AI Issues in 1.3.2 e13a

[Gwydion5]

Game Version : v1.3.2 (3A29)
DLC : Together for Victory
Mods : NONE (I have never used mods ever in HOI4, nor edited any files)

Playing Historical Focus Normal Difficulty Germany in observer mode to evaluate AI. I did buff Germany with Manpower, picked all the PP abilities, and maxed out Infrastructure, Airbases, Anti-Aircraft guns, CIC's, MIC's and Dockyards at the start.

1. Spanish Civilwar Started on February 26. When AI Germany sends volunteer forces to the Spanish Civil War and they arrive, the AI will remove or add the general at the 24:00 hours of each day. It will do Erich von Manstein for 1 day, then remove him for 1 day, then add him back in for 1 day, repeating this for months. Eventually it will add Hasso von Manteuffel and Sepp Dietrich to the rotation around June or July and sometimes they will stay for a few days before being removed. Erich von Manstein is being reassigned to a single army that has no commander in the main German theater. AI Germany also sends only 1 volunteer division, I think this is because the AI is trying to send the best general it has and Erich von Manstein is only commanding 1 division in his army.
   
   
2. AI Germany promotes Field Marshals irrationally. It promoted Paul Hausser (Who has the Panzer Commander trait) to Field Marshal in October of 1937, there are still plenty of rank 4 generals to create a new army commands, and plenty of rank 4 generals that have no traits. There are currently 6 armies in the main German Theater. Hausser's army of 26 divisions is doing Garrison for all of Germany, Gerd von Rundstedt's army of 32 divisions is commanding a front with the Czech's, Gunther von Kluge army of 30 divisions is commanding a front with France.
   
   
3. February 6th 1938, Hausser was removed as the General of the Garrison army, Rommel was removed from his army and assigned the Garrison Army. Rommel was then promoted to Field Marshal. Hausser is no longer commanding an Army and is available for assignment. February 15th or so Anschluss happens, Rommel is removed as the General of Garrison Army, Rundstedt's assigned to the Garrison army. Both Hausser and Rommel are currently unassigned. February 20th, Rundstedt's reassigned back to the Army front of the Czech's, Kluge is now assigned to the Garrison Army. Rommel has been assigned 30 divisions on the border with Italy in the Alps. April 1938, Sepp Deitrich and Heinz Guderian are now Field Marshals and currently unassigned along with Hausser.
   
   
4. Troop movements along fronts is still way too much. Even in peace time, watching the amount of troops moving along the Czech border is is puzzling to say the least. Screenshot below:
   
   What is it responding too? Ah I think I see what it's doing. It's pulling divisions off the front by reassigning them to a new army. Which then forces a realignment. The logic behind what fronts get what divisions needs to be tweaked. It should give priority during peace time to wait for the division to be created and deployed, rather than taking from an existing army. It should probably do these front/army considerations less often too during peace time. It's way too sensitive about it's army/front division allocation right now. Inserting another image on July 31st 1938. Troop movement along czech border pretty much constant.
   
   Oh and now Hasso, Weichs, Bock are now Field Marshals too.

I'm going to stop here, I have some other things to get to. But I may come back and add more to this thread.

 

[Gwydion5]

• January 4th 1939, Sudetenland has happened. The bulk of Army 2 is stationed on a single front with Poland next to Czechoslovakia highlighted by the red arrows in the screenshot. Army 2 has 3 fronts right now, 2 of which have no divisions assigned. The AI is still trying to move troops to manage these fronts as evidenced by the movement routes.
  
  
  
• Another screenshot showing the divisions themselves. Some of the divisions are not assigned to any front, indicated by their entrenchment bonus while bordering a front with no divisions assigned.
  
  
  
• On January 4th I selected Army 2 and left all those units selected. It is now January 30th 1939 and you can see that units have been reassigned from Army 2, to Army 1, Army 5 and Army 7. The logic behind the AI assignment is way too sensitive which is forcing it to constantly change the Theater/Army allocation of divisions. The amount of CPU that is used on this activity seems inefficient, especially if every nation is doing it. Also it is now almost a month later, and Army 2 still has not fixed it's fronts that have No Divisions and has stockpiled a mass of divisions in a single province. There are at least 30 divisions visible from Army 2 in in that one province bordering Czech and Poland. You can also see some of the divisions selected that were part of Army 2 originally, as well as the new units in Army 2 that are not selected. It should be noted that when Fate of Czechoslovakia happens this all resolved with the no division fronts. But because the no division fronts exists and is not fixed by the AI, as well as not seeing the folly of stock piling 30 divisions in a single province, I feel it is still an important bug to look at, simply because whatever is happening here, could happen elsewhere in the game.
  
  
  
• August 23rd 1939 Danzig or War : So here is the current Theater assignment. Rommel is on guard duty for Germany. Kluge is commanding the front with France, Rundstedt's commanding the western front for Poland. The following are generals promoted to field marshal that are currently unassigned. Albert Kesselring, Hasso von Manteuffel, Sepp Dietrich, Paul Hausser, Erwin von Witzleben, Kurt Student, Maximilian von Weichs, Fedor von Bock, Heinz Guderian, and Erich von Manstein.
  
  And here are the screenshots of all the promoted and unassigned field marshals:
  
  
  

So the AI's Field Marshal logic really needs to be improved, as it has handicapped Germany by promoting some of its best generals that have purpose orientated traits that can no longer be used. Kurt Student for example is great for Paratroops and Marines, and the AI should be evaluating that in terms of master strategy. But beyond that the AI was promoting rank 4 generals like Rommel and Erich von Manstein when it had Rank 4 generals that had no traits that it could've promoted instead for the same effect. But it should not be promoting generals to Field Marshals at this stage anyway when there are so many great generals for it to use in 24 division armies.

