Hi! Due to the recent rating system being developed, and many players' desire
for it to rate players by their playing skill (as opposed to amount of time
played), I have decided to make this thread.

Before you read any further:

1. Please do not post a reply in this thread unless you read the entire OP.
2. This OP is very long, and will take a while to read. For your own
convenience, prepare a sandwich and some coffee before reading. Take short
breaks while reading and take-in every section of information before you move
on to the next one.
3. While I try to keep most of this article objective, there is going to be
some opinion-based information. If you disagree with it and would like to
comment on that, please be specific and say exactly what part you disagree with,
and take the time to write a well-explained reason and counter-argument. Avoid
being vague or else it will be unclear what you mean.
4. Please stay on topic and don't derail this thread. If you're quoting someone,
make sure you stay on the main topic and don't start wandering off into some
sub-topic. To be clear, the main topic is: making a skill-based rating system.

Hopefully this thread will be insightful for all reader, and answer some
questions. I'll get straight to the point and cover the main one:

Q: Is it possible to make a skill-based rating system?
A: Yes. However, it won't be perfect - but it can be very close to perfect.

I'll place little bold lines with numbers to help you read this in parts if you want.
Again, please do not post any reply until you finished reading the whole thing.

------------01------------

As it stands right now, there is a pseudo player-made skill-based rating system.
In other words, if you ask a group of players to take a look at two players,
watch them play, and say which one is better - the players will be able to tell
you that "Player A generally plays better than Player B".
This system has two problems:
1. It's not automatic.
2. It's not accurate.

If we were to create an automatic skill-based rating system, that would rate
player skill with a number - I'll call this number SR from now on - the way to
know if this automatic system works would be to compare the SRs it gives
various players. If for every two players A and B that we take a look at, the
community says that A definitely plays better than B, and A's SR is higher than
B's SR, then we can conclude that the system works correctly.

Once proven to work, the system will shed insight that mere human observation
cannot reach. For example, if there are two players where the community can't
tell for sure who is better, but one of them has a slightly higher SR, than the
community will have to accept that the player with higher SR is slightly better.
Furthermore, this system will be automatic, so you won't have to poll people's
opinions every time you want to compare skills. The SR will always be
up-to-date for each player.

In other words, an automatic skill-based rating system should basically do the
same as asking experienced players how good is each player, except it would fix
the two problems of asking players:
It would be automatic.
It would be accurate.

------------02------------

* SR = Skill Rating. A number that indicates a player's skill in osu. Can be
compared against other players' SRs to know which player is more skilled.

Before we go on and create the skill-based rating system, its nature dictates
that we must understand what "skill" is. If you don't have a solid definition
for what "skill" means in osu, then there's no way you can make a system that
rates players by their "skill".

Is the score you get on a map related to skill? Is the accuracy you get on a
map related to skill? Is the amount of maps you can play before you get tired
related to skill? Is there a difference between the first run on a map and
repeated runs on a map in relation to skill?

Regardless of what skill is, one thing is clear: the only way to measure skill
is with information from playing osu. Nothing else is relevant to your skill in
osu. The only times when you can judge how good a player is, is the times when
he plays on a map, and you would judge it based on information from his results
on that run.

Lets think about all of those points. When you observe two players, is it
possible that one would get a lower score on a map, but you'd still consider
him more skilled in osu? It's possible that a player would miss a note in the
middle (because of a long repeating slider for example) but do perfectly on
almost the entire map, and get a bad score, while another player would do poorly
on the entire map, but manage to hold a longer combo and end up with a higher
score. Would the score help you indicate who played better here? No.

Conclusion: score on a map is not related to skill. Because of that, a skill
based rating system should not take scores into account.

- with this said, please keep scores as they are. They're exciting and fun,
and managing to get a full clear on a map you could barely pass is a nice
reward for your effort. Even that this shouldn't get involved in skill
calculation, don't just abandon the current scoring system. I think it's fine.

Next, if two players play the same map and get more or less equal results on it,
but one player has played it for the first time just now, and the other one
played that map 50 times before that run, who would you consider more skilled?
I believe most people would agree that the one who did better on the first run
is more skilled than the one who needed to practice the map to do well on it.

Conclusion: skill rating should take into account how many times you played on
a map while rating you for your plays on it.

- this means that for a skill-based rating system to exist, the game has to
keep record of how many times the player has played each map. The good news is
that the game already does this.

Next, what if two players play a series of maps in a row without breaks and
get various results on them, and do slightly better or worse during different
times of their sessions. Should the time they have been playing affect the
skill calculation of their plays?
Some might argue that the longer you play in a row, the more you get tired, and
it becomes harder to keep doing well. Others might say it has absolutely no
effect how long have you been playing. Some others can even say that as you go
on and play, you warm up and start doing better towards the end.

Conclusion: the amount you've been playing is unrelated to skill. Even if it is
related somehow, the relation is very vague and there is no good way for a
system to objectively rate you by it, so it shouldn't be included in skill
calculation.

- this was just one example to show that, there are many factors that some
people might consider that they're related to skill, while others would not.
Since the point is to create an objectively accurate system that everyone would
agree on, all vague and unmeasurable factors should be ignored.

Finally, the accuracy you get. This is the factor I believe is most closely and
objectively related to player skill in his run through a map. Regardless of
what score you get, or which parts you played better or worse, if two players
play the same map for the first time and one of them gets a higher accuracy
than the other, then it's evident that - at that run - the player with the
higher accuracy played better.

------------03------------

The goal in osu is to hit everything correctly. There will be obstacles, such
as jumps, stacks, streams, direction change, etc, but in the end - they're just
different levels of difficulty in keeping you away from hitting everything
correctly. If you manage to hit everything correctly, you get 100% accuracy.
For every inaccuracy in hitting a note, regardless of whether it was because of
a jump, or a stack, or whatever, you lose accuracy equally.
In the end, the final accuracy you get is what dictates how well you did.

Lets sum this up:

For the purpose of skill-rating, each player would have an SR. His skill rating.
This SR would be automatically changed every time the player finishes playing on
a map.
The factors that would affect how much his SR changes would be:
1. How many times has he played that map before*
*note: I'll later also introduce the importance of how long ago was the last
play on a map by that player.
2. How much accuracy did the player get on the map

Would this be enough, tho? No, there is one more important factor to take into
account, which is in fact the most important one:
The map difficulty.

If two players play a map for the first time, and get the same accuracy, but
one player played a harder map than the other player, whose SR will get higher?
Obviously, the player who played the harder map would get more SR.

