I think Gil's last post is pretty important. At the end of the day, this is a simulation game that is based on real world stats. No real world team goes into a game with 100% odds of winning. It might be 99% or 98%, but the other team has a percentage of a chance to win just by playing in that game. This simulation has to account for that which means unusual circumstances can hit even if you did everything right. When you do get that "bad luck" sim, it sucks and you won't be able to learn much from it like you might in a real world situation given our limited control of play-to-play coaching, but that doesn't mean it shouldn't exist in the simulation. All we can do is put our team in the best position to swing those odds as close to 100% as we can. Will we lose some? Sure. Just look at Top's unexplained loss to a bad sim. But it also sounds like he won a lot more games than he lost that season. So by and large he did put his team in the best position for success. This is similar to recruiting. We expect to win every roll we have the best odds in. Will that happen? No. But in the long run it will even out.

Of course, if the upset stats don't align with real world stats, then there would be an issue. But I haven't necessarily seen that when it comes to sims.

How can you be a 20+ year player when your account was made 15 years ago? I don't even think HD existed 20 years ago

Posted by Basketts on 10/24/2022 10:14:00 PM (view original):
I think Gil's last post is pretty important. At the end of the day, this is a simulation game that is based on real world stats. No real world team goes into a game with 100% odds of winning. It might be 99% or 98%, but the other team has a percentage of a chance to win just by playing in that game. This simulation has to account for that which means unusual circumstances can hit even if you did everything right. When you do get that "bad luck" sim, it sucks and you won't be able to learn much from it like you might in a real world situation given our limited control of play-to-play coaching, but that doesn't mean it shouldn't exist in the simulation. All we can do is put our team in the best position to swing those odds as close to 100% as we can. Will we lose some? Sure. Just look at Top's unexplained loss to a bad sim. But it also sounds like he won a lot more games than he lost that season. So by and large he did put his team in the best position for success. This is similar to recruiting. We expect to win every roll we have the best odds in. Will that happen? No. But in the long run it will even out.

Of course, if the upset stats don't align with real world stats, then there would be an issue. But I haven't necessarily seen that when it comes to sims.

In recruiting, there’s a cutoff at which point an upset is not possible. You have to be in range to expect to have a chance. And beyond that we have stretched odds, so the upsets that do happen look bigger than they actually were (a 75-25 battle is closer to 60-40 in terms of effort credit). The sim accounts for all the “luck” stuff in the attributes, and if we could trust that they are straight sims (they aren’t), there would still be plenty of upsets - most would be attributable to *some* coaching decisions, but there would still be some oddballs.

But there should be no question as to whether there is a “lucky day” battle roll component driving the direction. If it exists, it needs to be explicit, and the power needs to be very low.

Posted by sinatra on 10/25/2022 12:11:00 AM (view original):
How can you be a 20+ year player when your account was made 15 years ago? I don't even think HD existed 20 years ago

andrewfrance in Naismith has been a member since 2001, so I'd say it's just over 20 years old

If we want to talk about a hidden element that should be investigated, it’s the rubber band effect or whatever you want to call it that guarantees if my team played well above average in the first half that they will lay an absolute egg in the 2nd.

im not saying we know any of this for sure, but i do want to say... i agree with shoe that we should know if there is a 'lucky day' roll driving the direction, but also, CS has repeatedly and consistently said there is not. i think their position would be that there is none and that this is public, confirmed knowledge.

to baums's point, its really remaining user skepticism about the existing RNG and the impact on upsets, that caused seble to add feedback in the first place. basically, there were allegedly only 'non-artificial' upsets, and this still was too much randomness for folks. so seble added feedback, which is sort of analogous to how in recruiting, you can't have a 90/10 battle. in recruiting, its a hard cutoff, and in game sims, its a soft directional shift, making extreme upsets less likely, but with no hard cutoff.

its definitely a matter of opinion how well this all worked out. my take is that 1) there is no 'lucky day' roll by design. 2) feedback was moderately successful in its goal of 'increasing reversion to mean', if you will, in reducing crazy upsets. 3) overall, feedback is not something i'm a huge fan of and i definitely think there could be some bugs in there that screw things up. notwithstanding, IMO there are fewer upsets today than before feedback. 4) the base RNG itself is something i'd like to have more confidence in.

From what i remember in the forums years ago, the HD engine *does attempt* to soften first half blowouts. Generally.

So if a dynamite team is playing a horrible team, generally but with infrequent exceptions, the point spread at the half isn't too outrageous.

Like, on occasion i might be crushing a sucky Sim opponent 22-2 or 35-5... but by the first half buzzer, the sucky team will at least climb to say 18 points.

