AI strategy plan: main character syndrome

[bitmode]

If a (human) player knows the basics of the game, a match might go like this: they decide which path of which country to play as, map out a plan of the important focuses and advisors to pick, and off they go. Maybe they mess up at same point and start over or the plan goes to fruition.*

An AI can't function like that. Whatever original intent an AI country has, it *must* operate under the expectation that there is at least one country in the world that messes up other's plans big time. The simple fact of the matter is most countries will not be able to realize their original strategy and have to adapt accordingly. Most often, that will be AI countries.

Yet when it comes to national focuses, the devs decided to walk back the small amount of dynamic decision making the AI had and they literally just work off a single list of focuses from the start up until what is usually the end of the game (1942/43).

Even setting aside responding dynamically to what other countries are doing, some strategies are just much more risky than others for the AI. Sure, South Africa has a path for turning all of Africa into a union of socialist states. Should the AI try to do this in 25% of games, given what it knows about it's own capabilities? I'd argue no.


* I'm not saying everyone always plays like that, but this seems to be what AI strategy plans are modeled on

 

[kettyo]

The game needs a missing option in the first place. To be able to overthrow the ruling system in case it's agenta fails. Somewhat similar to what they are now doing with the balance of power system just on a general level. So in case a country stuck in a path impossible to be advanced upon anymore it could overthrow the regime and go down another path, not being stuck paralyzed forever.

 

[Corpse Fool]

I'd say that this is largely only a problem, specifically because players are given the power within the game to go so completely off the rails. The AI's being rigid and unable to respond to 'unique' situations is only made a problem, when those situations are allowed to exist. I'd imagine that if focuses and such didn't allow quite as much freedom or alt-hist and focused more on whatever predetermined scenario (ww2?), then everything would work a lot more smoothly. If the core devs focused on the main scenario, we could have allowed (third party?) expansion packs or mods worry about generating alt-hist and other scenarios, which seems like a win-win to me.

They once argued that they didn't put a communist branch in the chinese/warlord focus trees, because if you wanted a communist china you should play, communist china. I'm not sure why they haven't tried to extend that line of thought. Unless of course that line of though was just an after-the-fact attempt to come up with any argument at all to justify the decision, with out much thought to what else making that argument means.

 

[TheMeInTeam]

I'd say that this is largely only a problem, specifically because players are given the power within the game to go so completely off the rails.

In a game, you can have rules to abstract history wrt the mechanics you want, or you can have rails. Unless you remove player agency entirely and make a movie rather than a game, you can't have both.

When those things intersect, you get broken. Not in a subjective sense, but in an objective sense. Either the rules or the rails must break, because they are mutually exclusive.

Even if you start the game with all the relevant factions in place and pre-made divisions at WW2 start (skipping all buildup), a player controlling divisions under the present rules of the game can create massively different outcomes than historical through destroying divisions and getting more score. Now you must disregard that performance and make a "historical" peace deal anyway, or we're off the rails again.

Some of the worst interactions in HOI are when random focuses supersede gameplay rules, often w/o warning to the player. It happens in RT56. It happens in Kaiserreich, though they tried to "fix" some of that by blocking "unintended wars" before ww2 (aka you play a choose your adventure book rather than HOI for 3 years as many nations). And unfortunately, it also happens in vanilla. M-R usually counts, except when it doesn't and USSR can freely declare on Axis members w/o Germany getting involved. Puppets usually need civil war or liberty desire to get free, except when they don't. You usually need to fight for land, except when a focus takes it from player country w/o them being able to opt to fight for it. Then you also get crap like "Yalta conference" where player who wasn't even in on comintern or allies gets Germany stolen on some patches, ignoring the actual peace deal made.

The more rails a game uses at the expense of its rules, the more BS outcomes like this we'll see.

