It's been know for a while that the engine will provide compensation for a team in the 2nd half if it determines they have largely underperformed in the first half (or at least this is how I understand it). My question is if anybody knows what metrics/factors the engine uses to deduce whether or not a team is underperforming...such as team ratings in comparison to their opponent, or team statistics in comparison to that of the opponent, or something completely different?

Posted by nachopuzzle on 3/6/2014 6:06:00 PM (view original):
It's been know for a while that the engine will provide compensation for a team in the 2nd half if it determines they have largely underperformed in the first half (or at least this is how I understand it). My question is if anybody knows what metrics/factors the engine uses to deduce whether or not a team is underperforming...such as team ratings in comparison to their opponent, or team statistics in comparison to that of the opponent, or something completely different?

it doesn't work in the way you are suggesting. rather, it uses the true expected value of certain outcomes to assess performance. it is not relative to the other team, either. let me give you an example:

say you have taken 20 2 point shots this game. the engine clearly has to know the chance of the ball going in on each shot - for example, the first shot you might have 55%, so it creates a random number from 0-1, and if the number if under .55, you make the shot. now, after 20 2pt shots, suppose the expected value was 10pt shots made (50%), but you have only made 5 (25%). now, on your 21st shot, you will get some benefit to help the outcome move back to the norm - to reduce the chance of extreme variations. say on shot 21, you have a 45% chance, well the engine may now give you the make if the RNG produces anything under a .5 (50%).

the numbers here are, of course, hypothetical - but the mechanism is not. this is how feedback is introduced, and although their implementation may vary from what is described here in the details, it all boils down to the same thing.

somewhere, you can find a list from seble of some things (it may be complete or not) to which feedback applies. these are independent - evaluated in their own right. i don't know if they are on a team or player level, in some cases, it may be one or the other, or heck, even both (i kind of doubt its both). fouls and shooting are at least two areas to which feedback applies.

I tried to find the information in the old developer chats that influenced how I interpreted this compensation, but some of them don't have any of the conversation logs when you click on them for some reason, so I couldn't find any direct reference to it.

What you said makes sense, and that would be a good way of going about it, but it seems like you could play a numbers game to take advantage of these corrective measures.

It's a dumb thing to have implemented, in my opinion.  There shouldn't be any corrections for streaks, overperformance, or underperformance within games - the RNG events should be discrete.  The sample of events is just too small to extract any reliable benefit even if you believe there should be feedback.

Posted by llamanunts on 3/7/2014 10:51:00 AM (view original):
It's a dumb thing to have implemented, in my opinion.  There shouldn't be any corrections for streaks, overperformance, or underperformance within games - the RNG events should be discrete.  The sample of events is just too small to extract any reliable benefit even if you believe there should be feedback.

i mostly agree with the first, although i think it does help reduce complaining from users (to some extent). also in real life close games in theory often get far apart and seemingly usually converge back, because of the nature of an over performing team to take their foot off the gas, or whatever you want to call it. so i think there is some realistic basis. that said, im not really a fan...

on the second part, i do think there are strategies that work slightly differently with feedback. uptempo becomes more meaningful to allow feedback to play in over a longer term, slowdown becomes more critical against inferior teams, to reduce the impact of feedback. also, i believe shooting may work on the individual level, not the team level, which suggests having high scoring quality players shoot a lot might be more effective today than it used to be (i suspect it is), because that will help your team deviate to their normal level of play. i suppose this is only good for high end teams, low end teams might feel the reverse (hoping for the fluke upset). volatility is enemy #1 for top teams though so i certainly think there are ways feedback could play into the strategy of those teams (as well as for others, to a lesser extent). 

Great explanation gillespie!  I've been thinking about this effect on some of my bad beats (especially on the tempo front).  Query: could you clarify your meaning that a team, or individual player (I suppose), might deviate to their normal level of play?   I'm not sure whether it's the reality of the message or the phraseology that is sticking in my craw.

That is exactly what I mean by being able to take advantage of the numbers game...teams with a few very good scoring options could theoretically just run uptempo, chuck a trillion shots, eventually underperform, then have then engine compensate for them precisely when it matters (and possibly even moreso against good defenses precisely because they must run into that threshold). Of course, such a team would also have to have a good defense to make sure the other team can't run away with the game in the process.

Ironically, it's the inverted notion of bing ball.

