Quote:I mean with the conditions, apart from the year.
Calling the procedure means executing the whole thing, including the check for the year (turn count) so this wouldn't do what you want it seems.
Quote: And on the next turn the town is left empty? Super strange.
The nagas were pulled for doomstack creation or they were the doomstack themselves in the first place) - that can leave things empty. It's very noticable on nodes - early naga stacks clear them out but nothing is left for garrison so it's a "free-for-all" node where all AIs send spirits and meld until someone actually manages to push a unit on it to "claim" it. However, we seemed to agree it's better if the doomstack can keep going after a conquest instead having to stay as garrison so nothing can be done about this.
Quote:An RVL use shows that this AI already has 23 nagas of which 10 magic focused despite having no nodes. The cheating really needs to be toned down, at the beginning it's ridiculous.
If the purpose of the cheating is to fill up garrisons, why don't you base it on the number of the cities?
What difficulty was this? Also remember 10S books reduce the casting cost of spells by how much is it currently, 16%? 12? I don't remember lol, we change things too often.
No, the purpose of the cheating is to give the AI troops to be able to garrison AND fight with. That amount does seem a bit excessive although not sure how many turns and what difficulty...
The AI advantages are in a constant table where the row is the difficulty and the columns are the various types of bonus. It's not a variable amount (although for some of the easier ones like gold and mana it might be possible to ignore the table and hardcode a calculation for a variable amount. For the casting advantage that's not an opton because the value is used in a dozen places)
Year of surrender - that doesn't sound too bad a change though. You could move the check of the year out of the surrender system and add it to the turn procedure. Otherwise, the check of the year, in its current location, could be modified to add: OR (all the wizards banished/defeated). Of course I have no idea of the complexity of things, some apparently really easy things end up being super complicated.
Test game - master, around turn 50. I went back to 24, the turn of its attack to the halfling city, and it had 13 nagas and 5 focus magic, plus a decent amount of swordsmen. I can't know it lost any, which is possible as it found a druid somewhere. Just think of the maintenance of this, without any node... And the AI even has the luxury of 300 mana stashed ^_^'
Casting advantage means skill? Well, that shouldn't be so bad, limiting mana and gold could already fix the issue. I'd give the calculation a variable component (x per city) and a fixed component (+k, for the attack) to ensure that there's an advantage for the garrisons but also something left for the attack. With the good algorithms you've put in place for doomstack creation and targeting I don't think that you need a huge advantage anymore.
Next test game:
Good race, quick start realms, I'll go with master for the sake of it. What should be my win chance?
Overland casting is definitely too high. To cast 33 spells with a cost of 80, the human would need ~2640 overland casting skill (spread out over 3 years, means about an average of ~73 casting skill that whole time - far beyond what the human can get in the early game without nodes). The AI, even on expert, only requires ~1584 casting skill (spread out over the same 3 years, the AI only needs ~44 average overland casting skill, which is at least a possibility in early game.)
However the combination of gold/mana/maintenance is the other one that I think gets noticed a lot. 23 naga and 10 focus magic requires the human 66 mana per turn. Without nodes, that requires 3 mostly fully developed cities, even with high elf population bonus. However, for the AI, its much less (I think its 33?) and the AI is smart enough to use alchemy effectively, and the AI gets huge power and gold bonuses, so instead of needing 3 fully developed city, the AI, even on something like expert, probably only needs 1 reasonably developed city.
But then you combine these two things; if the AI has 2 cities, they suddenly have power production to spare for things like increasing skill, thus allowing them to get higher casting skill faster than the human can, even assuming that the maintenance is in a good place. Which allows the AI access to the overland casting skill they need.
Basically, the combination of cheating bonuses is working exactly as expected, but the AI has gotten smart enough, that it isn't using those cheating bonuses to maintain large armies that often don't have any synergy. Instead its started to use decent synergy, its good at doomstacks, and so the cheating bonuses are now being used to overwhelm the player early.
Note, if a human can get an advantage at all, they tend to do fine, because they're still more efficient. But the flat nature of the bonuses mean the AI gets way ahead until the human can get that advantage.
This is why I play node centric games - they let me get ahead of the AI, in a way the AI can't match.
