AI doesn't play fair

[Juswin24]

Has any one had three tanks full pips in the open go against one infantry group full pips not an elite group and have a draw? First after open phase and I can't beat an infantry group with three tanks.

[Topher]

Many players suspect an issue with the AI's combat rolls; however, the developers have made it clear that the AI neither cheats in any way nor have units with any built-in advantages over player units, it is, according to them, a level playing field.

That said I have an AI bug report in right now for a something very similar.

[Claymmore]

I'm pretty sure the AI does not play fair. I play the Russians and it seemed I would get crushed when the odds were in my favor quite a bit so I recorded the results of fourteen turns from one game and recorded the results as:
German did better than they should
Germans did worse than they should
The results were in line with the odds.

The game played like it normally did (the battles were in line with what I have grown to expect.

Here are the results of the battles

Germans did better than they should - 32
Germans did worse than they should - 11
The results were in line with the odds - 12

I know this is just one game but, as I said, the results were in line with what I've grown to expect. Anyone else want to give it a try and send in their results?

[Dbest]

Something seems wonky.

I replayed the first turns to get some easy data. I had the Germans attach in yelnya and zhukovka.

In 8 battles v. Armor (predicting 2 hits) 11 Russian inf got:
2
4
0
2
3
7
3
0

In 8 battles v. Inf (predicting 3 hits) 11 Russian infantry got:
6
5
2
6
3
2
3
4

[Dbest]

Rosskowksy almost never does worse than expected. He often does much, much better than expected. Granted, the sample size is small but this is close to getting heads 70% of the time. This backs up what I've experienced in dozens of outcomes.

In short, this ai seems to cheat. You have to get really lucky to win. I've done it, but I always feel like it isn't skill but just getting lucky that does it. Not fun.

[sa_gibson]

I posted this the other day in the new AI thread:

I felt my earlier post here might be unfair - memory is a funny thing & we tend to remember the poor results better than the good ones. So I decided to run some tests on opening planned offensives.

Halder's preferred opening offensives are 1) elite armor stack in Vladiminskoye, 2) infantry split into Andreapol & Olenino, & 3) armor / mech in Shostka. I started 10 games with Halder & he made these attacks every time. I tracked the results. In Vladiminskoye, Halder achieved 10 KIAs and got the armor exploit advantage. In Shostka, Halder achieved 8 KIAs with armor exploit, and 2 retreats by Sov defenders.

Then I switched sides and started 10 games vs Rokossovsky, using the same planned offensives. In Vladiminskoye, I got 6 KIAs and 4 retreats. In Shostka, I got 1 KIA, 7 retreats, and 2 failures to even dislodge the Soviet defenders.

I realize 10 tests doesn't prove anything, but this seems crazy to me. So, I stand by my earlier post. Something is wrong.

[blahblah3502]

I totally believe that the game developers didn't design a cheating AI , but I'm starting to believe too that there is something skewed/buggy in the random number generation for Halder.
Halder virtually always chooses the armor stack attack on Vladiminskoye in the opening prepared offensive moves, and he virtually always ( I have never seen otherwise after a dozen or so times) wins a total kill, gaining the armor breakthrough bonus.

I don't have an opinion about if Rosskovsky has similar issues or not.

[Lostmydice]

Being sick for a week I have played many full campaigns vs AI both sides and I am now absolutely positive that the battle outcome favours AI. I have turned in several bug reports.
I barely make any hit without being in overwhelming force. 8:1 (even in defenseless mode) in the plain is no warranty to overcome the AI. At best I can count on a (mostly unwelcome) retreat.
While the battle forecast seems to be balanced the outcome is more or less always in the worst case corner for me no matter which side.
Way too often AI manages a 100% score (at least not more...) being attacker, non-elite and outnumbered vs supperior, elite defender while on my counter attack I get hammered as well.
It must be a coding error or bug as after all the campaigns I have played it can't be random any longer.
It is still very manageable to beat AI but only under a complete different strategy as you would use expecting to play level field with AI.
Hope they can fix it...

[Lostmydice]

I have just thrown the towel in another sick game vs. AI where 1 single Chuck Norris pip defeated almost my complete forces.
I mean hey - is this a ridiculous US movie or is this meant to be a strategy game?
My personal stats say that the ratio is 1:4 at the best which is like playing Monopoly with a dice only showing 1.5 dots vs. AI playing with a full 6-dot dice.
I may not be the most brillant player in this community but I am deffo better than Hal and Rokky but it feels like I fight with bare knuckles against tanks no matter how smart I move.
Playing against human players would probably be the best solution but I bought DOM to play in airplanes and hotel rooms with no or bad/expensive wifi connection to kill my time while travelling not to frustrate myself with random matches...

[Miguel_TSS]

Just a reminder that we are exploring this with the help of your bug reports (In Game Menu -> Feedback). Thanks for your feedback and comments.

Get well soon, Lostmydice.

[Lostmydice]

Thanks mate.
I hope it's not AI making me sick
Will send you some more weird outcomes.
Thanks for caring.