There are a few other issues here:

• You can't "win" the game per the game's defines. The player will pick a "win" condition and pursue it, but the AI has nothing for which to optimize. There's nothing like "conquer all capitals" or "launch the spaceship" like we see in Civ or whatever. That might well be fine, but it's worth pointing out that in order to have a "long-term strategy", you actually need a goal for the concept of having a strategy to matter. AI in general is bad at long-term strategic planning, but creating an environment where that is literally impossible by default does not help!
• Focuses are rails by default, but insufficient regard has been given for how their evolution/additions interacts with other focuses or especially with core mechanics. Many nations can justify wars, in fact it's not clear why any non-democratic countries need to wait.
• What "should" happen when nations deviate from history is not something players will agree on. We can't even get people to agree on what the word "spearhead" means, or which factors actually gave USSR bargaining power after WW2. There are players unironically willing/desiring degenerate game incentives just to get outcomes that look more like historical outcomes, regardless of the process that gets the world to that state being inconsistent. That's an incoherent preference, but it's also a common one. Rails that create these outcomes can and must ignore or destroy player agency to get them.
• Dev design vision itself has shifted over time/been inconsistent. You can't do communist Chinese warlords, but you can do communist Japan. The penalty for micromanaging vs planning bonus was an objective deviation from the stated design intention of the battle planner. Armored cars were added, then obsoleted by a strictly dominating alternative and left in that state for a long time.
• HOI is reliably terrible at relaying when railroading will interfere with agency, and along with the meme controls this is one of the big sources of frustration in the game. Players get upset when they struggle for any reason. But in terms of design, players should struggle when they make a mistake, not because the developers made a mistake. This has memed to the point that in many cases, it's hard to discern unintended bugs from intended mechanics that rail their way past gameplay rules.
  • Compare Taureor's "order 66" to "civil war fires 300 days earlier than indicated, deletes half your army, and on top of that spawns units from nowhere in contrast to default civil wars". It is okay to telefrag countries with magic units, except when it isn't. Both or neither of these *should* be intended (neither is the reasonable answer), but patch behavior tells us otherwise.

 

[Corpse Fool]

In a game, you can have rules to abstract history wrt the mechanics you want, or you can have rails. Unless you remove player agency entirely and make a movie rather than a game, you can't have both.

I suppose this depends what either of us means by 'rails'. If we're talking strictly about the single track sequence of events sort of thing, then yeah. But if we're extending this train analogy, I find it's more the difference between supported and unsupported alt-hist stuff.

What I mean by that is I suppose players do have the control to take the train wherever they want (through certain mechanics like generating war goals or whatever), but trying to take a sharp left when the track goes straight will often lead to an uncontrolled disaster of a derailment (often regarded as an undesirable experience), rather than simply having the train jump the track and continue to chug along. We have to actually support the train with the rails, being supported by the ties, being supported by the ballast, and have a switch for the train to seamlessly transition onto the new track and be able to have it go in that direction any reasonable distance while maintaining a comfortable experience for the passengers.

What I meant by going off the rails wasn't WRT the specific 'main scenario', it was venturing into any unsupported territory and derailing the train. You can deviate from the main scenario, but you just have to support those outcomes. The more deviation you allow, the more work you have to do in supporting everything so that the user experience doesn't degrade. The devs don't seem to have given all of these possibilities an 'adequate' amount of support, which can (has) led to some confusing and frustrating situations that players occasionally come to the forums to complain about.

I'm not sure how well this communicates what I want it to, but I'm just going to hit send and see what sticks.

 

[FindFloppies]

If you remove the alt-history options from the game, you'll remove a lot of players as well. Simple as that.

I would get bored in a couple hundred hours if the 'historic' chain was the only choice to play.

So, when people rant about alt-hist choices being horrible things, I think they're asking for the life of the game to be shortened considerably.

YMMV.

 

[TheMeInTeam]

I suppose this depends what either of us means by 'rails'. If we're talking strictly about the single track sequence of events sort of thing, then yeah. But if we're extending this train analogy, I find it's more the difference between supported and unsupported alt-hist stuff.

