I feel like curent pp system favors short maps by a significant degree and I feel like a small change in the algorythym could fix that. Something like curent algorythym + lenght in seconds - lenght of map parts with no hp loss in secunds x2 = pp could propobly fix that problem. How would yo feel about that?
Note: I am cunting map lenght as time from first point with hp loss to the last point with hp loss for the sake of a fair algorythym.

maps with more circles (which longer maps usually have) already give more pp.

it's just that most of the shorter maps, that make it seem that long maps are underweighted, usually have short spikes in difficulty which the current pp system is not very good at handling.

also The Unforgiving [Marathon] (the longest ranked map in the game) would be around 2500pp for a nomod ss
(301 + 2877 - (300*2)) (i guessed that the break time (time with no hp loss) is around 300 seconds)

Ok, but still seeing how shorter maps tend to be the more played I would still like a change in favor of longer maps simular to what I proposed howewer giving much less aditional pp than what I proposed to make osu! less focused on short maps (maybe the same change but giving 0.5% of its initial ppp change + making it modified by accuracy). What would you think about that?

Including map length in pp is easy to say, but doing it properly is hard. Length is not the only thing additional pp needs to be given because of, but the difficulty of the sections during that time as well. For example, a short difficult section in the beginning of a long map while the rest is very easy is approximate to a short map containing just the difficult section. Where that difficult section occurs matters too. Move that short difficult section towards the end and difficulty should increase.

There is a formula to calculate this, but unfortunately it can be slow, and requires to plug in the probability of breaking combo for each score point. It's still unclear how to properly calculate the probability of breaking combo for each score point.

Expected time it would take to FC a map =


Where:
- N is number of score points
- t_n is time between current score point n and the previous
- p_m is probability of breaking combo at score point m

Maybe p_m should be based on an avrage but with cuting both extremes from the calculation? (just a sugestion)

Pioterrrr wrote:

Maybe p_m should be based on an avrage but with cuting both extremes from the calculation? (just a sugestion)

Average of what?

How commonly people break combo at score point m

Pioterrrr wrote:

How commonly people break combo at score point m

"How commonly people break combo at score point m" is the same as "probability of breaking combo at score point m", which I previously stated that it's unclear how to properly calculate.

Take data from players?

What data is required:
- Replays from various players, from low to high skill, with specific mods on specific maps
- Replays from specific players, from low to high skill, with specific mods on various maps

While data can be collected to help with making the formula, there needs to be a lot of data available to be able to figure out how various factors play into breaking combo. A lot of that data is inaccessible - only top 500 plays in leaderboards have replays, so there is no way to get data on lower skilled players, and no way to get data on plays using EZ mod.

As for data that is accessible, you need to filter out plays that have no combo breaks - meaning all maps less than 4 stars are ruled out because most top 500 plays on such easy maps tend to have no combo breaks. What you have remaining are plays with varied mix of various mods that make it hard to isolate factors, and 7 star+ maps for which most players cant FC no mod.

You also need to analyze players, not just the maps. This requires many replays from single players to figure out how they respond to various patterns - what their skills are - to build a player profile to test models against. Currently it's possible to collect this kind of data from top players only. Unfortunately the data would have various mods, so again, this makes it hard to isolate factors.

There has been a thought to get data from maps that have been recently ranked (before having 500 scores on it), but you need automation for that sort of thing and know which maps you want to target. It would an active effort of constantly going through qualified and modding queues to make a list of potential candidates to get data for. That would be quite a commitment.