Soviet Overwhelm After Patch 1.33

[Gwydion5]

Observe mode has never worked. Don't use it. It's broken.

Having done testing in both observe and human_ai modes, I feel fairly confident in saying that Russia consistently beating the snot out of Germany in 1941 is universal, which is why I include it in my assessment. At the time I did not know about the issues with observe mode and have requested that a stickied post be made on the general forum as well as the bug forum regarding the potential issues with observe, as well as guidelines on how to use it, or how to do testing without using it. As I was a bit frustrated that this known issue was not easily visible to people like myself who may not follow the forums / dev postings regularly enough to know about a post back in June of 2016. Take for example this thread on the bug reporting forum. 11 posts between January 12th to January 23rd, and nobody said "Hey I see in your screenshots you are using observe mode, you should know it has issues and can create problems". I even tagged podcat, and had a conversation with SteelVolt about that thread and never was told by anyone about using observe mode.

But yes, I do not use observe mode anymore due to potential unknown issues with using it. And if anyone notices someone else using it, please let them know.

 

[Kaosium]

Most of my tests have been conducted with the following settings:

Game Setup : New Game - 1936 Start - Regular Difficulty - Historical Focus - No Boosting - No Mods or custom edits of files.

So if you, or anyone else is observing differently, it would be helpful to support such claims with screenshots and baseline configurations/processes behind testing the game to draw conclusions from on why there is such a difference of experience. Without that kind of information and evidence though, I don't see how we can be constructive with such claims.

I have a game (playing France) with those exact same parameters saved in January, 1942 where Germany has conquered the UK and has the Soviets on the ropes, as in no matter what I do (send volunteers etc) the USSR falls within a couple weeks/months. Moscow has been conquered, as has St. Petersburg so volunteers show up in Archangel and by the time they get near to Moscow to try to retake it the Soviets capitulate. This game is definitely 1.3.3 but was started a couple weeks ago before so some tweaking (the NF Metropolitan France gives different provinces the additional infrastructure now) has taken place since then. This is not on observer mode which has been known to bollocks up the AI for whatever reason.

I don't really know how to make a screenshot and post it here, but if you'll give me step-by-step instructions I can probably figure it out. There must be a way to post the save game file too from what I have read about it over the years, I just have never done that and don't know how to do it off the top of my head.

 

[Gwydion5]

@Kaosium

Based on your response, I would suggest starting a new game with those settings and observing a nation that will survive until 1941 using the "human_ai" console command. Like say Switzerland or Germany or Russia or UK and see what happens with a sample of 5 games. Make a save in July 1941, and January 1st 1942, and upload them if your AI Germany beating the snot out of AI Russia. You should be able access your save games in your "Documents\Paradox Interactive\Hearts of Iron IV\save games" folder. Though I believe you need to zip to upload them directly to the forum.

 

[Kaosium]

@Kaosium

Based on your response, I would suggest starting a new game with those settings and observing a nation that will survive until 1941 using the "human_ai" console command. Like say Switzerland or Germany or Russia or UK and see what happens with a sample of 5 games. Make a save in July 1941, and January 1st 1942, and upload them if your AI Germany beating the snot out of AI Russia. You should be able access your save games in your "Documents\Paradox Interactive\Hearts of Iron IV\save games" folder. Though I believe you need to zip to upload them directly to the forum.

If you're using 'observer mode' that might be the problem. When I first started posting here several months ago I read numerous times that putting the game into observer mode would cause the AI to act nutty and not to report any 'bugs' or AI flaws found in observer mode as that was known to be screwed up.

 

[Gwydion5]

If you're using 'observer mode' that might be the problem. When I first started posting here several months ago I read numerous times that putting the game into observer mode would cause the AI to act nutty and not to report any 'bugs' or AI flaws found in observer mode as that was known to be screwed up.

There is a console command called "observe" which puts the selected nation in observe mode that Podcat has called the Heisenbug and SteelVolt has explained it some in this post. Then there is the console command "human_ai" that enables the AI for the active nation, which is what I now do all my testing with. There is also a command called "fow" which removes the fog of war, but Havebeard has said that command may influence AI behavior. So I do not use it. Basically I use human_ai and tag in my testing, and only when the game is paused do I do anything. As far as I know, nobody has said there are known problems with using these. Which is what a majority of my 1.3.3. tests have been with.

 

[Meglok]

There is a console command called "observe" which puts the selected nation in observe mode that Podcat has called the Heisenbug and SteelVolt has explained it some in this post. Then there is the console command "human_ai" that enables the AI for the active nation, which is what I now do all my testing with. There is also a command called "fow" which removes the fog of war, but Havebeard has said that command may influence AI behavior. So I do not use it. Basically I use human_ai and tag in my testing, and only when the game is paused do I do anything. As far as I know, nobody has said there are known problems with using these. Which is what a majority of my 1.3.3. tests have been with.