It might be true that "supported" (as in, there's a focus to do X) has somewhat less probability of badly interacting with another focus than if you justify and declare random wars. I haven't tallied it. But you can for sure get unforeseen/bad interactions using only "supported" (implied by doing stuff through focus) actions. And that's not surprising, if you don't constrain what focuses are allowed to do or interact with, a designer is stuck trying to think of every possible contingency in advance. That's not a task human beings can do. It's bad process and will set pretty much anybody up for failure.

What I meant by going off the rails wasn't WRT the specific 'main scenario', it was venturing into any unsupported territory and derailing the train. You can deviate from the main scenario, but you just have to support those outcomes.

IMO in a design context it has more or less been proven that the means whereby you can jump tracks and keep going is using base rules, not by trying to perfectly align focus interactions which create exceptions to rules. Not proven in the sense of scientific papers, but in the sense of "many different teams of devs, some vanilla and many more in mods have tried to "support" contingencies through focuses, and all have failed to varying degrees". The degree and frequency of failure tracks pretty closely to what focuses allow and don't allow when designed.

The more contingencies that need to be specially addressed, the more likely you're going to generate a run that screws the player over somehow. When you have 20 countries that can each take 1 of 3 paths, some of which have sub paths, and these all interact with each other there are too many permutations for an actual human to consider. If, by design, you constrain the ways these interact such that they are not contingent on other tree focuses, it works so much better. And when it comes time to push for territory, well...this *IS* a war game, right?

I just don't think "adequate support" is a thing Pdox (or any company made up of actual people) can attain with the way some focus trees are implemented. I also don't know how the AI could be taught to adjust to the game state/other focus trees selected, unless we can define a goal towards which the AI strives/tries to optimize to reach.

Some of this is outside the scope of what HOI 4 can actually do at this point. IMO focuses could get reworked to be blocked/bypassed if target(s) of land steal are at war (or just give a war goal then), some of the weird restrictions on justification could be eased, and then the AI could either pick historical focuses (if that's set) or whatever if it isn't defined for it. After which it still tries to win the wars that result, to the best of its very limited ability. That's about what I suspect we could REALISTICALLY see from HOI 4, and a lot of people would enjoy that (many players enjoy the game now, so improving its focus interactions/control/AI and not destroying the part of it people enjoy should be help in a strict sense).

So, when people rant about alt-hist choices being horrible things, I think they're asking for the life of the game to be shortened considerably.

It depends who is complaining and why. I recently experienced a scenario (I think it was vanilla ironman Tannu Tuva attempt, non-historical focuses) where one of the Chinese nations first joined CUF, called all of them in, and then during same war switched factions to allies and called them in too (???). Maybe that's a bug, maybe it's an intended interaction. I don't know. It's hard to tell in HOI. The problem with that isn't that it's not a historical path, but rather that the AI randomly used focuses to break what is normally a hard limit on what you can do with your faction alignment mid-war. It also obviously drastically altered how the game would progress from there, as Germany went oppose Hitler so I'd have been in a forever war + have to deal with government in exile in addition, all while also still fighting CUF and probably Japan.

There are many tags where a single interaction like that alters how the rest of the AIs in the entire world behave, and more with each patch/DLC cycle. Because focuses allow for rule bypasses like that, the odds of encountering one the moment you deviate from historical settings shoots through the roof. But focuses don't *have* to break established rules or hit you with "gotchas".

 

[Corpse Fool]

If you remove the alt-history options from the game, you'll remove a lot of players as well. Simple as that.

I would get bored in a couple hundred hours if the 'historic' chain was the only choice to play.

I'm not entirely convinced there would much of a dip. I'm not really sure what the demographics look like, but I don't imagine there are many people that would strictly play vanilla without mods, and only play hoi4 because of the alt-hist paths it contains. If you're looking for alt-hist, I would say there are a fair number of fairly well developed mods out there that can scratch those itches better than what vanilla does. I suppose the question hasn't really come up that often for those sorts of people to reveal themselves, but the people that are dissatisfied with the supposed ww2 scenario in vanilla or modded hoi4 have made themselves known a couple of times of the forum.