In the second half the HD engine let's you go wild. I only remember a couple of occurences being mentioned where Sim teams scored fewer than 10 points in a half, and both were in the second half.

This softening comes into play regarding Sim upsets, because if the human team has a decent but not overwhelming roster over the Sim, this softening keeps the Sim in the game. And then a string of bad luck in the 2nd half can screw the human.

So HD Sim manipulation does exist on a certain level.

Posted by npb7768 on 10/26/2022 5:31:00 PM (view original):
From what i remember in the forums years ago, the HD engine *does attempt* to soften first half blowouts. Generally.

So if a dynamite team is playing a horrible team, generally but with infrequent exceptions, the point spread at the half isn't too outrageous.

Like, on occasion i might be crushing a sucky Sim opponent 22-2 or 35-5... but by the first half buzzer, the sucky team will at least climb to say 18 points.

In the second half the HD engine let's you go wild. I only remember a couple of occurences being mentioned where Sim teams scored fewer than 10 points in a half, and both were in the second half.

This softening comes into play regarding Sim upsets, because if the human team has a decent but not overwhelming roster over the Sim, this softening keeps the Sim in the game. And then a string of bad luck in the 2nd half can screw the human.

So HD Sim manipulation does exist on a certain level.

this is partly correct. the sim attempts reversion to mean on a bunch of different items, like an individual player's field goal attempts. a player who should shoot about 50%, who starts 0 for 10, is going to roll a significantly better then even coin, on his 11th fga. this type of effect comes into play on a bunch of different game stats (its published exactly what, somewhere).

the above will cause an effect sort of like... attempting to soften first half blowouts. but that is a loose translation at best, and then you are drawing conclusions off it that end up being wrong. a better translation is, if a team is under or over performing, then over time, the sim pushes them back towards 'average'. that is relative to each team's inherent goodness, though, not about the overall margin in the game.

example: if team A is on average going to wreck team B, by 30 points per half, and they are up 15 points at 10 minutes - the game is not trying to soften that. if the first team is up 30 at half, roughly, everything is going as 'it should', and the sim engine is not pushing up or down for half 2. its not about the margin at half, its all about how team A is performing relative to the real expectations on team A, and how team B is performing relative to the real expectations on team B.

this is still a generalization... what is actually happening is what i said in the first paragraph. but its reasonable to think of the net effect as pushing teams to what their performance should be. its not about keeping games close or not close. but you could say... if a game should be close, then the sim tries to keep it close (generalization, but a fair one). if a game should be a blowout, then the sim tries to make it a blowout.

Posted by gillispie on 10/26/2022 7:03:00 PM (view original):

Posted by npb7768 on 10/26/2022 5:31:00 PM (view original):
From what i remember in the forums years ago, the HD engine *does attempt* to soften first half blowouts. Generally.

So if a dynamite team is playing a horrible team, generally but with infrequent exceptions, the point spread at the half isn't too outrageous.

Like, on occasion i might be crushing a sucky Sim opponent 22-2 or 35-5... but by the first half buzzer, the sucky team will at least climb to say 18 points.

In the second half the HD engine let's you go wild. I only remember a couple of occurences being mentioned where Sim teams scored fewer than 10 points in a half, and both were in the second half.

This softening comes into play regarding Sim upsets, because if the human team has a decent but not overwhelming roster over the Sim, this softening keeps the Sim in the game. And then a string of bad luck in the 2nd half can screw the human.

So HD Sim manipulation does exist on a certain level.

this is partly correct. the sim attempts reversion to mean on a bunch of different items, like an individual player's field goal attempts. a player who should shoot about 50%, who starts 0 for 10, is going to roll a significantly better then even coin, on his 11th fga. this type of effect comes into play on a bunch of different game stats (its published exactly what, somewhere).

the above will cause an effect sort of like... attempting to soften first half blowouts. but that is a loose translation at best, and then you are drawing conclusions off it that end up being wrong. a better translation is, if a team is under or over performing, then over time, the sim pushes them back towards 'average'. that is relative to each team's inherent goodness, though, not about the overall margin in the game.

example: if team A is on average going to wreck team B, by 30 points per half, and they are up 15 points at 10 minutes - the game is not trying to soften that. if the first team is up 30 at half, roughly, everything is going as 'it should', and the sim engine is not pushing up or down for half 2. its not about the margin at half, its all about how team A is performing relative to the real expectations on team A, and how team B is performing relative to the real expectations on team B.

this is still a generalization... what is actually happening is what i said in the first paragraph. but its reasonable to think of the net effect as pushing teams to what their performance should be. its not about keeping games close or not close. but you could say... if a game should be close, then the sim tries to keep it close (generalization, but a fair one). if a game should be a blowout, then the sim tries to make it a blowout.