Posted by gillispie1 on 3/7/2014 11:00:00 AM (view original):

Posted by llamanunts on 3/7/2014 10:51:00 AM (view original):
It's a dumb thing to have implemented, in my opinion.  There shouldn't be any corrections for streaks, overperformance, or underperformance within games - the RNG events should be discrete.  The sample of events is just too small to extract any reliable benefit even if you believe there should be feedback.

i mostly agree with the first, although i think it does help reduce complaining from users (to some extent). also in real life close games in theory often get far apart and seemingly usually converge back, because of the nature of an over performing team to take their foot off the gas, or whatever you want to call it. so i think there is some realistic basis. that said, im not really a fan...

on the second part, i do think there are strategies that work slightly differently with feedback. uptempo becomes more meaningful to allow feedback to play in over a longer term, slowdown becomes more critical against inferior teams, to reduce the impact of feedback. also, i believe shooting may work on the individual level, not the team level, which suggests having high scoring quality players shoot a lot might be more effective today than it used to be (i suspect it is), because that will help your team deviate to their normal level of play. i suppose this is only good for high end teams, low end teams might feel the reverse (hoping for the fluke upset). volatility is enemy #1 for top teams though so i certainly think there are ways feedback could play into the strategy of those teams (as well as for others, to a lesser extent). 

First part of the first part: Sure, I guess I can see it reducing complaining from some users, and increasing it from cranky grouches like myself.

Second part of the first part: Sure, realism.  I have two thoughts on that.  First thought about the second part of the first part: I think that might happen enough anyway, per the discussions about streaks occurring without needing to be programmed in.  Second thought about the second part of the first part: If the spreads were reined in some, it wouldn't be that bad... but winning by 40 against a 77 point spread can hurt your ranking, position on the projection report, blah blah more than I feel like it should.

On the second part - I meant benefit to HD, not ways for a coach to extract an advantage.  I see your point about that.

Posted by gillispie1 on 3/7/2014 12:26:00 AM (view original):

Posted by nachopuzzle on 3/6/2014 6:06:00 PM (view original):
It's been know for a while that the engine will provide compensation for a team in the 2nd half if it determines they have largely underperformed in the first half (or at least this is how I understand it). My question is if anybody knows what metrics/factors the engine uses to deduce whether or not a team is underperforming...such as team ratings in comparison to their opponent, or team statistics in comparison to that of the opponent, or something completely different?

it doesn't work in the way you are suggesting. rather, it uses the true expected value of certain outcomes to assess performance. it is not relative to the other team, either. let me give you an example:

say you have taken 20 2 point shots this game. the engine clearly has to know the chance of the ball going in on each shot - for example, the first shot you might have 55%, so it creates a random number from 0-1, and if the number if under .55, you make the shot. now, after 20 2pt shots, suppose the expected value was 10pt shots made (50%), but you have only made 5 (25%). now, on your 21st shot, you will get some benefit to help the outcome move back to the norm - to reduce the chance of extreme variations. say on shot 21, you have a 45% chance, well the engine may now give you the make if the RNG produces anything under a .5 (50%).

the numbers here are, of course, hypothetical - but the mechanism is not. this is how feedback is introduced, and although their implementation may vary from what is described here in the details, it all boils down to the same thing.

somewhere, you can find a list from seble of some things (it may be complete or not) to which feedback applies. these are independent - evaluated in their own right. i don't know if they are on a team or player level, in some cases, it may be one or the other, or heck, even both (i kind of doubt its both). fouls and shooting are at least two areas to which feedback applies.

Assuming you're assessment is correct, then the more I think about it the more I have to disagree with what I said about "this being a good way to go about it if you were gonna do something like this".

Why is even necessary to have "the true expected value of certain outcomes to assess performance, it is not relative to the other team either," in the first place? Why not a "true expected value of certain outcomes to assess performance, that is relative to the other team"? Hypothetically, giving a 1 ATH, 1 SPD, 1 LP, 1 PER, 1 BH, and F IQ player an expected chance of .55 when attempting a 2 foot  seems rational...however, if a 99 ATH, 99 SPD, 99 DEF, 99 BLK, and A IQ player is guarding them, then it is actually more unrealistic to give them an expected outcome of .55 to start averaging back towards.

And this part is just my personal opinion that is in no way verifiable, but I constantly see back-up players on bad teams teams go crazy, who seeming have a lower perceived chances of scoring against my back-ups then their starters do against my starters. Which is another reason why I think it is based on a team reaching some sort of criteria for it take affect.

Posted by rogelio on 3/7/2014 11:22:00 AM (view original):
Great explanation gillespie!  I've been thinking about this effect on some of my bad beats (especially on the tempo front).  Query: could you clarify your meaning that a team, or individual player (I suppose), might deviate to their normal level of play?   I'm not sure whether it's the reality of the message or the phraseology that is sticking in my craw.

sorry, i was sloppy - replace deviate with converge :) i was used to stating the inverse (reducing deviation) and failed in my attempt to convert to the positive form, or whatever you call it.