WHICH BRINGS UP THE FOLLOWING KEY POINT

A skill-based rating system cannot exist until osu is able to rate how difficult
a map is. Until the game is capable of telling apart every two maps in the
aspect of "which one is harder" correctly, the game won't be able to correctly
rate which player is more skilled.

The current map difficulty rating, the "star rating", is far from accurate.
Not only that it is not accurate, but it is also limited... it only goes up to
5 stars with 0.5 fractions. This is not enough to evaluate skill!

In order for automatic skill-based rating to exist, the game has to be able to
rate each map's difficulty with a number that is unrestricted by limitations
(large range variable), and this number would affect your SR after your plays.

There is going to be a large section in this post dedicated to explaining how
to correctly rate map difficulty. I will go over an algorithm that should rate
map difficulty very closely to perfection. However for now, lets put this aside
and assume this problem is already solved.

In other words, lets ASSUME for now that each map already has a certain MD
number attached to it, that is calculated by osu once every time the map is
saved in Edit mode.

* MD = Map difficulty. A variable attached to each map that indicates how hard
is it, or in other words, how much skill does the map require. MD will be
accurate enough to determine for every two random maps you pick which of the
two is harder. (higher MD = harder map). MD will be used for SR calculation.

------------04------------

So, lets recap:

Skill-based rating system would rate player skill based on their plays on maps.
Every time a player finishes playing a map, the following variables will be
available:
1. The player's current SR.
2. The map's MD.
3. The accuracy that the player got on that run on the map.
4. The amount of times that player has previously played the map.
5. The amount of time that passed since the last time that player has played
this map, if he played it at least once before.

Again, #5 is a new point that I mentioned but haven't covered yet, I'll cover it
shortly. Also, #1 is a point I was yet to mention - the way that skill based
rating would work, the SR addition or reduction you get from your plays on maps
will depends on your existing SR. For those familiar with ELO rating, this is
going to work pretty similar. It's the world's most proven skill-rating system,
and frankly it just makes sense; I'll cover it shortly later for the osu
perspective.

As far as last time played goes (point 5), playing a map a lot of times in a
row creates an element on memorization that eliminates the reaction time
requirement of maps, in other words, it affects that skills that are required
to get a good accuracy on the map. Regardless of whether it gets harder or
easier over time, it gets different, so skill-calculation must take note of
this.
Now, if you didn't play a map for a long time, you will eventually forget it,
and the memorization versus reaction ratio will once again lean towards reaction
instead of memorization. Because of this, osu should record when was your last
play on a map, and use (CurrentTime minus LastPlayed) as a variable in the
skill calculation system, as a factor for the AmountOfPlays variable.

------------05------------

At this point, here is where we stand:

> I have to introduce an algorithm to correctly rate map difficulty, or in
other words, skill requirement. This is vital for the skill calculation.

Other than this, everything else is in place. Lets assume this system exists
and give a few examples on how it works, to explain why existing SR affects
SR gain or loss from map runs.

Each player should start with a certain SR once he starts playing osu. The
actual number doesn't matter, so lets say 1000. When you create an account, it
has 1000 SR. Every time you play, this SR will either go up or down.

The harder a map you play is, the faster your SR will go up if you do well on
it, and the slower it would go down if you do badly on it.
The easier a map you play is, the faster your SR will go down on doing poorly,
and the slower it would go up if you do well on it.

The more times you have played a map recently, the slower of an SR gain you will
get from doing well on it, and doing poorly on it won't be affected by this.
If you played a map a lot of times, but not recently, the slowdown on the SR
gain by playing it will be more marginal.

How well you do on a map will be purely dictated by your accuracy. Every other
factor, such as score for example, will be saved for other ratings that can be
fun for people to follow, but they will be unrelated to the by-skill rating.

So basically, at this point, those are the entire foundations for the skill
based rating system. The actual factors and numbers you apply here don't matter,
as long as they're consistent within themselves (the only consistency you need
is between SR gain and SR loss).
Whether you end up with a system that gives player A 7213 SR and player B 6102,
or player A 1447 and player B 1413, the system will still equally spread the
entire playerbase on a graph from "most skilled" to "least skilled".

------------06------------

Before I go into the map difficulty rating issue, there are a few points left
to cover:

1. Skill decay over time
2. High level players playing easy maps

Lets say an SR system is already in place. And then a player with a very high
SR, who normally gets very good scores on very difficult maps, suddenly decides
to play an easy map, and of course he gets SS on it on his first try. Should
this increase his SR, at all?
If the SR system works correctly, then, it will barely increase his SR if at
all. He'll have to play 1000 [Easy] maps to barely get 1 extra point to his SR
if he's a player who can normally breeze through insanes.
Not only that, but if he ever gets below 100% accuracy on those easy maps,
it might actually hurt his SR and lower it slightly.

The result would be: players would be discouraged from playing maps too easy
for their level. Whether this is good or bad is up to you to decide, but it
does pose a big problem: once all maps are correctly rated by difficulty,
players will know which map is the hardest. Once a player beats the hardest map
perfectly, everything else he plays will only potentially reduce his SR, so
if he doesn't just "play for fun", but rather plays for rank, he'll have no
reason to keep playing.

This is just one corner case to demonstrate a big issue that comes up with this
system across the entire graph: players would be discouraged from playing osu
if the system perfectly rates their skill. It would make it too pressuring and
not fun to play many random maps or random levels.

Because of this, you must introduce SR decay. Basically, over time, the current
SR you have will decrease and go down until returning to the default (whether
it was zero, or 1000 like I suggested, or whatever number is the default).
Make this decay start after a week, and end after a few months. For example,
start a decay of 5% per day from the moment you haven't played for a week.

Note that 5% per day does not mean 100% loss in 20 days, it will chip away from
your SR for 5% per day at that point of time, so it will never actually return
all the way to "zero", it will just slowly close in on it.

Since the decay starts from the last time you played, playing more maps would
be useful to keep your SR high, even if they're below your level. Also, due to
the nature of skill calculation, it would be more rewarding to keep playing
newly released maps, instead of replaying an old hardass map you beat in the
past that you already have some runs on. So basically, this solves the issue.

However, a perfect reset isn't good enough. If you just reset the decay startup
to zero every time you play any map, then players would only need to play one
map per week to keep their SR from decaying. Instead, make playing a map belate
the decay startup time, but not fully. Apply a factor.
For example, lets say the factor is 50%. It's been a week since I last played,
and now I play a map. After I played it, the week of "time since I last played"
will not reset to zero, but be reduced to 3.5 days, which means my decay will
start in 3.5 days. When I play another map, it will be reduced to 1.75 days,
which means my decay will only start in a little over 5 days.

