On the 40 MSD limit

A common point of discussion and misconception in the Etterna community is the idea that 40 MSD is a hard limit for the calc that must never be broken no matter the circumstance. The actual explanation is more nuanced: there is currently a 40 MSD limit on scores, but it was implemented several years ago as a hack to minimize the amount overrated charts could be abused for rating.

The next question usually after this tends to be "Is the 40 MSD limit outdated?". The answer tends to be no; with most arguing that the best "real" scores are around 38 MSD currently.

It's not obvious what makes a score "real" though. Obviously farm can be excluded, but what about things like Myuka's Image Material 1.5x AA? That chart most people would agree is at least 28 MSD on 1.0, and MSD scales linearly with speed, so 28*1.5 = 42 MSD sounds reasonable, right?

The reason people don't agree with this reasoning is because of a simple fact of high level Etterna play: the faster the patterns are, the more you can mash while keeping good accuracy.

This is evident even at lower levels of play in some instances; for example, you can easily jumptrill the 32nd rolls in Cyber Induction on 1.0, but if you slow down the chart to 0.5x it becomes impossible. Hachigatsu no Yoru isn't jumptrillable on 1.0x, but it certainly is on 1.6x. The higher you increase the speed of patterns, the more you can approximate.

The calc doesn't account for this approximation however, outside of very simple checks to ensure charts like To Dimension and Comsten aren't extremely overrated. So the idea that Image Material 1.5 is worth 42 MSD carries the assumption that you need to play the chart on 1.5 with the same technique that you would on 1.0, just 50% faster.

This is evidently not the case, however. I have written a userscript that flags manipulated notes in Etterna scores by checking if two notes further than 10 milliseconds away from each other are hit within 10 milliseconds of each other. Out of the 4000+ notes in Image Material, over 1000 of them are flagged as manipulated in Myuka's 1.5x score. If you compare this to any 1.0x score, this is much higher; the average 1.0x 94% score on this chart has only around 60 notes flagged.

This argument unfortunately usually gets simplified into calling these scores "mashed" or the players that set them "mashers". Which isn't necessarily wrong but the lack of nuanced explanation makes these criticisms come off as baseless and buzzwordy, and likely contributes to the notion that people who care about manip are just bad players trying to find excuses for their ego.

Another interesting side effect of leniency at higher speeds is that for extremely high speeds (think 200+ BPM dense chordjack), you cannot actually CB rush because the next note is so close that you cannot actually get a good/boo judgement; you'll just get a great rush which is much less punishing. To actually CB rush you need to CB rush again on the same column before you cancel the great rush. This leniency is why vibrating 200+ BPM dense chordjack can work; vibrating jacks provides the speed to keep up with the pattern, while the leniency of the judgement system at high speeds compensates for vibrating's biggest downside: lack of control.

(If you want to see an example of this in action, look at this score on Vertex Beta vROFL 1.05x by FX8320. There's great rushes but almost no CB rushes.)

So, how can this be fixed? The most naive way would be a Wife4 that makes the scoring system even harsher than Wife3, or moving to a harder judge like J5, J6, or even J7. However, this is extremely punishing to new players, who naturally struggle with hitting accurately and also have no opportunity to manipulate patterns in the first place. Consequently, a better system would be one that scales the tightness of the judge window with the difficulty of the charts themselves.

Manip Factor is probably the closest thing to this that currently exists; it is a metric of how manipulated a score is that scales with the average NPS of the chart. However, average NPS is a very poor metric for chart difficulty; and it leads to the values being very underestimated for spiky charts. A better approach may be to base it on MSD instead; though the issue with this is that the scoring system would basically change every time MSD changes.

Another approach that has been considered is making the difficulty calculator assume pattern manipulation and treat notes that are close enough together as if they were chords. I think this approach is very liable to causing false positives and making the difficulty calculator less reliable for non-manipulating players, however.