One of the things that I would recommend that might make this better is to add options for theaters and army command to constrain/control the battlefront lines. Like the ability to soft limit a front for 24 provinces for offense or defense for a general. If offense then the front when created does not exceed 24 enemy provinces. Like when you have your single province sharing a front with 3 enemy provinces, the 3 enemy provinces would count toward the offense limit of front size creation. Field Marshals can be exempt from this, but perhaps a soft limit on how many field marshals manage fronts in a single theater. Now I say soft limit because fronts change obviously and a hard limit would be impractical. But at the same time the AI should be trying to work inside that soft limit as much as possible. So if the soft limit is set for a specific army at 24 Offensive provinces, and the front has expanded to 48 offensive provinces, it should be evaluating whether it assigns a Field Marshal and assigning units to that army, or if it should create a new front army to reinforce that front and shrink the front size for the army that has 48 offensive provinces.

I'm just spitballing on a possible solutions though. I have no idea what's going on with the AI. I just know it is very inefficient. From a technical stand point it's constantly moving troops around between the armies, which causes those fronts to realign their divisions along the front, in addition it is adding/creating armies greater than 24 divisions which I assume is triggering the Field Marshal promotion. The generals that it's promoting to Field Marshal tend to be the best generals, when there are better generals to use for that Promotion, like an equal rank general that has no traits should always be preferred over an equal rank general with traits. It should also evaluate how many traits that General has. So let's say Erich von Manstein is rank 8 with 6 traits and the AI wants to promote a general to Field Marshal, A rank 7 field Marshal would be rather decent, but if there is a rank 7 general with 2 traits, perhaps that should be the one it uses instead.

@podcat - Just so you are aware and should you find the information useful you can act on it.

Next post I will focus on war time AI from this game.

 

[Gwydion5]

With a new beta patch out, I decided to start over and check if the previous problems were still present.

Game Version : v1.3.2 (73da) (Open Beta)
DLC : Together for Victory
Mods : NONE (I have never used mods ever in HOI4, nor edited any files)

Playing Historical Focus Normal Difficulty Germany in observer mode to evaluate AI. I did buff Germany with all the PP abilities at the start, had 100 PP left over. None of the other buffs in the previous game were applied. So Germany has a slight boost in production and research. But it has ample PP to spend how ever it see's fit.



So I'm going to try and break this game down in to parts/posts.
This is Part 1: AI Political Power Expenditures. (I have 19 screenshots to provide as evidence, but I will spare you posting all 19.)

• On January 30th, 1936 the AI has it's first opportunity to spend 150 PP points. It chooses to change from Limited Exports, to Exports Focus. Which is not a bad way to go as Germany, since what they produce a lot of (aluminum and steel) they are not able to consume with military production.
  
  
• On May 12th, 1936 the AI again has an opportunity to spend 150 PP points. It chooses to change from Exports Focus to Free Trade. Now my question is, why not just go straight to Free Trade on January 30th for 150 PP?
  
  
• The AI will proceed to spend 150 PP's switching back and forth between Export Focus and Free Trade. (5-10-1936 Export Focus, 7-31-1936 Free Trade, 10-22-1936 Export Focus) until January 14th, 1937 where it will go to Extensive Conscription. Then it's back to Free Trade Focus on 4-6-1937, War Economy on 6-29-1937 because the Spanish Civil War finished, and then back to switching back and forth between Export and Free Trade focuses which it does 10 more times.
  
  
• On 8-21-1939 2 days before the Danzig or War National Focus is going to complete the AI has 149 Political Power points, and is currently in Export Trade Focus.
  
• On 8-22-1939, 1 day before the Danzig or War National Focus is going to complete, the AI spends the Political Power points on changing the trade policy to Free Trade Focus.
  

Now I get that by adding all the policies on day 1 doesn't give the AI a lot of sensible options that it would normally have. But it still demonstrates failings with the logic / intelligence on how to use those political power points. Inevitably the AI is going to reach this problem where it doesn't have anything to spend the points on, and I question whether it will ever save up enough points to spend on policies or ministers that cost more than 250 points. Furthermore the (German) AI doesn't seem to have any inclination to use diplomacy at all. Perhaps that's related to the historical focus for the game though.

One thing that seems clear to me is that during peace time the AI should pick a trade policy best suited for it and stick with it. In the case of Germany since it's building CIC's exclusively the first year or two, it should be in free trade only as it doesn't really have the MIC's to use all of the resources it produces in surplus. It should also do valuations of the resources it needs and the resources it produces to determine if the Trade Policy it is using is effective. For example if AI Germany is evaluating whether to be Export Focus or Free Trade Focus, shouldn't it consider the cost of CIC's needed to meet current demand? So if going to Free Trade was going to cost 2 CIC's to replace lost aluminum, and 2 CIC's to replace lost steel for current production needs, should the AI do it or not? That kind of logic seems to be missing from the political power AI.

Another thing worth pointing out is that the AI did not replace any ministers or military staff. Surely Rommel should replace Rundstedt with +8% division recovery rate, or Fritsch with -8% division attrition. So there should be a weighting to the ministers in terms of their value during peace time, war time, or in the case of Martin Bormann and Rudolph Hess, how long they should ideally be used before considering a Minister more suited to the issues at hand.

That concludes Part 1. Part 2 will be the Spanish Civil War, Daily Swapping of Generals, and inexplicably the volunteers theater disappearing!

 

[Gwydion5]

Part 2 : Spanish Civil War

In the previous 1.3.2 beta build there was some funky AI behavior with the Spanish Civil War. The same issues persisted in this build as well, this time with screenshots to support the claims.