With a 50% factor you only need to play about 3 maps a week, so maybe it's
better to use a much smaller factor, to encourage people to play more maps to
avoid losing their skill rating which they gained from beating hard maps.
It kinda makes sense, too. You keep playing to prove you're still solid.
If you quit playing for two months, the system would display you as not so
skilled anymore, which makes sense because you probably also started being worse
due to not playing for so long.

And if you quit for several months, the moment you start playing again, if you
just return to playing maps in your old level, you'll quickly regain the SR
you used to have, so you don't lose anything from this decay (unlike in a
rate-by-time-played system, where losing points over time would be a huge
annoyance because you'd need to spend a ton of time to regain points lost, in a
truly skill-based system the points would come back instantly as long as you
still have the skill to deserve them).

Also, you can apply a different decay-startup-pushaway factor for first runs on
maps and repeated runs on maps, to encourage players to play new maps. In other
words, make re-playing maps you already had lots of runs on have little effect
on pushing away your skill decay, at least, much less than playing new maps
that you haven't played before.

So, hopefully the importance of decay is clear now. It kills two birds in one
stone - it encourages players to keep playing maps of all levels, and trying
out new maps - while keeping the skill rating board up to date, and pushing
players who stopped playing down the board by gradually reducing their rating.

------------07------------

Here is the big moment you've all been waiting for: MD how?
How do we rate map difficulty? If we just solve this missing link, a perfect
automatic skill-based rating system can exist.
Before we go any further, let me say this: You can already try out the skill
based rating system to see how it works, by taking a small sample of maps and
giving them player-made number ratings of difficulty, and then just let players
play on those maps and see what results they get, compare this output to
opinions of spectators who saw them play, and see how accurately the systrem
rates people by skill.

However, for this system to truly be autonomous, osu itself has to be able to
numerically rate map difficulty. So, lets create this algorithm now.
Side bonus: once osu can rate map difficulty correctly, you can replace the
current star-rating with an improved star rating that's based on the MD numbers,
and then have correct star ratings for all maps to be used as map level
indication when you download/try out maps.
1star=easy, 2star=normal, 3star=hard, 4star=insane, 5star=veryXinsane,
won't that be nice?

------------08------------

So,

The best way to write a good algorithm is to collect a lot of information. I'll
cover this and guide you through it, to convince you in the correctness of it,
and to explain how it came to be.

Mission: to find out what makes maps difficult.
What we know: the goal in osu is to hit everything correctly, i.e. to get a
300 on every single note in the map and to spin as much as possible.
Everything that keeps the player away from getting any 300 is a difficulty.

So lets start by finding all of the things that keep people away from getting
300s. The first one is very obvious: the [Overall Difficulty] map property.
The higher it is, the less of a frame window you have to hit notes for a 300,
and thus this increases the map difficulty.

What about circle size, approach rate, and HP drain? Those are more complicated
in how they affect skill requirement, and I'll cover them over time as we go on.

Things that are known to add difficulty:
1. Jumps - movement between two points that is faster than the common movement
pace for that section of the map. Reason for difficulty: inconsistency
(different movement speed), speed (jumps are faster than regular movement),
cursor accuracy (the end of the jump - gets harder due to speed, since you have
less time to place your cursor on the correct spot).
2. Faster beats - doesn't nessecarily have to be in a stack, every time the
map suddenly changes from 1/1 into a series of 1/2s, or from 1/2s into a set
of 1/4s, etc, this adds difficulty. Reason for difficulty: inconsistency
(different clicking speed required), speed (need to click faster).
3. Patterns - sections in the map that repeat themselves. This can either
increase or decrease difficulty. For every time there's a very close pattern
repetition, the difficulty of it is reduced, because you know what to expect.
However, when the pattern happens partially and then breaks, this increases
the difficulty, because it's essentially a consistency trap.
4. The BPM of a map - in the end, no matter how you look at it, faster is
harder. Reason: speed (need to hit notes faster and move faster).

Note however that, something general like "BPM" isn't good for correct
difficulty calculation. You need to use raw information. By this I mean,
the actual time (in program loop frames) between actions. Similar to this,
every measure related to speed of clicking and cursor movement, should be
measured by program loop frames and not by musical-term numbers.

Also, it's important to take into account how the human mind works when you
consider this raw information. There are some breakpoints to consider here.

1. The human reaction time is approximately 1/4 of a second. On a 60FPS game,
this would translate into 15 frames, however osu does not have a set FPS,
however it still uses some sort of timing clock, so time should be evaluated on
a raw program scale and not on musical-term scale.

2. Humans can only process up to five items in one pattern. If you see six
items, your brain processes them as either 1+5, 2+4 or 3+3. However, you can
recursively see patterns inside patterns. i.e. a star of triangles, your mind
will process that as a 5pattern of five 3patterns. However if the triangles
are inconsistent with each other you will not see a 5pattern at all.

Patterns are an important part of difficulty, as difficulty basically comes
from three things:

1. The speed you need to do things
2. The accuracy level you need to correctly do those things
3. How easy is it to predict what to do
(4. length of a map)

#3 is decided by pattern creation and breaking during maps.
For example, lets say that for a large part of a map, notes appear more or less
in a line, with only under 60 degrees angle of a different between the movement
line you're required to do with the cursor from one note to another, and then
suddenly a direction change happens and a note appears at a stop that is more
than 60 degrees angle away from your current line. This specific movement will
be harder to do, even if it's not a jump and not a faster note, simply because
it doesn't follow a pattern your mind was following.

In order to have an MD rating for all maps that is consistent between different
maps of all types, osu has to be able to:

1. Rate how hard is it to accurately hit a 300 on a map. In other words,
measure the total frame window for hitting a 300. For example, if this is
tested in a 120FPS loop, there might be a global 6-frame window to hit a 300
per note, this would affect the MD.

2. Rate how fast of an action-taking the map requires you to do. This breaks
into several parts, as there are two actions you do: clicking and moving the
cursor. So basically, there's going to be a rating for clicking speed, and a
rating for movement speed. Note that difficulty isn't a consistent thing for a
map, some parts in the map can be harder or easier than others. You must make
an average out of all click speed and movement speed sections.

i.e. if I need to hit three notes: A...B...C
there are two speed sections and two movement sections here: the first "..."
and the second "...". Movement and clicking are related, and those factors
should be mutltiplied for exponential factoring.

Also, movement speed should be considered as movement between the middle of
one note to the edge of the next note (you can count this as: length between
the middles of both points, minus the circle radi of point2, which is consistent
for the entire map anyway, so basically just minus the circle radi). This is
where circle size will affect the difficulty rating, but not the only place.