Okay so im guessing, or hoping, that everything like fatigue is factored in after this "real expectation" is produced. For instance, if an awesome shooter started 0/5 but was now tired, would the sim first maybe change like a 50 percent chance of his shot going into lets say a 75 percent chance, then factor in fatigue afterwards? Or factor too many shots taken in one lineup, things like that.

Really in any way you'd like to share your knowledge of real expectations that the sim may have, I'd love to hear it. I thought the PBP itself was in question to have important each individual event was so now my brain is a bit mixed up I will admit. Wasn't that the theory as you can just generate non play by play games in modes such as dream teams in this site? I swear I read you say that the PBP was a lot of window dressing?

Please send help if you can, thank you as always

there used to be a big second half adjustment back in the day. That was curbed I think with an early engine rewrite.

I remember a D2 game between two somewhat evenly talented teams with human coaches. The score at half was 44-2. The losing team came roaring back and ran away with the second half 48-10. Not enough, as the team winning at half hung on and won 54-50. Nothing to see here. Just a typical 4 point win between two even matched teams.

Posted by gillispie on 10/26/2022 3:56:00 PM (view original):
im not saying we know any of this for sure, but i do want to say... i agree with shoe that we should know if there is a 'lucky day' roll driving the direction, but also, CS has repeatedly and consistently said there is not. i think their position would be that there is none and that this is public, confirmed knowledge.

to baums's point, its really remaining user skepticism about the existing RNG and the impact on upsets, that caused seble to add feedback in the first place. basically, there were allegedly only 'non-artificial' upsets, and this still was too much randomness for folks. so seble added feedback, which is sort of analogous to how in recruiting, you can't have a 90/10 battle. in recruiting, its a hard cutoff, and in game sims, its a soft directional shift, making extreme upsets less likely, but with no hard cutoff.

its definitely a matter of opinion how well this all worked out. my take is that 1) there is no 'lucky day' roll by design. 2) feedback was moderately successful in its goal of 'increasing reversion to mean', if you will, in reducing crazy upsets. 3) overall, feedback is not something i'm a huge fan of and i definitely think there could be some bugs in there that screw things up. notwithstanding, IMO there are fewer upsets today than before feedback. 4) the base RNG itself is something i'd like to have more confidence in.

The feedback to manipulate the RNG to reduce upsets really doesn't sit well with me and the fact that we know it exists is kind of shocking. Change the core game mechanics where slight differentials in ratings lead to more favorable outcomes for the better team or something but manipulating the RNG seems like a huge corner to cut with consequences that we can't see. The RNG should just be the RNG.

This reminds me of how streaming platforms have to manipulate the shuffle algorithm so that it seems random. Just make it ******* random.

I’m curious about how the feedback mechanism interacts with game planning. Like if I come out crushing it with a spot on game plan against a better team talent wise, will the feedback mechanism kick in and “soften” the better team?

I feel like my accuracy in predicting an 8, 10, 12-0 run to start the second half is near 95%. That shouldn't be.

Posted by oldwarrior on 10/27/2022 10:58:00 AM (view original):
there used to be a big second half adjustment back in the day. That was curbed I think with an early engine rewrite.

I remember a D2 game between two somewhat evenly talented teams with human coaches. The score at half was 44-2. The losing team came roaring back and ran away with the second half 48-10. Not enough, as the team winning at half hung on and won 54-50. Nothing to see here. Just a typical 4 point win between two even matched teams.

Do you believe the actual effect was curbed or just the extreme level it was presented to us?

Posted by tmacfan14 on 10/26/2022 8:11:00 PM (view original):

Posted by gillispie on 10/26/2022 7:03:00 PM (view original):

Posted by npb7768 on 10/26/2022 5:31:00 PM (view original):
From what i remember in the forums years ago, the HD engine *does attempt* to soften first half blowouts. Generally.

So if a dynamite team is playing a horrible team, generally but with infrequent exceptions, the point spread at the half isn't too outrageous.

Like, on occasion i might be crushing a sucky Sim opponent 22-2 or 35-5... but by the first half buzzer, the sucky team will at least climb to say 18 points.

In the second half the HD engine let's you go wild. I only remember a couple of occurences being mentioned where Sim teams scored fewer than 10 points in a half, and both were in the second half.

This softening comes into play regarding Sim upsets, because if the human team has a decent but not overwhelming roster over the Sim, this softening keeps the Sim in the game. And then a string of bad luck in the 2nd half can screw the human.