Posted by nachopuzzle on 3/7/2014 12:15:00 PM (view original):
That is exactly what I mean by being able to take advantage of the numbers game...teams with a few very good scoring options could theoretically just run uptempo, chuck a trillion shots, eventually underperform, then have then engine compensate for them precisely when it matters (and possibly even moreso against good defenses precisely because they must run into that threshold). Of course, such a team would also have to have a good defense to make sure the other team can't run away with the game in the process.

Ironically, it's the inverted notion of bing ball.

just to be clear, if this running uptempo chucking up a trillion shots strategy is not good in terms of expected value, feedback is not going to help you. however, yes, having a setup like this will cause your variance to reduce - you are going to be more likely to hit your expected performance. this not only reduces upsets against you, but also upsets you would manage yourself - so it cuts both ways. unless, of course, you are the best team - then it only cuts one way :) or the worst, i suppose (i like to be optimistic here though so lets stick with best ;)

also - isn't this like, the exact nature of bing ball? having a couple guys throw up a trillion shots, particularly when they are guards, is bing ball. what do you mean by its the inverted notion of bing ball?

Posted by llamanunts on 3/7/2014 1:53:00 PM (view original):

Posted by gillispie1 on 3/7/2014 11:00:00 AM (view original):

Posted by llamanunts on 3/7/2014 10:51:00 AM (view original):
It's a dumb thing to have implemented, in my opinion.  There shouldn't be any corrections for streaks, overperformance, or underperformance within games - the RNG events should be discrete.  The sample of events is just too small to extract any reliable benefit even if you believe there should be feedback.

i mostly agree with the first, although i think it does help reduce complaining from users (to some extent). also in real life close games in theory often get far apart and seemingly usually converge back, because of the nature of an over performing team to take their foot off the gas, or whatever you want to call it. so i think there is some realistic basis. that said, im not really a fan...

on the second part, i do think there are strategies that work slightly differently with feedback. uptempo becomes more meaningful to allow feedback to play in over a longer term, slowdown becomes more critical against inferior teams, to reduce the impact of feedback. also, i believe shooting may work on the individual level, not the team level, which suggests having high scoring quality players shoot a lot might be more effective today than it used to be (i suspect it is), because that will help your team deviate to their normal level of play. i suppose this is only good for high end teams, low end teams might feel the reverse (hoping for the fluke upset). volatility is enemy #1 for top teams though so i certainly think there are ways feedback could play into the strategy of those teams (as well as for others, to a lesser extent). 

First part of the first part: Sure, I guess I can see it reducing complaining from some users, and increasing it from cranky grouches like myself.

Second part of the first part: Sure, realism.  I have two thoughts on that.  First thought about the second part of the first part: I think that might happen enough anyway, per the discussions about streaks occurring without needing to be programmed in.  Second thought about the second part of the first part: If the spreads were reined in some, it wouldn't be that bad... but winning by 40 against a 77 point spread can hurt your ranking, position on the projection report, blah blah more than I feel like it should.

On the second part - I meant benefit to HD, not ways for a coach to extract an advantage.  I see your point about that.

i don't think winning by 40 or 80 makes a difference in HD on projection report or anything - there is zero impact from where im standing. playing a bad team that you should beat by 80... yes, that can hurt you :) thats true whether you crush them or... absolutely pulverize them?

anyway, its not like feedback causes you to beat a team you should beat by 77 by 40. rather, it makes it more likely you actually beat them by 77. its not about the spreads, it doesnt affect the EXPECTATION of how much you beat a team by. it does affect change in the deviation from that expectation. so instead of beating this team by 40, 60, 80, 100, and 120 in successive games, maybe now you beat them by 50, 70, 80, 90, 110. this is more appealing in a sense, despite the ridiculousness of the example :)

also i disagree with the premise even more when you say you meant benefit to HD. its debatable if the effect is a benefit or detriment - i don't like it - but see positives and negatives. however, i don't believe there is much question about the potential for the impact to be meaningful. you said, sample size is too small to extract any reliable impact (i changed benefit to impact in case that is an issue). this is not the case - there is PLENTY of time for feedback to kick in within a single game. its all about how hard you want it to kick - seble made it decently moderate - but just from the old eyeball test, its obvious its there. i personally think its too strong when its obviously there to the naked eye. this is pretty typical though, i often complain about imbalances and then complain some more that seble is too heavy handed. i think he fails to recognize the subtle nature of the game to competitive users... like when he made big men more effective by... increasing their fg% by about 10% across the board :O i don't think feedback is quite as severe but maybe half the impact would have been more appropriate? its hard to say, given that i instinctively recoil from the concept of feedback in a game like this in the first place.

IIRC, in regards to point differential impact toward projection report, it caps at 20 points. So a 50 point win has the same effect as a 20 point win.