Speed of clicking and movement, and accuracy of hitting a 300 - those are
consistent skills related to execution that are not affected by memorization,
although you could argue muscle memory affects this. Either way, they are not
skills related to reaction time.

Patterns and pattern break, though, is directly related to reaction time.
Patterns exist in two flavors: timing patterns and movement patterns. Both
should be taken into account seperately, although they still affect each other.
Since the game is able to play itself on Auto, I don't think it'll be a problem
to add a factor of info-gathering for auto. Basically, collect patterns.

For timing: collect timing sections. Place a limit on it to conserve memory.
Since humans can't process over 5 items simultaneously a limit wouldn't lose
much from real life difficulty; a limit of 15 for timing would be enough.

So for example, for a map with 20 notes, the patterns to try to find would be:
1 through 2, 1 through 3, 1 through 4, etc, until 1 through 15 (limit).
Then 2 through 3, 2 through 4, etc, until 2 through 16.
Once this list of timing patterns is saved, compare them against each other,
and find repetitions. If anything repeated itself exactly twice, it's a
questionable pattern, and if anything repeated at least 3 times it's a definite
pattern.

(as you're already guessing, this comparing is a N^2 loop on two lists, so
pressing "Save" while Editing a map would be time consuming... to avoid this,
simply have an option to "Save without MD calculation", which just saves the
map without calculating its difficulty. You'll only need to calculate a map's
MD once you finished mapping it anyway)

------------09------------

As for movement, the following factors affect patterns:
Type-1: movement speed - this pattern gives you a clue on where is the next
note going to appear, in a circle area at a specific radi from your position.
For every time the next note actually appears in that range, MD is reduced,
while for times it appears too close or too far, MD is increased, at that is
a pattern breaker.
Type-2: angle range between movement lines - this pattern requires long term
memory, you can place a limit of something like 15 on this too. Basically,
constantly remember the average, min, and max angle of difference between the
last 15 (~ or whatever limit) notes, and use this to predict the angle for the
next upcoming movement. When movement follow the recent consensus, MD is
lowered, while when movement suddenly behaves inconsistently, MD increase.

There is also the visual factor to take into account. When notes are stacked
on top of each other, patterns have bigger impact on your ability to either
300, or fail to 300, the stacked notes, especially when the circle approach
rate is below human reaction time.

NOTE THAT, the circle approach rate is never below human reaction time.
HOWEVER, it becomes that, once notes appear faster than human reaction time.
i.e. a stack of three notes with 1/8 timing difference appears in a 200+
BPM song. You're only going to be able to react to the approach circle of the
first circle, the remaining two you will time by pattern recognition, not by
individually reacting to them. At times when this pattern breaks, i.e.
you had several points with fast stacks and suddenly a slow stack appeared that
has the same shape - this kind of thing increases MD, as it's a pattern break.

This might seem trivial when the whole thing is above human reaction time,
but on very fast maps that have below human reaction speeds, patterns like this
are make-or-break and they test your reaction time to patterns.

------------10------------

Ok this is a big of a mess. Lets recap:

For osu to correctly calculate MD per map, it needs to do the following:

1. Rate the speed of the map in non-music terms based on non-program loops
between note hits.
To elaborate on this: you can't simply check the average from start to end.
You need to do a full collection of timing differences and keep them in an
array. i.e.
1/8s difference between two notes: 8 of those
1/4s: 215 of those
1/2s: 119 of those
1/1s: 23 of those
2 ticks: 1 of those
3 ticks: 7 of those
5 ticks: 1 of those
etc
And then take into account what was the most common timing difference, for
example 1/4s, and have the raw loop time of THIS speed requirement have the
biggest impact on the average calculation of the map's speed, while corner
cases with little appearances can be completely ignored as they will usually
be either breaks or a few slow sections in a map that don't truly indicate
the general speed of it.

2. Rate the speed of the movement the map requires. Since players can choose
whatever window size they want, the pixels-per-second doesn't really matter.
DPI is dynamic too. What does matter as far as distance go is movement
relative to circle size. i.e. if I need to move 20 pixels between two circles
with 5px radi in 1 second on one map, or 40 pixels between two circles with
10px radi in 1 second on another map, their movement speed is THE SAME.

3. Rate the level of pattern repeat and pattern break based on both timing
and movement patterns, by going over timing sections and comparing them, and
going over movement speed/direction and spotting inconsistencies.

------------11------------

Things that happen during one approach circles: this is the last difficulty
issue I'd like to talk about.
I hope I'm not the only one who thinks that the "Easy" mod actually makes maps
harder. This is for a good reason.
In the area of time between the appearance of one note and then another,
while you get full reaction time to the first approach circle, the second one
receives less reaction time. As long as you still didn't hit the first note,
you can keep building a queue of timing differences in your head based on
the showup of notes, but once you hit one note and have to start moving on to
the next one you enter a situation where you have to both empty a queue and
add items to a stack at the same time, which is problematic. Every pattern
inconsistency inside this realm has more impact than outside of it.

In other words, the way that MD addition from pattern breaks adds to the map's
MD will be affected by the amount of notes that appear during the timeframe of
the approach rate. Or in other words, sometimes approach rate will actually
make a map harder. If the approach rate is too slow for the map's timing speed,
then it has essentially the same effect as Hidden.

As long as approach rate is higher than borderline human reaction time -
which is 1/2 of a second - it doesn't make a map any easier, so it can only
potentially make it harder. If things inside approach rates happen faster than
human reaction time - which is 1/4 of a second - this adds difficulty too.

Again, this only refers to a case where enough things happen to have stuff
showing up after you already hit a first note of a queue of hits you had.
It's a common case but you need to address it correctly as far as finding each
time it starts and ends and a queue is restarted.

------------12------------

There's a few more things I wanted to talk about but I want to take a break
from writing this as it already got too long. I hope most of this made sense to
you so far.

To sum everything up:

1. A skill rating can exist. It will rate players every time they finish playing
a map. The factors it will use to rate them and either increase or decrease
their current SR would be:
1 - Their current SR.
2 - The map's MD.
3 - The accuracy the player had that run.
4 - Amount of times previously played on that map.
5 - Time of last play on that map versus the current time.

2. There has to be an SR decay system with fast decay, to keep SR tables up
to date with players' activity and to encourage players to keep playing.

3. To create a correct MD for maps that will be useful for SR calculation, the
MD has to be a large range variable, and it should be decided based on raw
speed/accuracy information, not based on musical terms. It should also take
into account patterns in the map, and increase difficulty for pattern breaks
while reducing or just not affecting difficulty for pattern maintaining.

