I’m a d1 coach and I have a decent team. I run zone/flex with normal pace most of the time. I keep getting beat the first half and the running away the second half.. I know I’m not wearing teams down.. any advice?

Posted by Indianajoinz on 7/10/2019 1:59:00 PM (view original):
I’m a d1 coach and I have a decent team. I run zone/flex with normal pace most of the time. I keep getting beat the first half and the running away the second half.. I know I’m not wearing teams down.. any advice?

Seems to be built into the engine to normalize actual results toward the expected result. I see this all the time in my games, and there have been other threads on the topic.

If you're upsetting your opponent at half, you'd rather be up by 9 than 11, because speculation is if you're up more than 10, the correction kicks in extra hard.

Posted by kcsundevil on 7/10/2019 3:03:00 PM (view original):

Posted by Indianajoinz on 7/10/2019 1:59:00 PM (view original):
I’m a d1 coach and I have a decent team. I run zone/flex with normal pace most of the time. I keep getting beat the first half and the running away the second half.. I know I’m not wearing teams down.. any advice?

Seems to be built into the engine to normalize actual results toward the expected result. I see this all the time in my games, and there have been other threads on the topic.

If you're upsetting your opponent at half, you'd rather be up by 9 than 11, because speculation is if you're up more than 10, the correction kicks in extra hard.

the engine has feedback in a handful of areas, to reduce the odds of extreme events. in real life, emotions and/or discipline and other things along these lines, maybe fear of your coach actually murdering you if you manage to foul out in 2 minutes of play time - it can limit the severity of outcomes. at a minimum, there is massive impact on team scoring, we all see that in how often a team down 10-20 in the 2nd half (which would suggest they should lose by even more, because they are getting crushed thus far) comes back and at least makes it close.

anyway, based on people complaining about bad results and making arguments along the lines of the above (i believe some stats were run and deviations between HD and CBB were shown), seble added this feedback stuff to the engine one day. it had a big impact - a major one really. it took the press from defunct (fatigue was made to be absolutely brutal) to a prize set again (well, fatigue was also adjusted around that time, to the more moderately increased impact i believe we still have today). the reason it was so important for press is players committing fouls was one of the factors, and extreme foul trouble in the press is so devastating.

one thing though, the 9 vs 11 thing, i don't know where that came from, and i've never heard it - but can pretty much guarantee its false. its documented (not sure where) what things are impacted by feedback; its a known quantity, seble told us. i don't recall all what it is, committing fouls is part of it, so is scoring, and there is other stuff. but it is all player level stuff, i believe - not team stuff. actually, press got shafted pretty hard early in potential because the impact of fatigue got way jacked up (first by an insane amount, then scaled back to a lot), and press became sort of awful. but the limiting of foul randomness helped make it dominant again for truly elite teams. i think there are thresholds or something, like an 0-2 guy may get no boost, but a 3-10 guy does. we were never told 100% how that stuff works, but i can say it had a very substantial and noticeable impact on reducing randomness at the game level (even though the impacts were all on single players - the sum of the reduction in crazy RNG helped moderate crazy games where typically many individual factors had totally bonkers RNGs), and on the player level.

to the OP - its possible its just luck - that you are getting really bad luck in the first half. in the old engine, you might have returned to normal luck for half 2, but still lost. but in today's game, you are more likely to return to above-average luck (from the standpoint of where your players should normally 'perform'), so its way easier for better teams to come back now than it was. i am not sure the half boundary actually has anything to do with anything, though - it might - but i think really there are just feedback loops introduced into equations and they take effect sometimes in the first half, sometimes in the second, it just depends on the particulars of the circumstances. im not sure anything changes from the last second of half 1 to the first of half 2 (assuming nothing happens, like your 0-6 guy is now 0-7), but im also not sure it doesn't. however, i do think the most constructive way to think of feedback, it to think about it having an impact over the course of the game - from whenever there is significant data to 'feedback on', if you will.

There are lots of different theories on how the programming nudge works, gillispie. I bumped a recent-ish thread on this for your reading enjoyment.

Posted by kcsundevil on 7/11/2019 12:28:00 AM (view original):
There are lots of different theories on how the programming nudge works, gillispie. I bumped a recent-ish thread on this for your reading enjoyment.

thanks! i'll check it out when i have more time before i'll be asleep :) i read a few posts, its a good topic, couple good posts (just from what i read, i get the sense it may devolve later - hoping some of it stays on track though, feedback is really interesting to me)

my favorite halftime turnaround was years ago under the old engine.

D2 game in Allen between two equally matched teams. One team jumped out to a 44-2 halftime lead. It must half been quite a halftime as the team trailing by 42 points at half, came back strong taking the second half by a 52-12 margin. Just ending up short in the comeback attempt with a final score of 56-54.