So HD Sim manipulation does exist on a certain level.

this is partly correct. the sim attempts reversion to mean on a bunch of different items, like an individual player's field goal attempts. a player who should shoot about 50%, who starts 0 for 10, is going to roll a significantly better then even coin, on his 11th fga. this type of effect comes into play on a bunch of different game stats (its published exactly what, somewhere).

the above will cause an effect sort of like... attempting to soften first half blowouts. but that is a loose translation at best, and then you are drawing conclusions off it that end up being wrong. a better translation is, if a team is under or over performing, then over time, the sim pushes them back towards 'average'. that is relative to each team's inherent goodness, though, not about the overall margin in the game.

example: if team A is on average going to wreck team B, by 30 points per half, and they are up 15 points at 10 minutes - the game is not trying to soften that. if the first team is up 30 at half, roughly, everything is going as 'it should', and the sim engine is not pushing up or down for half 2. its not about the margin at half, its all about how team A is performing relative to the real expectations on team A, and how team B is performing relative to the real expectations on team B.

this is still a generalization... what is actually happening is what i said in the first paragraph. but its reasonable to think of the net effect as pushing teams to what their performance should be. its not about keeping games close or not close. but you could say... if a game should be close, then the sim tries to keep it close (generalization, but a fair one). if a game should be a blowout, then the sim tries to make it a blowout.

Okay so im guessing, or hoping, that everything like fatigue is factored in after this "real expectation" is produced. For instance, if an awesome shooter started 0/5 but was now tired, would the sim first maybe change like a 50 percent chance of his shot going into lets say a 75 percent chance, then factor in fatigue afterwards? Or factor too many shots taken in one lineup, things like that.

Really in any way you'd like to share your knowledge of real expectations that the sim may have, I'd love to hear it. I thought the PBP itself was in question to have important each individual event was so now my brain is a bit mixed up I will admit. Wasn't that the theory as you can just generate non play by play games in modes such as dream teams in this site? I swear I read you say that the PBP was a lot of window dressing?

Please send help if you can, thank you as always

i tried to use the term that would be least confusion, but probably erred... anyway my understanding is, when a dude goes to take a shot - after everything is considered - a coin gets flipped. the odds have already been calculated, based on everything the sim engine does - distance to the basket, ratings, the quality of the defense, the works. some coin gets flipped, say the guy is 55% - the coin flips a 0.2, which is under .55, so the guy makes the shot. well, for feedback, the sim engine writes down, 55%, 1 shot made. let's say after 5 shots (feedback takes a while to kick in, probably? hopefully? not really that sure), the odds were 55%, 40%, 60%, 50%, 45%, then on average, the guy should have made 2.5 shots. let's say he really made 4 shots. the engine has just kept track of this - the 'real expectation', along the way.

then, when the guy takes his 6th shot, the engine figures the coin to flip like it always did / does, outside of feedback, and let's say its 55%. because of feedback, the engine goes well, this guy is making too many - so let's only give him a 50% chance. then, the coin gets flipped, let's say it comes back .52 - well, now the guy doesn't make the shot, when he would have pre-feedback. the feedback helps compensate a bit for the guy shooting hot earlier. if instead the coin flipped a .25, the guy makes the shot no matter what - so in this example, the feedback effect would be modest.

note it may be weird to call this a coin flip but in actual programming you basically just call the RNG and get a value 0-1. if you were a 55% to make it, you check if its under .55. so sorry if the explanation is strange in this regard but just thinking of it as a coin flip is perfectly reasonable, a weighted coin flip. it really makes no difference.

anyway. by this method, fatigue, everything, would be factored in. the engine is adjusting on the guy's real expectation. side note - when the game has no feedback, keeping track of the expectations and outcomes is fairly straight forward. still easy to make a mistake, but fairly straight forward. keeping track of the expectations and outcomes properly when you are applying feedback along the way, definitely more error prone. that is one area that would be pretty ripe for issues IMO. but if that is causing unrealistic sims, i think its a mistake, not by design.

a lot of pbp info can be meaningful, i think when a team makes a shot, that definitely happened so to speak. even if the pbp is generated after the game is over, just to make things interesting, it still can reflect some good info. substitutions and when fouls happened, who is getting double teamed, that stuff all seems to reflect what happened. some of the linguistic flair if you will, doesn't actually differentiate between different things happening. but i think even stuff like long 2s are probably getting their wording from underlying events that occured in the sim engine flow. i probably said something like, the pbp is half window dressing... but i think there is stuff in there that can provide insight. a lot of the pbp window dressing frankly relates to limitations in individual player stats, some stat are really meaningful while others are basically garbage, and the pbp works to represent all the stats - so its basically partly tainted even before it adds its own wording flair. still useful, though.