I just don't think "adequate support" is a thing Pdox (or any company made up of actual people) can attain with the way some focus trees are implemented.

Yes, this is something I was going to mention but cut from the final draft for brevity. Any team basically has a finite amount of resources to put towards a project, and the more different ways you try to split those resources, the less resources any particular part of the project is going to get. This is why I advocated for the devs to focus more on doing the ww2 scenario really well, and give the modders/third party expansionists the tools to realize their own scenarios.

It might be true that "supported" (as in, there's a focus to do X)

I don't think that simply having a focus to do a thing can be considered adequate support, but I didn't have the 'adequate' qualifier at the point in the post that you're responding to. Yeah, running two beams of metal that aren't quite shaped like rails or are a different/inconsistent gauge without any of the ties/ballast could technically bring that train further down that path than you otherwise would. But that isn't really a train that I'd want to be on. You need a combination of things all working in tandem to have everything run smoothly.

 

[Dryhad]

I agree somewhat but I think you are barking up the wrong tree in appealing to the old "dynamic decision making" system as you refer to it. You know as well as I do that that system was just as blind to the realities of a playthrough as any other, the only real difference is that under the current system, the AI at least isn't also blind to its own actions (mostly).

You can switch the underlying system all you want but the actual problem is that the AI needs to have its contingencies covered. That wasn't the case when each focus was evaluated independently and it's not the case when the AI picks a strategy plan to go with. I recently did a stocktake of the abort conditions and alternatives for historical AI strategies and it's pretty inconsistent. Spain and France have pretty robust reactions for Germany and Britain going off the rails, although France needs some updates to take into account changes to its communist focus tree (and, moreover, the logic of the Oppose Hitler scenario was disrupted by the new French focus tree and never fully amended - Historical France will always take the Popular Front first meaning that the scenario for fascist France vs Democratic Germany cannot happen). The two Chinas have some limited alternatives, Britain will fumble around if Germany opposes Hitler before committing to a plan based on which ideological path it finds itself on, and Mexico will abandon its historical plan if it finds itself flipped to fascism or communism, but curiously will only pick an appropriate alternate plan if historical mode is on. Nobody else has meaningful alternative selection criteria, and many do not have meaningful abort criteria. Any system is only as good as the use it is put to. The AI could be made much more responsive simply by scripting new abort conditions and associated strategy plans. It wouldn't even need to be done en masse (in contrast to completely replacing the underlying AI focus decision making system), devs could literally script one contingency at a time, as time and ability permitted. This is what I would prefer to see.

 

[bitmode]

You can switch the underlying system all you want but the actual problem is that the AI needs to have its contingencies covered. That wasn't the case when each focus was evaluated independently and it's not the case when the AI picks a strategy plan to go with. I recently did a stocktake of the abort conditions and alternatives for historical AI strategies and it's pretty inconsistent.

You are right that much more content (i.e. triggers, aborts and weights) is needed, but I do think the system being used plays a role in that. Just to name two points:

• Randomness is a useful tool to cover up holes in an incomplete AI. Aside from the initial selection with historical focus off, strategy plans don't have randomness, so the conditions and sequencing need to be well made. The only plan that the AI reliably can pull of is the historical one and that existed before the AI plan system, just as a simple list. I think in a game this complex it is often easier to iteratively shift the balance on a die roll rather than clearly define a particular contingency.
• Covering more contingencies will lead to more and more plans. It will becoming increasingly harder to keep their conditions and content in sync.

 

[Dryhad]

• Randomness is a useful tool to cover up holes in an incomplete AI. Aside from the initial selection with historical focus off, strategy plans don't have randomness, so the conditions and sequencing need to be well made. The only plan that the AI reliably can pull of is the historical one and that existed before the AI plan system, just as a simple list. I think in a game this complex it is often easier to iteratively shift the balance on a die roll rather than clearly define a particular contingency.