Posted by kcsundevil on 7/11/2019 12:28:00 AM (view original):
There are lots of different theories on how the programming nudge works, gillispie. I bumped a recent-ish thread on this for your reading enjoyment.

so basically, i just want to clarify - how this programming nudge works is less theory than a known quantity. somewhere on these forums, it was explicitly explained by seble how he made feedback work. it affected about 5 different areas, such as an individual player's shooting and an individual player's fouling (which i'm 99% sure are two of the actual examples). i kind of forget the other 3.

there is no first-half vs second-half type measure, it just doesn't work that way. basically, all that is happening is this - i'll explain via example because i'm not really prepared to explain in an absolute sense. suppose you have player X who is a starter and plays a good bit (just so the example makes more sense). what happened originally was, every time he took a shot, a % chance of making said shot was calculated, eventually, at some point in the process (the actual flow for shooting is weird, but that's not important - what matters is eventually, all the factors get mashed together into a simple % chance of making it - everything from the shooter's ratings to the defender's ratings, the systems played, the quality of the PG on the shooting team, etc, etc).

so, what happened next was, an RNG call was made saying ok, give me a value from 0-1, it could be 0.15 or .523 or .694251 or whatever. doesn't really matter. then, if % chance of making is > RNG return value, you make it. otherwise, you miss.

so, player 1 on his first shot, he might have had a really open look, and a 70% chance to make it! the RNG is called, it comes back .62, and the guy makes it. then, on shot 2, he's really well defended and the clock is really late, so he throws one up - 24% chance to make it. RNG is called, comes back .1, and he makes it again. next shot is a normal shot, average defense, hes a 53% chance to make it - RNG is called, comes back .7, he misses. and it went on like that throughout the game. his previous makes or misses had nothing to do with the current shot.

none of the core workflow described above was changed, but something was added. basically, the game is now tracking expected value vs actual value, on his shooting, and introducing feedback to cause a reversion to mean (to cause a reduce in variance, i.e. to reduce the frequency crazy **** happens, like guys fouling out in 3 minutes or going 13-13). there are a few ways, from a technical / statistical / programming standpoint, that one could accomplish this - but they all have the same effect and the details more or less, IMO, are meaningless to us as coaches. anyway, here's roughly what happens now. this gets tracked along the way:

player | expected makes | actual makes
shot 1 - .7 chance to make, shot made (doesn't matter if it was a .01 RNG or .69)
tracking - player 1 | .7 expected makes | 1 actual make

shot 2 - .24 chance to make, shot made
tracking - player 1 | .94 expected makes | 2 actual makes

shot 3 - .53 chance to make, shot made
tracking - player 1 | 1.47 expected makes | 2 actual makes

it carries on like this, tracking the expected makes vs actual makes throughout the process. at some point in the process (this may be from the beginning, it is not known), when the game goes and calculates those %s of making it - it will adjust based on the expected vs actual makes. let's say the engine does this after 3 shots on a player (it might be 0, it might be 5, but its somewhere in there).

shot 4 - would-be 49% chance to make, but because he's slightly exceeding his expected value, maybe the engine adjusts this (feedback) to 48%. RNG comes back, .2, shot made.
tracking - player 1 | 1.95 expected makes (you use the 48% that was really used, not the 49% that was based on the fundamentals, before feedback) | 3 actual makes

shot 5 - would-be 60% to make, but because hes moderately exceeding his value, let's make it 57%. RNG comes back, .5, shot made.
tracking - player 1 | 2.52 expected makes | 4 actual makes

it carries on like this. likely, the engine does more the more data it has. so suppose some time later its like this

tracking - player 1 (10 shots) | 5.2 expected makes | 9 actual makes
then -

shot 11 - would be 55% to make, but because hes way ahead of his expected value, change it to 35%. RNG comes back, .5, shot missed. this outcome was altered by the feedback.

so, that is basically how this works. it could be wrong, because seble has been wrong in explaining things before, but based on my own experience and observations, i am pretty confident it works this way (it could be confirmation bias, but i tend to have limited faith in the admins, so i sort of doubt it).

anyway, the outcome of all of this is that as time goes on, in a game, the opportunity for the results to be really skewed goes up. 2 shots in, the results just cant be that skewed. but by 10, they can be. so in a game which is really skewed, in the last 5-10 minutes, you can get some pretty weird results. on the whole, it does make the box score look 'better', as in, probably closer to real-world deviations, and more palatable to the users, both. real life has human emotions, which in some cases causes extreme events, but in other cases, avoids them. for example, its hard to try to crush your opponent when you are kicking the crap out of them and are up 15, as we see in like, half of all the NT games every single year. the engine does kind of a ****** job of emulating that effect, but its hard to model human emotions, so i think we should cut them some slack. also, they added feedback way after the game was built - you'd really want to plan something like this up front. so, without redoing the whole game, seble had to find a hack to approximate whatever it is he actually would have liked to do, and this is it.

some folks see this as wild swings from half 1 to half 2, but that is just because the results are presented to us in 2 halves. if they were presented in 4 quarters, we'd probably be like well, that 3rd quarter can get weird, but the 4th is where things really can get extreme. there are definitely more first vs second half swings, in this game, than there were before feedback. its just not really anything to do with the halves, as much as there are continuous effects in place which can build up over time, and thus, the distorting effects tend to be more noticeable in the late game.

just to note though, there were tons of first vs second half swings in the old game. its not like its changed drastically. the base RNG allows for a TON of randomness, and in HD, there just aren't that many coin flips. if you take 30 shots in a half, and should make 15, its relatively easy to make 10 or 20. do one of those one half, and the other the other half, and it really doesn't matter if feedback is in play or not - the two halves will look totally different. its just the way it is. real life seems to have a way of smoothing those curves out, at a macro level (game-wide), and so all HD is trying to do with feedback, is to approximate that. it just so happens that HD does it on a per-player level more than a per-team level, which is probably what would have been more real life like.

the net effect of all of this, it does make it harder to pull big upsets. if you were 50/50 or even 70/30 to 30/70, feedback doesn't really change the odds, in a material way. when you are 90/10, it might be 92/8 now. i don't think its more severe than that, if anything, i think its less. this isn't a huge factor, but its meaningful, is how i'd put it.