Human_ai is likely safe. I always choose Haiti as my nation when I am testing. I just save, pause, and tag to the nation I want to check in those games. Take a snap shot so to speak of the games progress at that point. Then go back to Haiti and restart.

 

[Meglok]

You know more about modding than I do. Is it possible to add a straight via mod? I feel that some system of closing off the eastern end of the Channel to Axis traffic would help. Though if the other anchor was Calais the problem would come back. From what I have seen of the Danish Straights, it is not always necessary to have two anchor points. A single one in Dover might be best.

Such a straight would not interfere with Sea Lion.

I guess the second question would be whether you can have a straight which restricts the passage of ships but does not allow crossing by ground units (unless on transports obviously)? This second point is less critical than the first as long as the RN keeps a strong presence in the Channel.

Yes, you can add or remove straits. It depends on the sea regions at the point you want to add the strait.

 

[Gwydion5]

Human_ai is likely safe. I always choose Haiti as my nation when I am testing. I just save, pause, and tag to the nation I want to check in those games. Take a snap shot so to speak of the games progress at that point. Then go back to Haiti and restart.

I'm starting to question if human_ai is even safe. I just ran a Guatemala game starting in 1936 and did not use any console commands. AI Germany performed a little better, but still lost Berlin in October of 41.

I ran a second game, loading from the June 1st 1939 save from the first game above and had different results.

Typically I see Soviets close to Berlin and more Axis territory lost by July 1st 1941. So I'm kind of wondering if perhaps using human_ai as Germany is inhibiting them in some way, or if AI logic/setup is being influenced / affected by your choice of human/starter nations? And you are already aware of the potential issues with loading a save game causing the AI to re-evaluate its position, potentially making changes and adjustments that probably would not have happened if the game wasn't reloaded.

 

[Dalwin]

I have seen many games since 1.3.3, even when I was supporting Germany as Hungary where they simply refuse to put up good resistance in the northern sector of the Russian Front. My small force of 10 divisions was reaching Kiev at the same time the Germans were falling back across the Danzig Corridor into Germany proper. They shift too many units to the Hungarian and Rumanian sections of the line while leaving the north too thin. This has nothing to do with the Human_AI command. It is all about the balance having shifted radically against Germany.

They will happily ship units from Norway which is only partially conquered down to Africa via the Channel. They already did not have enough units (in no small part because of new resource penalties). The way they send their units to lower priority areas and throw them away on boat rides is not something they can afford to be doing.

To get a good SP game now you must either play Germany or one of their allies or give Germany at least +3 on the slider. That is if they are facing a Soviet AI. I don't think there is any point at all in playing the Soviets in an SP game until 1.4 comes out. Even then we can only keep our fingers crossed in hopes that they can fix this. It got this bad by fixing other things which were obviously needing to be fixed.

The number of divisions that Russia was using to garrison borders in central Asia was silly. Fixing that makes them stronger which is a good thing if you are playing Germany. Fixing the bug that had the Allies running out of convoys so that hundreds of divisions were stuck at sea taking attrition also fixed their inability to launch invasions. That is trading a big problem for a smaller one, but the difference is overwhelmingly against the Germans.

The new resource penalties are very much against the Axis. This is dead on from an historical standpoint, but combined with other factors it is no wonder that the German AI is virtually always dead before the end of '42 now.

 

[Slimmins]

@Gwydion5 You're pretty spot on, I ran a game as you suggested and this seems to be the case. I'm gonna run another game when I get the chance, I have a feeling that the German AI stupidly attacks the Stalin line with soldiers that are under equipped (I think it's called the Stalin Line, please correct if wrong). The game I had just watched, the AI threw inf at the damn forts with dug in SU soldiers. They lost bitterly then from what I can do the SU just punched through here made their way to Berlin by early 41.

Also in this game, Finland never ended up fighting the SU.

 

[Gwydion5]

They shift too many units to the Hungarian and Rumanian sections of the line while leaving the north too thin. This has nothing to do with the Human_AI command. It is all about the balance having shifted radically against Germany.

That may be true, but would not be exclusive to AI Germany. In my human_ai games observing Russia, I've seen the Russian front/BP do some similar things, like shifting 203 divisions to the southern front as evidenced in the second and third screenshots in this post. Now mind you that's AI Russia fighting max boosted AI Germany from late January 41 to November 41 before suddenly deciding the south needs 90% of the western front divisions. So to me, the variances we see may just be a result of RNG/Bugs in the AI front/BP management. I've also seen AI Germany games where AI Russia does a steady march west effectively and Berlin has fallen by May of 41, half of Hungary, all of Slovakia, all of former Poland. I want to say this is the most common result, but I think it might be my cognitive bias showing from the traumatic imprinting of AI Germany's worst defeats to the AI Soviets.