Its also why razing cities feels so much more effective than conquering. If the AI steals a city, its 3+ times as effective as when the human steals one. So if you conquer a city, keep it for ten turns, then the AI conquers it back for 4 turns, then you raze it, the AI is ahead. So, if you're not already winning, its virtually always safer to just destroy every AI city - which has the huge advantage that you don't need to garrison it (since you have far fewer units than the ai anyway.)
Which is where my playstyle of 'get just enough cities that you can support node garrisons, then raze every AI city on both planes' came from - and why I win those games when I have 1/5 the cities of my enemies even if I declare war on all 4 at once.
Quote:You could move the check of the year out of the surrender system and add it to the turn procedure.
That would make the astrologer update only once every 4 turns, so that's not an option. Remember the astrologer data update and the surrender system are in the same procedure.
Quote:Just think of the maintenance of this
Don't. The AI pays very low maintenance. It's not a concept it can handle at all since it can't hold back on producing units once the costs are too high (and it needs to have more units than a human anyway - doomstacks definitely improved it a lot but it's still far less efficient at using units on the overland map). Ultimately, the AI can maintain about 5-6 times as much as the human on Master from the same amount of cities/nodes and they need that.
Casting advantage means reduced overland casting costs, in particular Expert and above pays 50% of the normal cost, Advanced pays 66%, Normal pays 80%. This makes up for the AI not playing a coherent strategy but having to do a little bit of everything they have.
Quote:but the AI has gotten smart enough, that it isn't using those cheating bonuses to maintain large armies that often don't have any synergy.
No, it still does that. Except it's literally impossible to produce those large armies of normal units in the early game so you're only seeing the summons of it - especially because the production bonus and population bonus has been toned down most in the past-, and those are (sometimes) good as a strategy on their own. If the wizard plays Chaos and has Hell Hounds in that quantity, it suddenly isn't looking that smart with those 5*9 hounds doing nothing at all around their capital. If those Focus Magics would be on those Hell Hounds, or some Swordsmen, it also wouldn't make a doomstack and would just feel like a bunch of worthless units.
The difference is, thanks to doomstacks, sometimes, some of the AI's resources do get pushed into what resembles a working strategy - but this is entirely unreliable and up to chance. If the AI only knows Nagas then sure, but if they also know Hell Hounds and War Bears then they only get 1/3 as many Nagas...
Also, while Naga doomstacks do look intimidating, if the player has the counter to them, it's just as wasteful to produce them as any other summon. (It only takes like, 1-2 gargoyles or a stack of ghouls with darkness to take out such a doomstack without losses. Ranged units work too, albeit not as well as when nagas has 2 moves.)
In the end, the AI is still only capable of spamming the same "tactic" over and over again, instead of adapting to the situation and trying something else if the first attempt fails.
I think where we disagree is on scale. Due to doomstacks, I don't think the AI needs 5-6 as many units anymore. I think 3-4 on master would be fine. This would have the advantage of reducing early game craziness if the AI does get a combo like mono sorcery.
While they can maintain 5-6 times as much, that doesn't mean they can also produce that many. If looking at the bonuses relevant to producing/summoning units, you'll find it's around 3x probably even less. The maintenance becomes relevant if the units keep getting produced with no wars to use them up - we don't want the AI to collapse and lose the game due to its own units and having constant zero or negative income does that - it prevents casting overland and combat spells entirely. So the maintenance bonus is the most significant of all categories.
Also, this isn't just the units. Buildings are an even greater risk : The AI can't "disband" them if the gold cost total is higher than the income, so they'll disband units instead. So if the AI has much more buildings than what it could maintain, the end result is they lose all normal units while they build even more buildings...and while this doesn't sounds a likely outcome, think about it, how much does all buildings in the game cost to maintain total, compared to the income from a pop 7 desert city? The AI has to be able to pay that.
Right, but the most likely situation where that will happen is the case where the human has both armageddon and doomsday in play. (And I'm up to over 3 years of both spells in play, so its not like I'm just getting unlucky and not affecting their cities much.) Yes it CAN happen naturally but even on dry poor maps, its extraordinarily unlikely.
So what you're really doing is saying 'if the human player is using the most powerful economic destruction spells together' (and throw in chaos rift on every single city too), then the ai should he able to.. Keep paying for it anyway.
