed averages seem to make it harder to grind for (average) records just by going to loads of comps and getting lucky on a few of them.)
It’s not exactly fair if somebody averaging 1 minute has worked hard to get here, then messes up his H Perm at the end, and gets a DNF, which results in a DNF for the entire solve. This applies to everybody, actually.
If you've messed up a solve, you've messed up the solve. If the rules had been to use the arithmetic mean (which they aren't, but just for the sake of argument), I don't think it would be unfair per se to make the whole average a DNF too. This is already the case for all the events ranked by mo3 (FMC, 666, 777); I DNFed out of an FMC podium once, but I know that was completely on me and I don't consider that unfair.