In my 2 Guatemala games though, I couldn't see what the AI was doing over there, so I have no idea about the differences and front dispositions, encirclements or routs, as the war went on. So I can't really say why.

The way they send their units to lower priority areas and throw them away on boat rides is not something they can afford to be doing.

Yep. It's just too easy to play the UK and convoy raid the east side of the Atlantic while the Axis send units. I think that might be why the AI in my Human Germany 1940 game might do a bit better than my Human 1939 game, in that I don't call Italy in to the war with the Allies in 39/40, so it can send all the divisions it thinks it needs to Ethiopia and Africa without being sunk. I think that also explains why in my Human 1940 game AI Germany successfully invades the UK, because I've kept the Kreigsmarine in the Baltic and is in good shape when I hand it over to the AI, where as AI Germany in the 1939 save will lose a lot of it's navy and air power in 39 before Germany finishes conquering France and friends in western Europe.

The new resource penalties are very much against the Axis. This is dead on from an historical standpoint, but combined with other factors it is no wonder that the German AI is virtually always dead before the end of '42 now.

I know this is probably a factor. But from the human controlled games I've done, AI Germany still manages to bungle it up when given control in 1939 vs 1940. That to me indicates a production / inventory management issue, in that the AI is consuming inventory and does not have enough surplus for a sustained war against a capable opponent. If AI Germany and AI Russia are at the same level of equipment for the field armies, AI Germany loses since AI Russia will have more divisions/better resistance, short of doing dumb things like 60 divisions in a single province that has like 4 infrastructure for a month or three. Hmmmm there should be a way to test this. I have no doubt you and Meglok are right about the resource shortages and penalties having a role in all this. But the AI just seems so much more aggressive on recruitment that there is never really a surplus of inventory, which means the combat effectiveness per division tends to be lacking. IMHO anyway.

The game I had just watched, the AI threw inf at the damn forts with dug in SU soldiers. They lost bitterly then from what I can do the SU just punched through here made their way to Berlin by early 41.

As Dalwin pointed out, perhaps AI Germany has an issue with protecting the northern front. So if you see AI Russia beating AI Germany, make note of the lands it has taken. There could be a bug or something going on here where AI Germany is abandoning Northern Poland for whatever reason. (Sweet Sweet Oil in Caucasus? maybe)

@Gwydion5 You're pretty spot on,

Thanks, I try to be thorough in my testing so my assertions ring true. Double checking my assertions is always a good thing, it will either build consensus or it will reveal a flaw in my assessment. Both are win win in my opinion. So thank you for taking the time in doing your own tests.

 

[Dalwin]

I know this is probably a factor. But from the human controlled games I've done, AI Germany still manages to bungle it up when given control in 1939 vs 1940. That to me indicates a production / inventory management issue, in that the AI is consuming inventory and does not have enough surplus for a sustained war against a capable opponent. If AI Germany and AI Russia are at the same level of equipment for the field armies, AI Germany loses since AI Russia will have more divisions/better resistance, short of doing dumb things like 60 divisions in a single province that has like 4 infrastructure for a month or three. Hmmmm there should be a way to test this. I have no doubt you and Meglok are right about the resource shortages and penalties having a role in all this. But the AI just seems so much more aggressive on recruitment that there is never really a surplus of inventory, which means the combat effectiveness per division tends to be lacking. IMHO anyway.

I don't think your logic is sound here. Yes by handling the early years for the AI you have no doubt given it a better starting position than if it had done those years itself. That does not mean that it is in a better position than it would have been in 1.3.2. It still seems to have not enough units to do all the things that need doing. The new resource system means that Germany will have less to work with once the war starts. In my German games in 1.3.3 I have noticed that it is much harder to get what I think are adequate forces to hold the west and attack the east as well as having a decent air force. Much harder than it was previously.

If it is harder for me I have zero doubt that it is also harder for the AI, much more so based on some screen shots showing how many German factories were offline in 1940 due to resource shortages.

The Ai has always had trouble prioritizing the use of its forces. It is very wasteful. When it has less than it needs, it collapses very quickly. Give Germany +2 or +3 and have them fight against the Soviets. If it is all bad strategy they will quickly collapse anyway. If they do well then economics was more than just a slight contributing factor to their collapse.