I'm a bit confused as to where you're coming from here, because you're saying at the same time that it is necessary to introduce randomness to cover the AI's flaws, and also that the AI can reliably pull off the historical script. Perhaps I'm misunderstanding you, but in my view the only perspective from which that makes sense is one that internalises lower expectations for historical mode: This happened in history, therefore it happens in game, no matter what else is happening in game. The same problem of the AI ignoring world events to follow its script exists, except the "problem" has been redefined as a goal to be achieved.

Besides this, you're glossing over the problems the AI faces will analysing everything in the moment. Randomness can be a useful tool but focus trees don't lend themselves to true randomness. The die roll is only ever on the focuses available to the AI at any given point. It can't understand the concept of completing one focus to unlock another and it can't understand the concept of delaying a focus until an alternative becomes available. This is why you used to see Hungary form the Pact of Rome all the time, because it would get to a point in its tree where it could either Renounce the Treaty of Trianon or Renew the Rome Protocols and it wouldn't have enough fascism support to do the former so it wouldn't even analyse the possibility of taking it. The complexity of the game, and of focus trees in particular, argues against a system of dynamic analysis imo.

• Covering more contingencies will lead to more and more plans. It will becoming increasingly harder to keep their conditions and content in sync.

Perhaps, but there are ways around this and cheats that can be employed. Contingency plans could prioritise submissive or non-interfering options. Behaviour could be dependent on who the player is. A few Chinese strategy plans consist only of weightings rather than a full script, you might look at these as something of a compromise between both systems. Despite everything else I've said, I do think there's gameplay value in having the AI be less predictable, so some more of these might not be out of place.

Also, as I said above, I maintain that this issue also applies to the analysis at the focus level - perhaps moreso. With a strategy plan, the AI has an end goal and can be guided by the hand to reach it, with fully dynamic selection the AI has no real concept of where it's going or where it's been, and it will treat a focus that will unlock in an in-game week as equally unachievable as one that is hard-blocked by previous decisions. You would need to introduce contingencies for these just the same as for broader strategy plans, and there are many more focuses than plans.

 

[Voigt]

For my improved AI in IHMP I mostly work with historical paths, and even there I added some contingency, since we allow slight historical deviations as a group, and even that was quite the undertaking to script.

For one Japan and China. To improve the performance of China AI I scripted the AI of China, Communist China and the Warlords to make peace quickly and together prepare for the invasion of Japan and form their faction sooner than later.
If Japan now goes down Embrace Kodoha, instead of Purging them, I also expect them to follow the lore of that branch and sign a NAP with China. This means I can let the Civil War in China flame up again, but to make it more interesting, China needs to ignore Communist China for a while, while the Communists attack and integrate Shanxi and Xibei San Ma. Sadly communist China utterly fails to take them over on their own, so I had to fudge the AI and force Shanxi to put all their divisions on the border with Mengoku/Japan/China, and keep their border to Communist China completely open. Even in war, over half their army is stuck to neutral borders, so Communist China can take them over consistently.

This overall was one of the easier paths to do, the much worse thing is Spain. If Republicans Win by Soviet help, they become a puppet of them, which protects them from the Axis while the Molotov Pact is still active, but when it is broken, Spains fate is also sealed, and with that Gibraltar. So I want them to revolt, but if they revolt to soon, Soviet will just crush them and annex them. So I let them revolt at the same time as Germany attacks Soviet Union. Again I had to cheat a bit and let Spain check Germany's active focus even when Spains Intel wouldn't allow it normally. Soviet Union can oppose their independence, then Rep. Spain should join the Axis, or they let them go, and then they should join the Allies without join the war on themselves, just giving the Allies the chance to quickly station troops, before Germany turns around and attacks the now independent Rep. Spain.

If Nationalist Spain wins, it is a bit more simple on the outset, since Nat. Spain just stays neutral and nothing else. Until I added a small diplomacy minigame, where Germany can try to influence Spain to join the Axis, which they will do, unless UK influences Spain to stay neutral. But if they influence Spain, they are locked out of the now mutually exclusive focus to invoke the 1337 Alliance with Portugal, to stop the Tungsten trade or fortify Gibraltar against Naval Invasions.