[Guderian24]

I feel the multiplayer experience has changed as well as the AI play--- I have given up for the most part on any new multiplayer games. When I start Operation Typhoon or Moscow campaign as the Axis against I now cheat, since it takes about 20 attempts of replaying the 1st turn before I get results that were typical before the 2.0.166 release. Where results on the three planned offenses were favorable in 1.x, it now takes at least 20 attempts know to get even close to what used to be normal. I loved the game before 2.x and now am just frustrated... I wish we had the option of individually rolling back to the old version.

[daviddunham]

When I start Operation Typhoon or Moscow campaign as the Axis against I now cheat, since it takes about 20 attempts of replaying the 1st turn before I get results that were typical before the 2.0.166 release. ... I wish we had the option of individually rolling back to the old version.

We spent a fair amount of time last week reviewing this, and none of the three people looking found anything that was out of the ordinary.

(I just now started Typhoon in 2.0, attacked in Vladiminsokoye (sic) and Shostka and got armored breakthroughs in each, and a retreat in Zhukovka. Purely anecdotal since the sample size is 1, but this is a bit better than expected.)

Apple does not support rolling back.

As a game developer, “individually rolling back” sounds like a bad idea, in that different players would be on different versions. (We have not taken strong measures to ensure that all players are on the same version; perhaps we should. Except that might mean everyone moves at the speed of the slowest upgraded, and that would mean that bugs don’t get fixed.)

Anyway, we always appreciate the reports (especially when sent directly via the Feedback button, because we are lazy and like having all the necessary information in one place).

[sa_gibson]

I'm confused - I saw a post from Miguel earlier which made it seem you guys were pursuing this as a problem, but this most recent post sounds like you're saying there isn't a problem.

Does Shenandoah have any actual data on combat results? Are the outcomes consistent with the expected results?

[sa_gibson]

I ran the same test I ran earlier.

In 15 attacks on Vladiminskoye, Halder got 14 eliminations and 1 retreat. For 15 attacks against Rokossovsky in Vladiminskoye, I got 10 eliminations and 5 retreats.

In 13(*) attacks in Shostka, Halder got 9 eliminations and 4 retreats. In my 15 attacks I got 3 eliminations, 10 retreats, and 2 cases where the defenders held.

It seems clear there's a bug here.

(*) in 2 cases Halder opted to attack somewhere other than Shostka.

[glenndrover]

I was seeing some pretty unexpected combat results in my first game against the new Russian AI, but wanted to wait until I'd played two full campaign games to comment here. In my first game, my Germans were doing OK and had advanced to take all objectives except Moscow and the surrounding VP locations by the time frost set in. There had been more than a few battles where the combat results elicited an "oh, come on!" from me when I took losses that seemed unusually high. Then, when the frost turns started and I was preparing to make a push on Moscow, the combat results anomalies became the norm. Battles where I had a great superiority resulted in disproportionate losses for me. So much so that I became convinced that the AI was cheating.
I decided to start a new game and see if it happened again. Playing the Germans in the campaign against the new Russian AI general, I was playing unusually well and really sticking it to the Russians on the first three turns. The combat results seemed, for the most part, to be within expected norms. Late on turn three things got crazy. I attacked a 3 pip Soviet tank unit by itself in an open terrain space with three German panzer units. It was 12 German pips against 3 Soviet. The result was one pip lost for the Soviet Tank and two pips lost for the Germans. Over the next two turns I attacked this unit twice more and it attacked me once. Each time I had three panzer units (I brought up a fresh three units so there were actually six German panzer units attacking the single Soviet Tank unit over three turns). The final tally of losses over four battles was German losses=10 pips, Soviet losses=1 pip, and in each battle, the Soviet tank unit was outnumbered 3-1.
There have been plenty of other anomalous results in my two games so far, almost all of them in favor of the AI, but this one was the most ridiculous and bothersome. If the combat results have not been intentionally tweaked to gain these results then there is a serious bug in the new version!

[Dbest]

I hate to sound like a whiner, but there's definitely a bug with the new AI's.

I replayed 20 games as the Russians and the axis, only watching for the initial prepared offensive at vladimskoya. When fighting rokossovsky I was successful 25% of the time at eliminating the Russian infantry. Halder is successful 90% of the time.

This isn't possible if the game is working as intended. Something happened.

[Claymmore]

There is definitely something wrong with the AI. I've learned to win with the Russians anyway by just sitting until late fall/early winter and then surround enemy units while avoiding combat as much as possible, but that means a lot of turns pressing Pass which is not very exciting. When you looked at this, did you play actual games on an iPad, or run simulations or use the development hardware? Maybe there's a difference. One thing I noticed, (this may just be random weirdness) but my cavalry seems to take fewer hits than my other units. Thanks for looking at this.

[Claymmore]

Okay, this is odd. I just played a game and every battle was in line with the stated odds. There was some variation, of course, but the game played perfectly. This was how it played the first few times I played it, before the odds started going wonky. It seems that when the odds are against you, they are for the entire game but if it starts out well, it stays that way. Hope this helps.

[Claymmore]

Okay, three weeks later and the AI has been playing fair the entire time. No idea what happened to change it. I played unfairly for about 25 games in a row and then it was fine.