------------13------------

Ok, one more thing I forgot to mention: apply a factor to SR change from map
completion. i.e. if the MD and accuracy of a run I did is supposed to give me
2500 SR, and I currently have 2000 SR, I wouldn't jump to 2500 instantly.
I would get maybe 10% from this run, so my +500 would only give me +50 and I'd
move to 2050. This is to avoid flukes or random mishaps from having drastic
effect on players' SR. Your SR will slowly crawl towards the number you truly
deserve as you play more, and if you have random good or bad runs along the
way they won't instantly break your SR. Pretty similar to ELO? Simpler tho.

------------14 (final)------------

I hope this was insightful for anybody who read this far.
It would be nice if osu had a true skill-based rating system, and I believe
this is the closest you can get. If anybody thinks you can get any closer than
this, please tell, because I'm curious.
It's not too hard to implement this, but there's a lot of corner cases to cover
and factor-effect testing to do to apply this, so it would take at least a few
months to create this.

Would you support a system like this? Do you like this idea?
Do you think this can rate player skill well?
Can you bring up any example where this would rate player skill incorrectly
and give one player a higher skill rating over time than another player?

Note how this system is barely affected by the amount you play, and just playing
a few hard maps will instantly shoot your SR higher than players who play a lot
of easy maps, but at the same time this system still encourages you to play a
lot to truly reach the edge of the SR you can get on your play level.

Good? Bad? Feedback please -> gogogo

i believe you forgot 1 very important thing
mods
if i play a map with doubletime and hidden and get an accuracy or 95%
yet another person plays the same map with no mods and gets 100%
it's obvious the player who played with mods is more skilled (only in extreme cases is this not true)
so there needs to be an algorithm that also takes mods into account

I believe a system like this is what is needed to take osu! to the next level, great job you obviously put a lot of thought into it

Thanks, I really forgot to mention mods. I also forgot to talk about spinners.

I'll write a more elaborate convention for them later, but right now, what you can do with them is pretty straightforward: keep the current percentage bonuses and don't think about it too much. Although this is not very accurate, it can work as a temporary solution until this can be properly analyzed.

What this means is, for mods: if I'm supposed to get 2500 SR for SSing a map, and I have 2000 SR right now, normally I would get +50, but if I played the same map with the doubletime mod I'd get +12% to 2500 which turns it into 2800, so I'd get +80 from this run. If my with-DT run was worse than my no-DT run, and would only have gotten me to 2400 SR, the +12% bonus still makes this worth more than the higher accuracy on a modless run.

The fact is, though, that mods affect skill requirement for maps very differently. For some maps for example, hidden barely adds any challenge at all, while for other it makes them much harder. Same for HR and DT. If you really want to make it fair and correct to skill, then instead of calculating one MD, a map would have to calculate several different MDs - one for each mod, where it would take into account stuff like timing property changes from HR or DT, and take reaction to pattern time difference from Hidden into account. This would make the algorithm even more complicated to write than right now, though, but this would drastically increase accuracy for mod ratings.

Another approach would be to only include HR and DT in skill rating scores, and keep Hidden and FL runs in a separate rating table of their own, since they change the entire nature of skill requirement for map completion. In other words, keep two tables for skill:
1 - normal / hr / dt / hr+dt
2 - hidden / fl / hidden + fl (with any hr/dt variations you want)

As for spinners, to calculate skill from spinning, simply use percentage of spin done by the player relative to auto's spinning or any arbitrary above-player-ability spinning level as a bonus to SR gain from the run. This is to motivate people to actually spin a lot and not just clear spinners. Although frankly, I don't know how much you can say skill is involved in doing long spinners. It's fun :p but it doesn't take a lot of skill regardless of map difficulty. So any additional spin beyond spinner clear should be a very marginal or negligent addition to skill rating calculation.

Epic tl;dr

I guarantee my system is more accurate.
I guarantee that no matter how long you make your explanation of a potential system, mine will work better as I have hands-on experience with data.

If you want to throw your time and effort somewhere, take a look at the current AiMod difficulty conditions and work on improving them (hopefully in less words). Unfortunately I can't link the page I wanted to with information on this, since it was nuked in the previous wiki iteration. I'll work on getting it up soon.

Well done, seriously.
The issue with the mods is pretty obvious. Also it's pretty hard to automatically rate patterns. But this may be done by calculating the distance spacing compared to the slider velocity and the coordinates of the hitobjects.

Whatsoever. I would not seperate FL and HIdden from HR and DT (many people play all stuff together, so how would you rate that?)
But right before the "DT is harder than FL" discussion begins, I'd suggest that the factor being used right now is appropiate enough to influence the SR at all.

Even if DT may have a DIFFERENT kind of difficulty than FL may have, the skill can be valued equal (same goes for Hidden and Hardrock). If you think, Hidden is pretty easier than Hardrock, you should try playing syncopic maps, where Hidden gets a pain and Hardrock may makes stuff easier and don't add a single challenge because the AR is very low and Circle Size is 1.
The current factors given to the mods are appropiate imho, they may require different kind of skills, but their value is the same. (I hope you get what I meant)

We should spread this thread in order to may get this game and ??? ranking to the next level, maybe revamping the whole ??? stuff again with also an renewed Star Difficulty rating.

peppy wrote:

I guarantee my system is more accurate.

No offense, and maybe the ??? has still weird algorithms that turn out to have weird results and still needs adjustments because it's still in dev-phase. But - no. Right now, it's not.

I guarantee it is. I guarantee the above system would give you even more warped and unreliable results. My system will also naturally improve with improvements to AiMod difficulty criteria once they are completed and implemented. About half of the text-wall in the initial post could be split into discussion of that, since it's focussing on classifying the difficulty of a beatmap, not how to rank players.

Problem is its very hard to test a system like the aqo's
So I suggest taking points from this system that are good and implementing them on the ??? system
One issue that needs attention though is map difficulty, I wonder if a non overkill solution even exists

A non-overkill solution does not exist, and that's one area I agree with the OP. It is why I began the AiMod difficulty criteria movement, which has specific rules that class maps into difficulties. It still needs more refining, so is not yet in use.

I don't believe anything above with regards to how to rank players is superior to what I have in place. For now, you'll have to trust me on that one .

Like Aqo stated, it is essential for rating the skill of the player, that it is required to clarify what "skill" actually means. And to know what it means to "pass a hard map", you need to make the game actually recognize what "a hard map" is.

I can just hope that with further development of your AiMod and also of the messed up star difficulty rating, your ??? rating will be more appropiate.
From now on, the weight of Normal/Easy maps with nomods and the influence of Accuracy and high-mapranks are too high.
I don't know how your algorithm works at all (and I doubt you want to release that, so I won't ask any further), so I won't call it crap (like it has been done before).