So I understand your original goal, but I believe you've gone too far. You've made it impossible for the ai to have a bad ecenomy, ehich makes any spells that attack economy literally useless to the player (*cough death/chaos get royally shafted*) and leads to the early game AI being extraordinarily powerful if they get the right combination of factors (such as mono sorcery. This is not a rare occurrence, nor should it weaken sorcery by a HUGE amount to GET the ability to summon war bears.)
As a further note, my chaos game with all the very rare globals (except chaos surge.. 8 known very rates, no chaos surge) isn't even on lunatic. So the invincible economy isn't because I'm playing max difficulty.
Let's not drag armageddon or doomsday in the discussion. The problem is that at turn 50, there's such an advantage of units that it's simply impossible to have a reasonable assurance of victory. I've played the next game till 1404 and it's the same. The AI is good enough with stacks, and the player can't do much to beat it without using the usual overpowered strategies - some of which you've just named...
By the way, take a look at this case, speaking of doom-stack targeting. The yellow AI loses 2 cities to the purple AI. I was happy, for the first time I saw a meaningful war. The 2 cities had 2 and 3 war-bears left in them, and the yellow had a nice naga doomstack wandering very close. Cool! I'll just wait that it gets the 2 cities back and then strike - I can get in a wizard's pact with the other one. Right?
Wrong. The doomstack ignored the 2 cities, despite them being at 5 and 9 distance from the capital!, went for a node, then a neutral city much farther away, then a node even farther again away and finally... A city of mine, with 6 units as garrison, and we're not even at war. I am now convined that there's still some bug in the targeting, this is completely unreasonable - 2 closer cities at war with smaller garrison? Why ignore them?
Fix this and I bet that we can reduce the cheating. In the meanwhile I consider this game another loss, it's the third in a row, one at expert and 2 at master. This one was with nature, and despite a good start - I cracked a couple lairs and took a neutral klackon city, plus 2 heroes offered their services - I was soon after blocked by the disparity of resources available.
To go back to the objective of this testing. Is it to nerf the OP strategies and then nerf the cheating, making more strategies available and the game more diverse? If not, what are we doing?
Edit: and back to the game I just posted, another perverse effect of the cheating that can be observed here is: if an AI loses a city early on, it gets stronger, because it has the same resources but needs to waste a lot less on garrisoning.
I don't remember ever mentioning Armageddon or the like.
The AI builds units and buildings faster than its population grows, and builds everything while the player picks useful buildings and does not build the rest. So the AI has proportionally more things to maintain and fewer population to support it. On top of being designed to unit spam.
Economic attacks work on the AI, we tested that extensively, both the Chaos and Death spells. I don't see where this is coming from. Using the Death spells can reduce their income to below zero in a given category (Evil Presence+Warp Node for power, Famine+Evil Presence for food, and Pestilence for gold) causing them to lose all units that require that type of maintenance. Chaos spells are a bit more subtle - you don't see much visually out of the AI having to play on near zero taxes instead of maximal, nor the wasted resources of having to rebuild lost buildings and troops. All you notice is the AI sending less things to attack you but it's easy to think "hey, meteor is killing them all" instead of realizing they are also producing less. However you can lock a city into a state where it's unable to produce anything if you also cast some Call the Voids on it to destroy the religious and/or food buildings and shrink the population. At a low population the +7 rebels from the two spells are enough to make sure every remaining citizen becomes a rebel, even if there are units in the garrison and taxes are zero. Then the city will shrink to 1 pop and stay there.
We don't have time for this prior to the 5.0 release but feel free to change maintenance advantage to 100 (full cost) and see how that affects the AI. I bet you'll start to see AI with 0 gold and 0 mana on high difficulties.
Eh, on high difficulty and Archmage/Conjurer/Spellweaver AI has 0 mana even with the advantage - they summon enough creatures that they cost of summoning and the maintenance together exceeds their income, yet they keep summoning more.
This was most noticeable when Archmage was +10 starting skill, nowadays it's probably less likely, especially as the AI has more starting gold and mana than it used to.
![[Image: Ba8gOAs.png]](https://i.imgur.com/Ba8gOAs.png)