• 3-17-1936 : Spanish Civil War happens. Below is the current German army layout. As I expected Erich von Manstein only has 1 division under his command, and there are others with 2-4 divisions under their command. The AI is going to choose Manstein as a volunteer force.
  
  I'm not sure if it does this as a historical focus or not. It's not like Nationalist Spain needs much help as they tend to win in 1.3.x. But if this is not working as intended, then it's something to look at as AI Germany has plenty of options for supporting Nationalist Spain better.
  
  
• 3-22-1936 : After the AI sends Manstein in command of 1 infantry division, it then reassigns Manstein back to the German theater, before the volunteers even arrive.
  
  
• 3-31-1936 : Volunteer forces arrive and as expected there is no general. But the AI will assign Manstein on April 1st, and then reassign Manstein back to the German theater, leaving the lone infantry division and Spanish Volunteers theater without a general. This will repeat rather consistently for months, sometimes though it skips a few days of this behavior.
  
  If you pay attention to your battles against the AI you may notice this happen as well. Where one day it's a rank 5 general, and the next day it's a rank 3 or vice versa. I always assumed the general of a battle changes because different divisions under different armies come in to the combat. But now I want to look more closely and see if it isn't changing the generals out, and if this type of AI behavior is affecting other things as well. But regardless whatever is happening here is not right. For some reason the AI has a high priority to use Manstein in both theaters. In my previous game it eventually adds Dietrich and Hasso to the rotation. But Germany has so many good generals to pick from it should not need to keep reassigning, it should be able to put another competent general in Germany to command that 1 division army.
  
  
• 5-28-1936 : I went to go check on the Spanish Civil war and just noticed that the volunteer theater is gone, yet the German Infantry division is still there. Also on the diplomacy screen it still shows that Germany is supporting Nationalist Spain with volunteers. This did not happen in the previous beta build. It also appears that the infantry division just sits there and does nothing during this time.
  

That concludes Part 2. Part 3 will be the Sudetenland Front Changes and observations about the AI in general in how it manages divisions in multiple armies.

 

[Gwydion5]

Part 3 : Sudetenland Front Changes

This post is going to be a bit more nuanced on changes in the theater/army division allocation and it starts with Demanding Sudetenland.

• 9-7-1938 : Rundstedt is currently commanding Army 2 which has 31 division assigned to the Czech front prior to Sudetenland happening. After Sudetenland happens the front shrinks accordingly. What happens next though is a bit perplexing.
  
  1st screenshot shows Army 2 selected divisions and movement routes. I assume this is an adjustment due to the front change, but it's possible these were it's movement orders for the old front before Demand Sudetenland finishes.
  
  2nd screenshot shows that the AI basically reduces Army 2 from 31 divisions down to 4 divisions. Because I had all of Army 2's divisions selected, I can see which armies received them as evidenced in screenshot 2. Army 5 and Army 7 are new, Army 10 appears to be old, and 27 divisions have been reassigned to them in some quantity. Another thing to note is the army division counts. Kluge stays the same at 38, Rundstedt is down to 4 divisions, Rommel stays the same with 19, Hausser replaces Kurt Student as commander and increases division count from 11 to 20, Manstein stays the same with 3 divisions, Student replaces Guderian as commander and decreases divisions down from 12 to 10, Bock reduces from 25 divisions to 19. These realignments really come off as nonsensical to me, and they happen often.
  
  If you look at the 1st screenshot and then look at the 2nd and 3rd, you will notice that the front that has "no divisions" was initially Czechoslovakia land to the east, but now has switched to Poland, which I think is what causes the break in the front. Not sure if that land is supposed to switch or not. The main front for Army 2 is surrounding the rest of Czechoslovakia with 4 divisions.
  
  
• 9-8-1938 : Kluge is replaced by Rundstedt, 21 divisions assigned to garrison duty are reassigned to Army 2. Guderian now commands Army 2 with 25 divisions. Rommel and Manstein swap army commands. Bock in Army 10 is replaced by Witzleben(?).
  
  1st screenshot shows all the garrison units now assigned and moving towards Czechoslovakia.
  
  2nd screenshot shows a zoomed in view of the front with detail on the division assignments to the two fronts. It seems all 21 divisions that were assigned to Army 2 from garrison duty were assigned to the front that had no divisions. This front only has 2 friendly provinces adjacent to 2 polish provinces (formerly Czech provinces).
  
  3rd screenshot shows the original 4 divisions assigned by Army 2.
  
  
• 9-9-1938 : Hausser is now assigned to the garrison army, 19 divisions have been added to this army. Rundstedt is back to commanding Army 2 reducing it's divisions from 25 to 21. Witzleben from Army 10 replaces Hausser's command. Kuchler assumes command of Army 10. Guderian assumes command of Army 5 and Hell assumes command of Army 7. Army 5 now has front with western Poland.
  
  1st screenshot shows that units that were in Army 2 have been moved back to Army 1. Since Army 2 has gone from 25 to 21, as well as the main large front around Czechoslovakia now was no divisions. I'm going to to assume it was those 4 divisions that were reassigned.
  
  2nd screenshot shows that divisions assigned to this front hasn't changed.
  
  
• 9-11-1938 : Rundstedt is now back in command of Army 1 which is doing garrison duty. Hausser is now commanding Army 2 for the Czech front. Hell has replaced Witzleben, Witzleben has replaced Kuchler, Reinhardt has replaced Hell.
  
  In this screenshot you can see the divisions assigned to the small front along Poland are trying position themselves along the entire border with Czechoslovakia. Try as it might, without divisions being assigned to it, while also belonging to that small border seems to be an exercise in futility.
  