But I hope that if you give the ??? actually a name and put it officially onto the site, the stuff really evaluated the skill correctly. Because right now, it makes me cry. But I respect it's state of being in development.

I see this thoughts of Aqo more than a idea pool of stuff you may havn't taken into consideration, yet. (no one is free of errors)
You should not screw the current ??? if you think your's is better, but you should at least read it and maybe you get the one or another idea how to improve AiMod and the ??? rating at all.

You forgot to take slider velocity and slider tick rate into account for calculating MD.

Even if the above seems kinda reasonable to me I'd still got something to complain about:

What is skill?
You are saying score is irrelevant to skill, right? Reason: One might screw up at one single note and ruin a combo, even if you are better.
Then I'll give you my answer: That's what "retry" is for. The game osu! has some very basic and well defined concepts. One of them is that score equals skill. Accuracy does equal skill indirectly aswell of course, that's how it's defined in osu!.

The flaw of the current ranking system is that it rewards farming ranked maps without mods over actually getting top scores on many maps, not anything about "skill". Just check Lunah's recently posted video of the community-voted top20.
They are all capable of getting insane top40 ranks.

No offense to you, but ??? takes exactly this approach of correctly classifying those ranks and setup a more or less correct ladder and wins at least for me as a new ranking metric.
Also peppy doesn't say "I guarantee it" when he can't. I'm pretty sure he got his reasons for being so confident, he knows more than anyone here about all that data after all.

To peppy: First, thanks for taking the time to read and reply. Second, I haven't heard of the "AiMod" yet, so I'd like to check this out.
As for your confidence in the accuracy of your system over what I suggested, I'd love to have access to some of the data you're speaking of to experiment with. I'm guessing you don't want to share the inner workings of your system right now, and it's unclear if you will share it in the future...
So let me bring up a question: why not share how your system works? Is there any reason you believe it's a good idea to hide the development process of it from the players? If there is, what is it?
My own personal belief is that sharing information with more people will help collect insight and come up with a more refined solution. If you disagree, I'm very interested in hearing your counter argument.
It's not that I don't trust you :p I'm just curious; and I want the best for osu's future, I love this game.

Thanks for the replies everybody else I don't really have something refined to reply, but thanks for bringing up some issues; you're right about that one Tshemmp; I tried to keep my post as broad as possible and summarize the key points so a lot of minor details like that were left out.

I plan on sharing how it works once it is finalised, to an extent. Exact numbers will not be shared, but you will know how to increase your score. I believe it is not exploitable so it should be fine to give near-full details once I'm ready.

As for the AiMod difficulty system -- you can already see it in action from the editor, but the implementation details of it are no longer available (wiki got deleted). I will make them available on the wiki again in the near future. I'd appreciate if you could focus your efforts there since I agree with much of how you classify the difficulty of maps.

Showing the details of the metric will invite people to figure out how to exploit it.

Well Tom to be fair he did say he forgot to take mods into account, so don't judge the system based on that yet, refining it may solve your issue

And also if what we are measuring is pure skills here, then accuracy is the most important deciding factor, why?
Say I get 3000 combo on a map with 93%, accuracy after 50 plays, if another person gets a near perfect accuracy but only missed on one note making his score far lower than me, and on his first play as well, who would you say is better?
The retry thing you're talking about is a measure of hardwork, we already have a perfect system for that

peppy's system even though incomplete has proven to work quite well though, so I have no problems at all

Inviting people to try to exploit a system openly is a good thing, as it helps you spot exploits faster and thus come up with a solution to them faster.
Peppy, you avoided the question of "why do you prefer to hide the development process from players over share it", but I guess you just don't want to go there so I'll leave it at that. I'll keep lurking around to see info on the AiMod when info shows up again.

To the OP: Why do you think it's more skillful if a player beats a map in less tries? Maybe the guy was drunk, maybe he was messing around, maybe he was trying it with flashlight or something. This kind of game is ALL about trying and retrying maps over and over all day. That's all you do. If there was a system that based skill around attempts you'd have people making brand new accounts to 1 shot SS insane maps for extra points. Which is bad, obviously. I also love your avatar <3

The development process is trial and error, and specific test cases I'd rather not reveal at the chance of offending the users involved. Exposing it would be horribly messy for both me and you. I've been through at least 50 iterations to reach what I have currently. I am yet to attack taiko/ctb, so there's some way to go.

Unkind: if you were messing around, then you kinda have to accept that your skill-based rating would drop because of it. In the end, if two players play seriously, and one of them takes 100 tries to have a similar run to another one who had a run like that on his first try, it's undeniable that the latter is generally better at osu, at least from that map's indication. It's not like you're not allowed to mess around, but you kinda have to accept the results of it as far as ladder rank goes. If you don't mind playing around, then you wouldn't mind losing some on some fictional number that doesn't affect your enjoyment from the game, right?

-A t H e N a- wrote:

Well Tom to be fair he did say he forgot to take mods into account, so don't judge the system based on that yet, refining it may solve your issue

And also if what we are measuring is pure skills here, then accuracy is the most important deciding factor, why?
Say I get 3000 combo on a map with 93%, accuracy after 50 plays, if another person gets a near perfect accuracy but only missed on one note making his score far lower than me, and on his first play as well, who would you say is better?
The retry thing you're talking about is a measure of hardwork, we already have a perfect system for that

peppy's system even though incomplete has proven to work quite well though, so I have no problems at all

The perfect system you're talking about is perfect if you got the time to retry every single map and rank on it. So far only cookiezi and whitewolf came close to that.
Nearly all top ranks in the current system are just farming S in ranked maps. At a not so high skill level you can do this with very little retrys already.

Regarding your other matter, that is a matter of opinion to me. I personally wouldn't know any case where happened what you mentioned. I for my part would gladly replay a map once if I get 99% and just one miss while the score w/o the miss would be an easy high rank. After all the guy with 93% accuracy and 50 plays ovbiously got a rank aswell, else he wouldn't even qualify for comparison.

Also, who would decide the borderline... at which accuracy is nomod SS vs XX accuracy DT better? 95? 92? Never? Who knows...

Tom, if you think you'd need to play all the maps to get a good score with my system then you have misunderstood how it works. The very nature of my system allows you to get a high rating by just playing a very few difficult maps, without having to go over most of the maps available, that would have almost no effect on your score very quickly.

