Originally Posted by
Plastkin
Except it is because that's how you exclude things like parses of people who die (a 95% player dies and still does > 80%) which contribute a great deal to the bloat. Anything below 70% is completely unacceptable for a mythic raider, so you're really looking at about 30% of parses being actually useful.
It's also not an RNG thing, it's how probability works. If your sample size is too small, outliers are harder to spot, but even at a handful above 99% you're still getting relevant information. The skew looks exactly the same as sisters for instance. There are 6600 logs for Havoc on sisters, the 99% mark is 66 which is 1.471 million DPS. 99.5% is 33 which is 1.501 which is ~2% more DPS. 100% is 1.623 which is 10% more than the 99% figure. This large skew happens generally around the top 10-15 with this many logs, where perfect kill times and/or RNG line up to give someone outlier-level DPS, and you tend to see this happen in heroic a lot more because guilds that stomp mythic don't generally try to speed kill heroic, it's less pronounced in mythic because most top guilds that log are always trying to kill bosses faster, so you're not looking at thousands of shitty guilds barely killing the boss and a dozen guilds killing it minutes faster. The outlier range for this starts somewhere around 99.75%, which is looking at the top quartile of 99%. The numbers at 99% are reasonable, and comparing them say to 95% (330 or 1.387 million DPS, or 94.3% of the 99% parse, or in other words the 99% cutoff is only 6% higher than the 95% cutoff, but the 100% is 10% higher than the 99% due to it generally being an outlier) you find that there's a smoother gradient there than in the top 50% or so of the top 1%, which means the issues you raised aren't really a concern at that point.
If you exclude Avatar and KJ due to sample issues, the standings remain the same with DH at the bottom in the 99%, I don't think sample size is an issue here. And it's not 4 or 5 people are good and the rest are trash, it's 4 or 5 people are realistically close to perfect and are what the spec is capable of with the best gear you can obtain, without dying, without bad RNG (not outlier RNG), and with a kill time that isn't completely fucked.
If everyone had the exact same gear, exact same kill times, nobody ever died, and everyone played at the same skill level, I'd say using a 50% would be reasonable, just like simcraft does, but realistically to normalize for all of that you have to look at the top few percentile.