I worked for quite a while on those Spanish paths, and they are still not fully bugfree. I can't really imagine making all that effort for one far more ahistorical paths those nations themselves could take, or even all the other nations that also exist.

For Paradox, maaybe if they had a dedicated AI scripter, only and only for those interactions it could have worked, but now it is probably to late for that.

 

[bitmode]

but in my view the only perspective from which that makes sense is one that internalises lower expectations for historical mode: This happened in history, therefore it happens in game, no matter what else is happening in game.

In some places the devs deviated from what happened historically as a result from testing, like the Phoney War. So what I meant to say: in contrast to the other AI plans that were introduced, the historical plan (especially the situation of ALL countries using the historical plan) receives some continuous (re-)consideration. And that's needed for a rigid plan to keep making sense. I just don't see that happening for the other plans.

The die roll is only ever on the focuses available to the AI at any given point. It can't understand the concept of completing one focus to unlock another and it can't understand the concept of delaying a focus until an alternative becomes available.

It does not currently understand that automatically. But for one it is possible to include them manually in the weight of a focus. And it wouldn't be hard to change the AI to calculate the weights of all unfinished focuses and fold those into the weights of the available ones based on relation.

This is why you used to see Hungary form the Pact of Rome all the time, because it would get to a point in its tree where it could either Renounce the Treaty of Trianon or Renew the Rome Protocols and it wouldn't have enough fascism support to do the former so it wouldn't even analyse the possibility of taking it.

In my opinion, Renew The Rome Protocols simply had a too large weight, it should have been close to negative. The branch is based on the premise that Hungary of all countries prevents Anschluss by giving a guarantee. Without adding a lot of checks that detect an unusually favorable situation, that's just a suicide plan even if no mutually exclusive alternative existed. The AI also lacked the most basic checks like Austria and Italy still being independent countries.

The complexity of the game, and of focus trees in particular, argues against a system of dynamic analysis imo.

I'd argue on the contrary that one of the main purposes of the focus tree is to streamline complex choices for the AI and new players. I.e. let's say a German AI picks focuses completely at random, except starting new wars while still having fronts in its home area. It is not perfect and somewhat slow but very well could reach its goals because the focus trees are set up like that.

Contingency plans could prioritise submissive or non-interfering options.

I'd be interested to see something like that but I don't think the plan system is designed for it.

With a strategy plan, the AI has an end goal and can be guided by the hand to reach it, with fully dynamic selection the AI has no real concept of where it's going or where it's been

But that's my main point, as stated in the title: the AI reaching its end goal to me is a nice-to-have at best. Renewing the Rome Protocols as Hungary is a possible goal. But doing whatever works and doesn't likely get one killed in the current situation trumps having a goal. Pursuing whatever goal and making it work at the same time is a luxury that human players have.

 

[Dryhad]

In some places the devs deviated from what happened historically as a result from testing, like the Phoney War. So what I meant to say: in contrast to the other AI plans that were introduced, the historical plan (especially the situation of ALL countries using the historical plan) receives some continuous (re-)consideration. And that's needed for a rigid plan to keep making sense. I just don't see that happening for the other plans.

Ok, that is fair and that is more or less a description of the current state of affairs. But any improvement is going to require constant attention, the remedy is to have a custodian team taking a serious look at AI strategy plans on a regular basis. Switching to the old system will not eliminate the need for this, it will at best shift the type of work such a team would be doing (and, I maintain, greatly increase the amount of work they'd need to do).

It does not currently understand that automatically. But for one it is possible to include them manually in the weight of a focus. And it wouldn't be hard to change the AI to calculate the weights of all unfinished focuses and fold those into the weights of the available ones based on relation.