  
• 9-13-1938 : Rundstedt was at 36 divisions on garrison duty is now down to 28. Hausser was at 21 divisions is now up to 24 divisions in Army 2 with some divisions of Army 2 being assigned to Army 5. Witzleben was at 19 divisions, now at 20. Guderian was at 21 divisions, now is at 27 divisions for the front along western Poland. Reinhardt was at 5 divisions now at 3 divisions.
  
  
  
• 9-14-1938 : Kluge replaces Rundstedt for the army doing garrison duty, also increases division count from 28 back to 36. Guderian replaces Hausser for the Czech front, while reducing divisions from 24 to 22. Hausser replace Hell. Witzleben reduces division count from 20 to 18. Rundstedt replaces Guderian in Army 5, reduces division count from 27 to 22 for the western Poland front. Hell replaces Reinhardt and increases division count from 3 to 5.
  
  In this screenshot I still have all the original units selected from Army 2 on 9-11-1938. You can see some of those original units have now been assigned to Army 1 which is doing the garrison duty. This will be the last of the screenshots showing the nonsensical realignments of divisions and commanders. Anyone who sits in observe mode and watches the AI for a target country will see things like this happening far too often then they should. I think the above provides good evidence of there needing to be some improvements to the logic of managing commanders, theaters, armies and divisions.
  
  
• 11-24-1938 : Even a couple months later, the AI still has not fixed the front issues it has with Czechoslovakia.
  
  Look at all those units piled up to cover 2 provinces, while the rest of the front has no divisions.
  
  
• 2-2-1939 : Here is the AI again trying to manually assign the troops to provinces bordering Czechoslovakia while belonging to an existing front.
  
  
  
• 3-17-1939 : Front issue still unresolved, while it tries to manually deploy troops along the front with Czechoslovakia.
  

Now obviously once Fate of Czechoslovakia happens, this becomes a non issue, unless they resist. But even then Germany should quickly contain any vulnerability they have to not being able to assign divisions to the front. The bigger issue here is what does this mean? Is this isolated to the Demand Sudetenland event, or can front issues like this happen elsewhere, where the AI has broken/multiple fronts from a single front and can't resolve the issue? Clearly what ever is happening here is worth looking at as it could provide some valuable insight to lapses or loops in the programming that are systemic, and ultimately counter productive to a challenging AI that knows how to build armies, assign them properly, and clean them up when they become disorganized or broken.

 

[Gwydion5]

Part 4: Political Power Redux

I was going to try and do a Polish front breakdown for August 1939. But it's just far too pedantic trying to assess and conclude what the AI is doing. Plus it would be a lot of work to break it all down in to screenshots. As I observed above with the Czech front, the AI does a lot of division reassignments and commander replacements that seem nonsensical to me. My advice to PDS would be to have the AI programmers responsible for this to just sit in observer mode, watch what the AI does and ask themselves if that is working as intended and if that is the optimal/desired behavior.

So I have started a new game running v1.3.2(73da), regular difficulty and historical focuses on. I am not buffing Germany at all, and just looking at the decisions the AI is making regarding political power expenditures. Also something to be mindful of when looking at these pics are armies that are training. Considering it's peace time, no war justifications are happening against Germany, and no national focuses are being pursued that could put Germany in to war, I would think the AI should put more emphasis on army divisions training. Lastly, I'm not sure if these choices are due to historical focus or not.

• 3-12-1936 : Rhineland NF completes, AI's first choice for political power is Heinz Guderain for Theorist.
  
  Now this isn't a horrible choice, and with the AI not maximizing training for the armies, it seems like a good one. But I could see better choices being made, such as changing trade policy to export focus or free trade, or picking Schacht since Germany is at this point building only CIC's.
  
  
• 5-1-1936 : AI's second choice for political power is Walther Funk for Political Advisor.
  
  Going forward I will include relevant screens/information in to a single screenshot so you can see what the AI is doing without having to post a separate screenshot each time. In this screenshot I've copied the construction screen information and pasted it to the right of the political screen. Now this choice isn't bad either, if the AI was building MIC's rather than CIC's, but since it is not, Schacht seems the clear choice for political advisor.
  
  
• 8-9-1936 : AI's third choice for political advisor is Schacht.
  
  Nothing to say about this one, other than perhaps it should've been done sooner.
  
  
• 11-17-1936 : The AI's fourth choice is for Military High Command and it picks Schomer.
  
  Trade Laws would be a much better choice, or Research and Production. Even adding a political advisor for +15% Political Power Gain would be worth considering. If we take a step back and consider what the AI has done so far and what we would do differently, I would argue 1st pick : Free Trade Focus for +10% Factory Output, +10% Construction Speed, and +9% Research Time. 2nd pick would be Schacht for +10% CIC, Infrastructure, Refinery construction. 3rd Pick for Guderian as theorist, 4th pick would probably be Krupp for +10% Industrial Research.
  
  
• 2-24-1937 : 2 Screenshots this time based on the choice it's going to make.
  
  In the 1st Screenshot the Manpower is at 908.78k, not horrible, and not pressing to make a conscription law choice at this point in time. But as shown in the 2nd screenshot, it chooses Extensive Conscription, which adds +10% training time to divisions. Which increases the manpower of Germany to 2.57 million, and as we look at more screenshots look at how the manpower changes as I think this choice for changing the conscription law is happening a bit too early.
  
  
• 6-5-1937 : With the Spanish Civil War concluded in favor of the Nationalists, the AI chooses War Economy.
  
  Considering the AI is now constructing MIC's and this policy change frees up 4 CIC's from consumer goods, it's a solid choice.
  