I must admit i was exaggerating to get my point across, but such cases do actually happen, no proof but I recently watched a friend of mine get 16xx combo on world's end with hidden and 98% accuracy only to fail to get a top rank because of 1 miss, I would say he definitely has the skills for it, but sometimes luck isn't on your side and things like this may prevent you from ever ranking high, by basing the system on accuracy his score wouldn't matter and since he used hidden he would definitely get a high SR for that map.

peppy wrote:

I guarantee my system is more accurate.

I hope so.

Aqo: I had written a huge post, but in the end it would come down to how you implemented it.

If you gave someone a small bonus for beating the map the first try, good, huge bonus, very bad. And don't penalize people that take longer, good. As long as you don't make the bonus too ridiculous it wouldn't matter.

In the end this would still probably just result in people playing the map 2000 times with spunout beforehand lol. (ugh)

Rename "Spunout" to "Play Unranked" ^_^
If people secretly practice a map offline until they can do well on it and get a high SR on their first ranked run of it, the truth is, all of that practice time will really make them better at osu, so they'll deserve whatever they get. If they keep doing this for every single map they play, they will actually get really good at the game and deserve all the SR they get, and if they only do it once in a while, it won't have too big of an effect on their SR due to partial/gradual increase per map played. So basically this doesn't work as an exploit... I don't think it's an exploit to practice being better, even if it might slightly bend the percentages in your favor.
In the end, it's really just a small bonus like you said it should be.

Aqo wrote:

Tom, if you think you'd need to play all the maps to get a good score with my system then you have misunderstood how it works. The very nature of my system allows you to get a high rating by just playing a very few difficult maps, without having to go over most of the maps available, that would have almost no effect on your score very quickly.

Not talking about your system though.

-A t H e N a- wrote:

I must admit i was exaggerating to get my point across, but such cases do actually happen, no proof but I recently watched a friend of mine get 16xx combo on world's end with hidden and 98% accuracy only to fail to get a top rank because of 1 miss, I would say he definitely has the skills for it, but sometimes luck isn't on your side and things like this may prevent you from ever ranking high, by basing the system on accuracy his score wouldn't matter and since he used hidden he would definitely get a high SR for that map.

Then let me give another example. Me getting 99% on Airman would be worth the same as cookiezi getting an SS and fcing?
Having 95% on Big Black with a 200 combo is worth the same as the same accuracy with a 500 combo?
Let's just assume both people tried an equal amount of times.

Using accuracy is a good idea, I can't deny that, but 100s giving 33% and Misses 0% is just a too little difference of 2 totally different things in osu!. I'd support your idea a lot more if misses decreased your accuracy by a more fair amount. That couldn't be called accuracy anymore though.
Also, I know that 99% of the maps require an FC to rank on, however in my opinion those mostly are the maps which prove way less skill than the "real deal".

As far as accuracy rating worth goes, accuracy goes from up to down, not from down to up.
i.e.
Getting 96% versus 92% isn't ~4% better, it's /twice better/ (in the skill-calculation for ratings). By the same logic, the difference between 99% and an SS is huge.

Yes, getting a few random small combos on The Big Black can take the same skill as getting one big combo on it. Just because you were less lucky on the times you accidentally let go of a repeating held slider and ended up with poorer combo doesn't mean you played worse than somebody else who was more lucky in the timings of his misses.

Aqo wrote:

99% and an SS is huge.

You are contradicting yourself with that. So someone who had "bad luck" in his run and got just one 100 will get way less skill points? Wasn't that what you wanted not to happen?

Aqo wrote:

Yes, getting a few random small combos on The Big Black can take the same skill as getting one big combo on it. Just because you were less lucky on the times you accidentally let go of a repeating held slider and ended up with poorer combo doesn't mean you played worse than somebody else who was more lucky in the timings of his misses.

Let's get this realistic. You don't play those maps once. I took about 50 tries for getting a high combo. And if somebody else gets a combo lower than even half of my combo with the same amount of tries, you are seriously telling me he is equally good?
Even with 5 to 10 tries it would be clear to me who is better. Not rythm-wise but definitely movement-wise.


If my example for Cookiezis airman record was not appropriate due to the 100% let's take this:
Someone can 98% on high ODs (like 9 and 10) with HD and HR. However he got serious issues with certain jumps in a map and just doesn't manage to FC it, no matter how hard he tries.
If that guy takes 100 tries and manages to get a run with say 5 misses. Wouldn't someone who got a 95% FC with in his 5th try get a lower amount of points in your system?


I am saying it again: Accuracy rankings won't in any way be a metric of calculating movement skill, because misses drop your accuracy way too less than mere 100s.
This is a rythm game, but you can't deny that hitting the circles is just as important as pressing the buttons at the proper time.
While your system does a great job at scenarios where everyone in the top40 can FC a map it is performing poorly on the extremely hard maps imo.

tom94 why are you using me as an example?

Holy mother of tl;dr

ragelewa wrote:

tom94 why are you using me as an example?

I actually thought about writing your name in specific, but wanted to keep it general in the end. XD
Also you're even getting above 99% with HD HR, GRRRRR.

Tom94 wrote:

Aqo wrote:

99% and an SS is huge.

You are contradicting yourself with that. So someone who had "bad luck" in his run and got just one 100 will get way less skill points? Wasn't that what you wanted not to happen?

Aqo wrote:

Yes, getting a few random small combos on The Big Black can take the same skill as getting one big combo on it. Just because you were less lucky on the times you accidentally let go of a repeating held slider and ended up with poorer combo doesn't mean you played worse than somebody else who was more lucky in the timings of his misses.

Let's get this realistic. You don't play those maps once. I took about 50 tries for getting a high combo. And if somebody else gets a combo lower than even half of my combo with the same amount of tries, you are seriously telling me he is equally good?
Even with 5 to 10 tries it would be clear to me who is better. Not rythm-wise but definitely movement-wise.


If my example for Cookiezis airman record was not appropriate due to the 100% let's take this:
Someone can 98% on high ODs (like 9 and 10) with HD and HR. However he got serious issues with certain jumps in a map and just doesn't manage to FC it, no matter how hard he tries.
If that guy takes 100 tries and manages to get a run with say 5 misses. Wouldn't someone who got a 95% FC with in his 5th try get a lower amount of points in your system?


I am saying it again: Accuracy rankings won't in any way be a metric of calculating movement skill, because misses drop your accuracy way too less than mere 100s.
This is a rythm game, but you can't deny that hitting the circles is just as important as pressing the buttons at the proper time.
While your system does a great job at scenarios where everyone in the top40 can FC a map it is performing poorly on the extremely hard maps imo.