Of course it's possible to include it manually in the weight of a focus, it's just really tedious. You are the one arguing that it is implausible to cover sufficient contingencies via strategy plan abort and enable conditions, yet you think it is plausible to encode the same kind of contingencies in every relevant focus? My argument has never been that this system cannot be made to function as you would like, my argument is that doing so would be much more difficult, and much easier for a mistake to break the intended functionality. The idea of having subsequent focus weightings impact the prerequisite focuses dynamically is an interesting one, but I feel like it would make the task of setting weights in the first place very difficult to navigate. Plus it would also undermine the advantage you see in that system, that of the AI selecting focuses on the basis of what is happening at that instant. If weightings are set up to facilitate rapid reaction to changes in the gamestate, but also to filter up to prerequisite focuses, you have weightings designed for immediate reactions (e.g. checks on who is an ally or potential ally) having an impact on preparation focuses months or years prior.

In my opinion, Renew The Rome Protocols simply had a too large weight, it should have been close to negative. The branch is based on the premise that Hungary of all countries prevents Anschluss by giving a guarantee. Without adding a lot of checks that detect an unusually favorable situation, that's just a suicide plan even if no mutually exclusive alternative existed. The AI also lacked the most basic checks like Austria and Italy still being independent countries.

It's not the only example, just one that occurred to me recently and was fresh in my mind. The issue is not why the AI's typical course of action was stupid, the issue is what caused it to act stupidly. Your baseline proposal is to go back to the old system, where this kind of thing was endemic, and you are making the argument that this would be preferable even without these whack-a-mole bandage solutions to specific focuses. The strategy plans solve this in the general case, not in specific individual cases.

I'd argue on the contrary that one of the main purposes of the focus tree is to streamline complex choices for the AI and new players. I.e. let's say a German AI picks focuses completely at random, except starting new wars while still having fronts in its home area. It is not perfect and somewhat slow but very well could reach its goals because the focus trees are set up like that.

And if a Mexican AI picks focuses completely at random then it gets into one civil war or another pretty quick. There is a good point that we haven't been discussing that focus tree design plays a role, but main character syndrome falls out of focus tree design because focus trees are designed first and foremost for players to have fun. I don't anticipate that changing, so I think it is better to look for solutions elsewhere, but there are some cases (*cough*China*cough*) that could perhaps benefit from some AI-oriented options being added.

Regarding Germany, I put it to you that there isn't an issue in your hypothetical because Germany is supposed to be the main character and the focus tree's design reflects that.

I'd be interested to see something like that but I don't think the plan system is designed for it.

It is designed for exactly that! The first implementation of the strategy plans was to have France switch to a "holding pattern" if Germany Opposed Hitler and then to subsequently choose an appropriately opposed ideology once Germany committed to a path after that. It is absolutely possible to script plans for countries to turtle up, or seek to join the player's faction, or to take no substantive focuses and presumably not impact the game much. The specifics will depend on the focus tree but it's definitely what the plan system was made for.

But that's my main point, as stated in the title: the AI reaching its end goal to me is a nice-to-have at best. Renewing the Rome Protocols as Hungary is a possible goal. But doing whatever works and doesn't likely get one killed in the current situation trumps having a goal. Pursuing whatever goal and making it work at the same time is a luxury that human players have.

Well, this comes back to focus tree design largely, that most alt-history options are designed for players to enjoy rather than for the AI to randomly attempt. No matter how you set up AI focus selection this is going to be a problem. But that's also not what I mean by a goal. I don't mean a capstone focus of a tree, or some Turanist world conquest or whatever. I mean quite simply that strategy plans allow the AI to be guided through a series of several focuses to achieve something. That something could be exactly what you want: whatever works and doesn't likely get one killed. I also think you're putting a lot of faith in focus tree design and/or the proliferation of prophylactic weightings if you think weighted random selection is going to remotely approximate "whatever works and doesn't likely get one killed". I bring up an example of how this was clearly not the case and you just casually assume it will be fixed in script. But the strategy plans are a much more powerful tool for fixing this, and much easier to implement.