  
• 9-16-1937 : The AI chooses Himmler as a Political Advisor.
  
  Again this choice seems too early, and there are better choices to consider, like trade laws or research and production.
  
  
• 12-21-1937 : The AI chooses Goring for Chief of Airforce.
  
  Again better choices are available.
  
  
• 2-8-1938 : This is an interesting one and I'm not sure if this is undocumented feature or bug.
  
  1st screenshot shows that AI Germany has 75 political power, with NF Anschluss about to be completed.
  2nd screenshot shows that Anschluss has completed and Germany gained +75 political power from it. I approve that there should be a PP gain related to this event as it makes perfect sense. But this effect is not listed when you mouse over the NF Anschluss.
  3rd screenshot shows the AI picked Fritsch for Military High Command. Excluding the normal picks that would be better, I think with the first medium tank being researched, that the AI should pick a Tank focus under research and production.
  
  An observation of note in the 1st screenshot we have 2.25 million manpower available, meaning since we changed conscription laws, Germany has used around 250k from this pool. Now 908k subtracted by 250k means we still have around 650k manpower available proving that the conscription law change on 2-24-1937 is too early. In addition the second screenshot shows that with Anschluss we've increased manpower back up to 2.5 million effectively negating the manpower consumption done since the change.
  
  
• 5-19-1938 : Being almost 2 and half years now without a change to Trade Policy, and the AI having constructed some MIC's to boost consumption of resources, I decided I would see what a Free Trade policy would look like by saving the game and adding 150 PP to make the change manually.
  
  1st screenshot shows that the AI picked Rundstedt for Military High Command, I've also included the Trade Screen to see the current resource utilization values. As you can see, Germany has 96 surplus in aluminum, and 304 in steel, with 13 CIC's being used to import trade goods. Also with 32 days remaining on medium tank research, the AI has another opportunity to pick a tank focus under research and production that it does not do.
  2nd screenshot shows what our trade / resource consumption looks like after switching to Free Trade. It seems clear that the valuation of trade policies by the AI is needing some tweaking. As the AI should be considering the "cost" of a trade policy and this screenshot shows that at this point, and probably any point before this, it would cost Germany nothing for all the gains to be had. I reload the game undoing the trade policy change I did after this screenshot.
  
  
• 8-27-1938 : The AI chooses Beck as Chief of Army.
  
  Again better choices are available, but as we get closer to Danzig or War the benefit of those choices become less and less.
  
  
• 12-5-1938 : The AI chooses Erich Raeder for Chief of Navy.
  
  *insert redundant comment here*

I'm going to stop here since I think I've produced enough evidence to show that the AI is picking some sub optimal choices for policies. I'm also not sure if Historical Focus is forcing some of these choices or not. Feeling somewhat curious I went ahead and created 2 new games, one without historical focus and another with historical focus. Rather then wait and see what the AI does normally, I injected them with the PP at the start to make decisions, and then just monitored the order of its choices, by putting a red alpha numeric character in the upper right corner of the portrait. The order from first to last is " 1,2,3,4,5,6,7,8,9,0,A,B,C,D,E,F,G,H ".

• No Historical Focus
  
  
  
• Historical Focus
  

For the most part it starts out the same for both. Theorist, All Political Advisors, All Military Staff, 3 Research and Production Techs starting with Industry, then Aircraft, then Infantry tech, and then Trade Policy Export Focus. After this is deviates (Oops forgot to use "D" in the historical focus screenshot and jumped to "E". ) with Historical Focus going to Free Trade, Tank Tech, Navy Tech. While No Historical Focus chooses Tank Tech, then Free Trade and then Navy Tech.

I'm not sure Manstein as a theorist is worth 250 points for +3% research bonus over +10% Max Speed from Guderian. I think it should really depend on the AI's strategy, if it's not going to use a significant amount of armor divisions, then I suppose Manstein could be a decent choice. But I have strong doubts about whether the AI would ever wait or have the opportunity to pick Manstein since it always seems to spend 150 PP when able. I also have not seen the AI ever change a minister once assigned.

Another thing that is worth noting is that for the screenshot on 11-17-1936 above, it picks Schomer over Himmler as it's 4th pick, and I imagine the reason why it does that is because of the Spanish Civil War. So clearly there is some context valuation happening on what it chooses for political point expenditures. Also in historical or non-historical mode, when the AI Germany had greater than 500 PP it would try to influence nations with diplomacy. Once it fell below 500 PP it started removing it's diplomacy options until there were none left. Since the AI switches back and forth between Export Focus and Free Trade whenever it has 150 or more points and nothing else it deems worth spending points on, it doesn't seem capable of ever utilizing diplomacy for meaningful effect.

Historical Focus does prioritize Funk over Schacht which may be a bug, especially considering historical focus builds CIC's only for the first year and half or so. Another thing that I wonder is if things observed with AI Germany are universal or not for other major nations and how they spend their political points. It's quite a bit of working breaking down and documenting just one nation to this level of detail, I imagine it's a lot of work for PDS to do the whole game, which does not make me envy the process. Hopefully some of this proves useful for quick easy fixes though.

 

[Gwydion5]

New official patch so a TLDR recap of issues still present in this build.

Game Version : v1.3.2 (e13a) - New Game (as in not loading from a saved game) - Regular difficulty - Historical Focus
DLC : Together for Victory
Mods : NONE (I have never used mods ever in HOI4, nor edited any files)
Buffs : None (except for specific testing like 5.6. 5.7. 5.8. and 5.9.)
Issues Resolved : Schacht is now prioritized over Funk as 2nd PP expenditure. EDIT: Saw another game where AI picked Funk over Schacht as it's 2nd choice. So not resolved.