In regards to te player with five misses, you yourself said he took 100 tries to get five misses, that would impact his score heavily because of the amount of tries he required, meaning there wouldn't be such a large gap between their sr, but frankly if he did the map with hr hd, (I'm assuming that because you said he was good with it), then of course if he gets a higher accuracy he would have a better sr than the other player

Sorry if that's not what you mean and I misunderstood you

He's talking about a player that can get good accuracy at hd+hr but can't play jumps at all even without mods

Then he would get a low accuracy on maps like that, since he can't even hit the notes on extreme jumps, an he would do badly in terms of accuracy for the less demanding jumps, a person who can FC it 95% would definitely have a higher accuracy

Meaning this person who can't do jumps would get a low sr for this particular map, but since sr doesn't depend on 1 or 2 maps only at the same time it doesn't depend on playing every single beatmaps in existence, he could easily make up for it with hd hr high accuracy ranks on other maps. This is why the sr system is very fair in my opinion, atsunjin it seems the onlly downfall to this is on stream maps, since you lose much more accuracy from not being able to stream well(consistent huge rows of beats versus plenty but individual beats) but I don't think the difference is too much, especially since I'm a mouse only player used to seeing low acc on very fast streams

-A t H e N a- wrote:

Then he would get a low accuracy on maps like that, since he can't even hit the notes on extreme jumps, an he would do badly in terms of accuracy for the less demanding jumps, a person who can FC it 95% would definitely have a higher accuracy

Meaning this person who can't do jumps would get a low sr for this particular map, but since sr doesn't depend on 1 or 2 maps only at the same time it doesn't depend on playing every single beatmaps in existence, he could easily make up for it with hd hr high accuracy ranks on other maps. This is why the sr system is very fair in my opinion, atsunjin it seems the onlly downfall to this is on stream maps, since you lose much more accuracy from not being able to stream well(consistent huge rows of beats versus plenty but individual beats) but I don't think the difference is too much, especially since I'm a mouse only player used to seeing low acc on very fast streams

No, you are just wrong. There are players getting higher accuracies while not being able to FC certain maps. Also you are not per definition doing bad in terms of accuracy if you barely manage to clear jumps, that's a totally wrong assumption.

Let's just take mittens as an example.

Perhaps I took moelewa's statement of not being able to do jumps at all too directly, since everyone at the top at least has a certain degree of skill with them, otherwise any jump however small would either make them miss or lose a lot of accuracy, my mistake sorry.

Seems I'm not all familiar with aqo's system, I'll just wait for his response so that I won't need to fear making a fool of myself.

i can mash buttons at high BPM
(kokou no sousei [insane])

but im not very good at insane jumps

(gold dust)

how would that rank me compared to someone who is the other way around?
which track is "harder"?
same with streams. i really cant stream longer than 5-7 notes over 185 bpm (only with luck) so a song with 200 bpm streams is impossible for me, yet someone can easily stream 200 bpm while not even being able to clear gold dust. how about that?

for all intensive purposes trial and error to figure out the pseudo-player-based system is probably the best solution, which peppy is doing.

I guarantee anything you explain/propose could be resumed on a single word: ladder.

In the end, it doesn't matter if a player is good at jumps or at streams. You have to think of it in the simplest way possible: the goal in osu is to hit everything for a 300. Some maps will make getting those 300s hard because of jumps and others will make it hard because of stream speed, but in the end it's just different difficulties for the same goal. You don't need to compare whether jumps or streams are harder, just let the system work out who played better per map and obviously each player will have their own strengths and it'll be represented in map-specific ranks.

The accuracy does indicate one thing: How close you were to getting 300s on all notes in the map. Regardless of whether you're strong in stream or in jumps and weak in the other, the acc you get per map shows how well you were able to play /that one map/.

As far as losing more accuracy for not being able to stream over not being able to jump... three 100s are equal to one miss in terms of accuracy loss (from a 300), right? If you think about it like this, it's not really that different. If you completely get a 12stream in 100s it would equal to failing 4 jumps - which is just 5 notes to travel between. And frankly even if you get 100s in a stream you'll usually still 300 most of it, even if you're bad at streams.

Again, as far as the whole "how would it rank people good at jumps vs people good at streams"... it would just rate everybody fairly, regardless of what they're good at, based on how good at they at *osu* in general. It's as simple as that. Osu requires both jumps and streams. Some maps focus more on one than the other. I already explained why ranking depends heavily on what maps are available, so that's the deal basically.

Wishy22 wrote:

I guarantee anything you explain/propose could be resumed on a single word: ladder.

We get that it's good, problem is that it's a pain in the ass to implement, and would require much more "Internet" or bandwidth or something, dunno the term but it costs money dude

Wishy22 wrote:

I guarantee anything you explain/propose could be resumed on a single word: ladder.

How can a game like this benefit from a ladder? Why does playing a single player game at the same time as someone else make it somehow more significant than when you play at the same time or 20 minutes apart or 2 years apart? This isn't a competitive multiplayer game, it's a single player game with online ranking functionality.

That being said, it might be a cool mode or something. But it wouldn't in ANY way determine who's more skilled just because they can do better in an isolated instance.

edit: "ladder" is pretty vague, i'm assuming you mean something like how starcraft 2 is setup, with ladder maps and matchmaking etc, if not nevermind.

Wishy22 wrote:

I guarantee anything you explain/propose could be resumed on a single word: ladder.

Not with current combo based scoring.

I suggested adding this as a new play mode in the ??? Thread, with separate rankings, I doubt this will ever be implemented though, and I dot really think its necassary, seems a little overkill as well once you consider that you aren't affecting your opponent in anyway(you are competing, but you aren't really playing againts each other) in this game unlike other games where this system is used.

The original ranking system was really bad, and anything else seemed like a better idea to me at that time lol

-A t H e N a- wrote:

I suggested adding this as a new play mode in the ??? Thread, with separate rankings, I doubt this will ever be implemented though, and I dot really think its necassary, seems a little overkill as well once you consider that you aren't affecting your opponent in anyway(you are competing, but you aren't really playing againts each other) in this game unlike other games where this system is used

exactly.

Now if peppy wanted to come up with some super awesome mode where you can somehow affect your opponent based on how well you play, then it would have an actual reason to exist.

Well, I think even this score-based ranking system tells us who is good player enough.
How? Look;
1-) Your score, you get more score then others.
2-) Your accuracy, it says how you good at playing map correctly.
3-) Play count, that means experience you get.
4-) Your rank count, SS, S and A ranks are shown in your profile.
no one works?
Watch how people played and get scores in maps at #40 list with these 4.

but I don't say you are wrong or something, I just think this is enough and we don't know what is ??? yet.. (I don't)