 

[Slimmins]

@Gwydion5 Yea, I'm gonna try note the paths the AI tends to take/lose

My issue with my idea though is that it seems inconsistent and I need to check some facts about how the AI works in HOI. For instance are the AI's for each countries the same and their actions dependant on their choice tree? Or are some more simply aggressive than others (I imagine it's a combo).

Then you have the Maginot line, I have rarely seen the Germans attack through here. So why do they attack the Stalin Line? Especially factoring in the obvious resource shortages? They are both fortified lines with soldiers present, it just seems abusrd

 

[Gwydion5]

I don't think your logic is sound here. Yes by handling the early years for the AI you have no doubt given it a better starting position than if it had done those years itself. That does not mean that it is in a better position than it would have been in 1.3.2.

It's clear that the AI Production has changed in 1.3.3. There is definitely more priority / sensitivity to the needs of tanks and artillery over air power, ships are also being built as well. So something has changed in that area. In 1.3.2. I don't recall AI Germany having the equipment shortages it does now in the war with Russia in 1.3.3 which makes me think it might be trying too hard to utilize maximum inventory/logistics during the calm before the storm (1940-1941), which translates to equipment shortages in war time. I guess the only way to know would be to load up 1.3.2 and take monthly snap shots of the logistics screen / division overview screens to get an idea.

Then you have the Maginot line, I have rarely seen the Germans attack through here.

From my 2 Guatemala games today.

First game :

Second Game :

In most of my games that I've observed, I'd say there is a 60-75% chance that AI Germany will break through the maginot line before Netherlands or Belgium have capitulated. I think this is mostly because France removes so many divisions from the Maginot line for Early Naval invasions of Germany, as well as moving troops to fight in Belgium and Netherlands. In the first game France capitulated before Belgium or Netherlands. So if you are seeing differently, I wonder why.

 

[Slimmins]

Far out, I'd love to see the manpower loses.

 

[Gwydion5]

Far out, I'd love to see the manpower loses.

There really isn't much to talk about. France is basically abandoning the Maginot Line, leaving one division per province, and even then some of those divisions are being moved and losing their entrenchment bonuses. So in terms of casualties, you are going to have a hard time identifying them over active battles happening in Belgium and Netherlands which often puts up more resistance than the Maginot Line.

But if you want to check it out, here is the Guatemala June 1st 1939 save, just "tag ger" or "tag fra" and "human_ai" from the console and you should be all set for observing the the fall of the Maginot Line and possible casualties related to it.

 

[Gwydion5]

So reverting to 1.3.2 to try to compile some numbers on production, logistics and division overview for AI Germany and compare to 1.3.3. AI Germany, I have a screenshot to show how AI Germany has done at the start of 1942. This is a 1936 start as Haiti, no console commands other than to show the version at the start of the game, I did an integrity to check to validate the 1.3.2 reversion, and finally this is single play through so at no point did I load/reload the game.

Also AI Germany did bust through the Maginot Line well before Netherlands and Belgium fell, and Netherlands and Belgium fell before France. FWIW anyway.

 

[RforRush]

1.3.3, 1939 start, not-historical, no buffs.
Big issue on start of the war is that Germany places troops in Hungary and Romania prepared for attack, but it takes them couple of weeks to actually invite into war, so there're only 50 divs in Poland area versus 100+ soviets on the start of the war.
Also I noticed German AI no longer declares war just because it can. Germany actually waits for good moment, in screenshots below Germany finished UK before declaring on SOV (waited for year after taking focus).
Screenshots here, Germans "defending" vs USA and SOV at the same time. SOV manages to defend well while taking triple casualties. I think Axis will win eventually, they are too strong this time.

Spoiler: Screens

Anyway, I am happy that Soviet is no longer super easy to beat, it was a lot worse in 1.3.2.

So reverting to 1.3.2 to try to compile some numbers on production, logistics and division overview for AI Germany and compare to 1.3.3. AI Germany, I have a screenshot to show how AI Germany has done at the start of 1942.

In 1.3.3 AI no longer defends peaceful borders, this is why SOVs lose in 1.3.2.

 

[Gwydion5]

1.3.3, 1939 start, not-historical, no buffs.

I'm not sure how that is relevant considering all my tests, and by extension all my arguments, are based on 1936 starts, and historical focus.

 

[Dalwin]

I'm not sure how that is relevant considering all my tests, and by extension all my arguments, are based on 1936 starts, and historical focus.

The vast majority of games played are '36 starts. Over 95%.

It is also preferable to use historical focus for any comparative testing. I know a fairly large percentage of players turn off historical focus, but doing so makes the game much more volatile and this randomness makes it harder to compare one run to another without some fluke invalidating a run.

1936 start with historical focus is exactly the right choice for the testing that you are doing.