/jp/ - Otaku Culture

>>18029773
Your example changes the offset by about 0.2, not 0.1, and it even helps illustrates the point that you probably shouldn't bother with 0.1 increments until you're already very confident that you're close.

Going off of your example (-19ms~9ms), let's assume a player has timing that can be modeled with a normal distribution with a 14ms standard deviation, does not drift based on in-game early/late feedback, and already has things like WN/GN set correctly. He is playing on an unfamiliar machine, so he's 5ms off on average. Ideally, he should be getting 87.44% scores, or high AAs, but he's getting 85.61%. Something's off, so he adjusts offset by 0.1. On a 1000 note song, he sees an improvement of about 20 EX, or a 7% reduction in lost EX. That's already not that much, but it's still pretty good, and you might think that the adjustment of 0.1 was a huge success. Instead, consider that such a "large" improvement with 0.1 is just evidence that the adjustment should have been much larger to begin with, since a 0.3 change would have had almost double the reduction in lost EX.

In reality, using 0.1 increments is even less useful, since any reasonable player would adjust themselves based on the judgement feedback while playing the game. At best, this means that the score differences between using different offsets would be even more subtle, and at worst, the player may be hurting themselves in the long run. If the offset is off by a lot, like 1.0, as a player makes 0.1 increments, they're also very likely to be slowly adjusting themselves. After they're satisfied with the timing, they may have drifted to the point of meeting somewhere halfway, where a 0.5 total adjustment feels fine. After returning to a familiar machine and using the offset they normally would, it might feel strange and require yet another period of adjustment.

For most players, I would advocate adjusting offset in 0.5 increments, or 0.3 at the very least. Most of the time, you'll more quickly find the offset you need, and overshooting and scaling back helps the player avoid drifting too much in any one direction while searching for the correct offset. Then stick with an offset once you're confident enough. I often see less skilled players constantly adjusting their offset up and down by 0.1 between songs, and it's silly.

After experimenting with some numbers, 0.1 offset error appears to have the most noticeable total EX improvements only for 95+% score players that have a standard deviation of 10ms or less, and even then it's still within the noise/placebo range.

>>18031991
Post scores.