1. Germany AI does not use army training/exercise to a reasonable degree from 1936-1939
2. Germany AI only sends 1 division of volunteers to Nationalist Spain. (It may do this to manage World Tension / Threat? Nationalist Spain always wins)
3. Germany AI swaps out the commander for the Spanish Volunteers daily. Often 24 hour cycles where no commander is present.
4. Germany AI reorganizes it's army divisions and commanders too often and nonsensically.
5. Germany AI does not spend PP's optimally.
   5.1. It does not switch trade focus between 1936-1939
   5.2. It chooses Extensive Conscription when Manpower is less than 1 million troops. It does this early to mid 1937.
   5.3. It often picks a military theorist as it's first pick, I think Schacht or a Trade Policy would be more appropriate.
   5.4. It is now picking Himmler as it's 4th pick as a political advisor. (Note: It is now picking Schacht before Funk as 2nd pick)
   5.5. It prioritizes Military Staff over Trade Policies and Research & Production.
   5.6. It almost always switches between Export Focus and Free Trade Policies whenever it has 150 or more points, when all policies/positions are filled or optimal ( IE conscription / economy policies are ideal. )
   5.7. AI Germany will not do diplomacy unless PP is greater than 1000. (at least at start of 1936 campaign)
   5.8. AI Germany will cancel all diplomacy once it falls below 150 PP. (Hard to test exact value because of daily policy change stated in 5.6. )
   5.9. AI Germany rarely (if ever?) changes out Policy Advisors or Military Staff / Theorist once picked. No matter how poorly current ministers are.
   
6. Germany AI fails to manage fronts with Czechoslovakia after Demand Sudetenland event.
7. Germany AI promotes Generals to Field Marshals sub-optimally.
   7.1. Generals Promoted have traits when there are Generals of equal rank without traits to be used instead.
   7.2. Germany AI promotes 4-5 Generals to Field Marshal prior to Danzig or War.
   7.3. Of the 4-5 Field Marshals promoted, 3-4 are unassigned when Danzig or War happens.

That's about it for a recap of issues still present in this build. I've observed 2 games and everything listed above has been present and consistent.

My next post in this thread will absolutely be post Danzig or War and focus on War Time AI behavior.

 

[Gwydion5]

Part 5 : Division and Command Changes on Danzig or War

Game Version : v1.3.2 (e13a) - Loading from v1.3.2 (e13a) save game on 8-22-1939 - Regular difficulty - Historical Focus
DLC : Together for Victory
Mods : NONE (I have never used mods ever in HOI4, nor edited any files)

This post is only going to cover a few days. I kind of got sucked in to the stuff I've already posted about regarding fronts and reassignments of divisions and commanders and I think this is a huge issue that really cripples the AI from being competitive. So I guess I'll post about it again, just that this time it will be during war with Poland.

• 8-22-1939 / 11:00 hours : 1 day before the NF focus "Danzig or War" completes.
  
  So this is just an overview of going in to the war. Would like to see the number of tank divisions increased. Also Rundstedt is commanding the western front for Poland with 109 divisions and they are training. Once Germany AI picks Danzig or War NF, it should be better about what it is doing. Like perhaps hard code a stop for training/exercise of armies on fronts with Poland on the 40th day of the NF?
  
  
• 8-23-1939 / 1:00 hours
  
  Danzig or War NF Complete. Some units that are assigned to Rundstedt are not training. I've pointed them out with red boxes. Might be a bug?
  
  
• 8-23-1939 / 2:00 Hours
  
  The AI scraps Rundstedt's battleplans for Poland, decomissions two armies.
  
  
• 8-23-1939 / 3:00 Hours
  
  With Rundstedt's 109 divisions selected, we see a major division realignment among the armies. Rundstedt's divisions go from 109 down to 48, with 61 divisions being reassigned to army's 1, 3, 5, 2 and 6. 2 new armies are created.
  
  
• 8-23-1939 / 4:00 Hours
  
  1st screenshot is a total view of what is happening for all armies except the garrison army.
  2nd screenshot shows that it is actually pulling troops off a front with Poland with less than 24 hours before war.
  
  
• 8-23-1939 / 5:00 Hours
  
  More Spaghetti as units are being shifted along all fronts.
  
  
• 8-24-1939 / 1:00 Hours
  
  War is declared. I should've moved the notification window to Poland so we have visibility of units in Germany.
  
  
• 8-24-1939 / 3:00 Hours (I must've missed 2:00 or nothing happened that was eventful)
  
  A bunch of divisions are moving back to fronts with Poland and for whatever reason the garrison mission went from Kluge with 46 divisions, to Hell with 21 divisions.
  
  
• 8-24-1939 / 4:00 Hours
  
  Now a bunch of divisions are moving back to western fronts across Germany.
  
  
• 8-24-1939 / 5:00 hours
  
  More divisions moving across Germany. Some but not the majority are due to the change of the army doing garrison duty.
  
  
• 8-24-1939 / 6:00 Hours
  
  Division movement and assignment is mostly the same, but the commanders have changed.
  
  
• 8-22-1939 to 8-24-1939 / 6:00 Hours Division/Army command changes in the theater.
  
  One of the things that is hard to visualize or understand is all the changes the AI is doing in terms of division/commander organization for the theater. Going back and forth between screenshots and such is a lot of work trying to assess the changes. So I created a single image to demonstrate the changes. I think it is a bit much, and based on my observations this is happen alot, peace time or war time, the AI likes to change things up.

One thing to consider is that the logic that is driving these reassignments and changes in command, may apply to all AI. I should actually test that before making the speculation. But if there is logic in the AI to match enemy front positions and/or improve its front strength (green/red bar at the bottom of their portraits) then you have to wonder if there isn't a cascade / feedback effect happening to some degree? I don't know just speculating really at what might be driving such changes.

Another thing worth noting is AI Germany needs to come up with some calvary divisions for suppression. Using normal infantry is going to be tough given their manpower constraints.

 

[Gwydion5]

Part 6 : Italy Moving Troops to Ethopia / German Spaghetti with France

Game Version : v1.3.2 (e13a) - Loading from v1.3.2 (e13a) save game on 9-1-1939 - Regular difficulty - Historical Focus
DLC : Together for Victory
Mods : NONE (I have never used mods ever in HOI4, nor edited any files)

I'm positive this has been reported before. But it seems Italy sends troops to northern Germany and then tries to send them Ethopia through the english channel. I'm just going to attach all the screenshots in order and you can scroll through them and look at the dates if so inclined.

• 12-5-1939 : German Front with France.
  
  In these two screenshots we are looking at southern and northern France division movements/allocation. In terms of movement, we have a lot of divisions moving 10+ provinces, some are even being ordered from west/southwest of Switzerland to attack provinces adjacent to Paris. We also have provinces that are stacked with axis forces against the French that are not attacking. Like the very bottom of France/Italy where Germany alone has 2-1 division count against France and empty provinces adjacent to it's divisions. Unfortunately I have 3 armies selected that appear to be orangeish, so I can't make a determination about Rundstedt's aggressive battle plan stance.
  
  
• 12-15-1939 to 12-17-1939 :
  
  In the first screenshot we can see 6 units with no movement orders assigned to Rundstedt with aggressive battle planner stance that are adjacent to empty provinces. Ideally what I would like to see happen is the province with 3 divisions to send 1 division in to each (there is only 2) of the adjacent french provinces that have no troops, leaving 1 division behind for defense and to continue entrenchment. Worst case scenario is they fail and retreat back from where they came, best case scenario they take both provinces which betters the German position for almost no risk.
  
  In the third screenshot we see that the AI has reassigned one of the divisions to Weichs and is sending in to take an undefended province. Division reassignment was probably not necessary, but at least the AI is trying to take advantage of the situation. Weichs is currently doing garrison duty though, so it's a very odd choice to reassign that division.
  
  In the fifth screenshot we see that it has aborted it's attack and as far as I can tell reassigned that division back to Rundstedt's army. I see no reason why it should've stopped given it had a 1 day lead or so into invading that province and I'm skeptical french mobile infantry could've got there in time.
  
  
• 12-20-1939 to 1-5-1940
  
  1st and 2nd screenshots are just showing a view of the front and the movement orders being assigned along it.
  The third screenshot shows that Rundstedt is now executing his battleplan, so France should fall relatively quickly.

Conclusions :

• Italy needs to de-prioritize Ethopia once it is at war with the Allies. It should prioritize taking the Suez and it should put emphasis on having a significant portion of troops already in northern africa for that end. That should hopefully end the Northern Germany to Ethopia route it is taking.
  
  
• Ideally I would like to see the AI improve it's use of smaller (12-24 division) armies. Right now it seems too dependent on huge large fronts.
  
  
• Battle Planner Stances/Policies : I'd really like to see "aggressive" policy stance be more aggressive with undefended provinces. It should not stop moving divisions if there is an undefended adjacent province AND no known adjacent enemy divisions. Regardless if the AI is sending divisions into that province from 1 or 10 provinces away.
  
  
• I'd really like to be able to control front redeployment policy / sensitivity. I get that the AI has some optimal distribution method, but a lot of times, especially in an active war, it can be completely unnecessary. Perhaps we can bake this in to the current stances, where cautious will move troops along the entire front to meet it's idea of optimal distribution. Balanced policy will limit redeployments/movements to half the front. Like if a front covers 24 provinces, it will not move troops more than 12 provinces to reinforce. And aggressive will be like 25% of the front. And that is a possible quick fix situation to the problems I'm seeing. The long term fix in my opinion is understanding that moving 1 division 12 provinces is to reinforce a position along the front is way more inefficient than moving 12 divisions 1 province all at the same time. Effectively shifting the line by 1 in whatever direction it needs.

 

[Gwydion5]

Part 7 : Loading Save Game AI Issues - Garrison Issues

Game Version : v1.3.2 (e13a) - Loading from v1.3.2 (e13a) save games - Regular difficulty - Historical Focus
DLC : Together for Victory
Mods : NONE (I have never used mods ever in HOI4, nor edited any files)

In this thread I've been doing a lot of starting new games, making saves at important junctions, and loading these games to retest certain aspects of the AI. One thing that I have noticed is that on loading a savegame, the AI does a big "reset" of whatever it is doing. You can really notice this with division / command changes, but also Part 5 where AI Germany scraps it's battle plans as well as major command/division reassignments on the day before Danzig or War NF completing. I'm going to break this down in to 3 different parts by save game. Also make note that the AI Garrison orders are often only filled to half of the minimum capacity. So if the minimum capacity is 100 divisions, the AI will have roughly half of the divisions required.

• Part 1 : Loading a Germany Save Game from May 19 1938 and going in to Observer mode.
  
  Starting in December 1939 with Western Europe capitulated I take a monthly screenshot showing the AI Garrison command.
  
  
• Part 2 : Loading a Germany Save Game from May 1st 1940 and going in to Observer mode.
  
  1st Screenshot shows the current AI garrison deployment. Also make note of the command changes that follow.
  2nd Screenshot shows the AI abandoning it's garrison deployment in occupied countries.
  4th, 5th, and 6th screenshot show May, June, July and how the AI doesn't change it's AI Garrison orders.
  
  
• Part 3 : Loading a Germany Save Game from June 19 1939, switching to Switzerland in Observer Mode.
  
  1st Screenshot showing the start of the game.
  2nd Screenshot showing Western Europe capitulated, I also save the game here as the Swiss.
  3rd Screenshot shows me switching to AI Germany in Observer Mode to see the current garrison command.
  4th Screenshot shows a month passing by with the observer switch from Swiss to Germany done in the 3rd screenshot.
  5th Screenshot shows what happens when I load from the December 9th Save Game made in Screenshot 2 and let time advance.

Conclusions :

• Loading from a Save Game breaks the AI's Garrison ability for occupied countries.
• Loading from a Save Game causes major command changes. (Commanders, Armies, Divisions, Fronts, Battle plans)
• AI Garrison only does half of the minimum divisions required.
• Unknown if loading from a save game effects other aspects of AI, such as policy, diplomacy, division design, production, construction etc...
• Partisan Activity cripples AI Germany in occupied territories. I didn't post the screenshots, but I was counting the CIC's needing to be repaired month to month, and even without the Loading Save Game Garrison bug, AI Germany was steadily falling behind. CIC's were being destroyed faster than Germany could repair them.

Attaching 2 save games if anyone want's to test / reproduce.

 

[Gwydion5]

Part 8 : AI Construction Issues

Game Version : v1.3.2 (e13a) - New Game - Regular difficulty - Historical Focus
DLC : Together for Victory
Mods : NONE (I have never used mods ever in HOI4, nor edited any files)

TLDR Summary : The AI only builds CIC's and MIC's if it can't build either of those it will convert CIC's to MIC's. Germany AI has a bug where it won't ramp up construction (Conversion of CIC's to MIC's) until March 1936 / Rhineland. AI will not convert MIC's to CIC's. The AI will start building other things (Radar, Airbases, Dockyards, etc...) in 1940. The problem with that, especially for AI Germany, is that as evidenced in Part 8, it can't suppress resistance enough to get ahead of all the repairs in the queue. The AI will not change the build queue despite being counter productive. The AI will not replace staff/ministers no matter how poorly they are for the given situation. I also did a new game with Russia and the United States with boosting their CIC's at the beginning and they demonstrate the same behavior. In addition I also gave the AI plenty of MIC's at the start to the point that it was 3 short on CIC's to meet consumer goods. It still tried to convert a CIC to MIC that would go nowhere.

Now the long version:

• Germany 1936 January to March
  
  1st screenshot shows the start where I have max Germany on CIC's.
  2nd screenshot shows its first build project.
  3rd screenshot shows that it does not add another build project.
  4th, 5th, 6th screenshot shows the completion of the build project without adding a new build project.
  7th screenshot shows it had queued up 5 CIC to MIC conversions. Incrementing 1 per day.
  8th screenshot shows that it is still adding CIC to MIC conversions to the queue.
  
  
  
• September 1936 to August 1937
  
  1st screenshot shows that the AI is still converting CIC's to MIC's.
  2nd, 3rd and 4th screenshot shows what the AI is building with Dispersed Industry I technology researched.
  
  
• April 1938 to September 1938
  
  I went ahead and took control of Germany did instant construction to max out the CIC's in April to see if the AI would prioritize differently. It did not.
  3rd Screenshot shows the AI building a CIC after Demand Sudetenland NF. Does not cancel the CIC to MIC conversion builds before it.
  AI doesn't always use all the availabe CIC it has when doing CIC to MIC Conversions.
  
  
• January 1939 to May 1939
  
  1st and 2nd screenshot again showing the AI not changing the production queue. Seems to me it should cancel the CIC to MIC conversions if it's building CIC's.
  3rd, 4th, and 5th screenshot is me again maxing out CIC's with instant construction, hoping in 1939 the AI might try to build something else. It does not.
  
  
• September 1939 to February 1940
  
  1st screenshot shows the capitulation of Poland. Repair queue starts increasing.
  2nd screenshot shows the repair queue taking all the CIC's. Nothing is being build. Western Europe capitulated early in January.
  3rd screenshot shows me again maxing out the CIC's in the occupied countries. It might be enough to get ahead of the repairs but it's going to take months to do.
  
  
• May 1940
  
  1st screenshot : Growing impatient I decided to annex the Soviet Union and max out their CIC's.
  2nd and 3rd Screenshot show that the AI is finally building things other than CIC's, MIC's or Conversions.
  Of course the problem with this is that normally Germany will be forever stuck in repair because it doesn't do garrison duty very well. Especially if you load from a saved game after it has conquered territory.
  
  
• Ministers and Staff
  
  One thing I did do was start off AI Germany with the worst ministers I could, just to see if it would replace any of them during peace or war time. It does not. This isn't that big of a deal, considering the AI does a decent job with its selection, though its order/priority could use a bit of improvement. Still it would be nice if the AI could make valuations on its situation and make changes accordingly. Like the silent work horses in the early game aren't a bad option for German if it had a strategy to use them. But the AI for politics seems limited, it only makes changes to conscription and economy when necessary. Trade policy is its goto when it has nothing better to spend it on. It did switch down to closed, then back up to limited, and then back up to export during the last 3 screenshots.

I'll say it again, while the situation I've created by boosting Germany is kind of a non-issue in that normally the AI will not be in this situation. This is still important because it shows the limitation of the AI to make context based valuations of what it wants and what it needs. The AI seems to be lacking desired goals based on strategy, or just goals in general, like having 50 CIC's before switching to constructing MIC's, or perhaps a AI Germany that emphasizes a little more on naval power by building dockyards sooner, or an AI Germany that builds MIC's first or a more balanced policy of 1 CIC 1 MIC. etc...