Caster of Magic II Game balance brainstorming.

[Seravy]

Instead of looking for a solution involving how hard it is to take a city, maybe think of changing the value of a captured city. If you feel that captured cities are too valuable at present, maybe think of some ideas to make them less valuable, such as a longer period of turmoil that prevents them from being a quick boost to your empire or being more costly to defend (defense penalty, higher chance of raiders attacking).

Well, I have already done that in CoM I in two ways : building destruction rate in general, and Smithy destruction specifically.
I don't think any further action is necessary, although maybe requiring neutrals to have at least 4 swordsmen (instead of starting at 1-2 then growing to 4 over time and pop 1 only having spearmen) might be an improvement.
Cities larger than those already require significant effort to take out. (the pop 3 Dwarf city I found had 4 steam cannons in it. Most races get some ranged unit at pop 3 or 4...)

In COM, raiders seem less crippling to the AIs, but an annoyance for me. Part of the problem with raiders is that they tend to be a rare catastrophe, so investing in garrisons mostly feels like a waste.

Sounds like it's working as intended? (except that it shouldn't be a "rare" one but something that happens often if all cities are undefended - but that's higher difficulty, on normal or fair, "rare" might be appropriate. Also, that's the role for rampaging monsters - raiders do come from neutral cities but there isn't always a neutral city near the player's empire or it might not have dangerous units available.)

[jhsidi]

I wanted to play a few more games at higher levels before commenting on these two.

3. Is it too hard or easy for the AI to take out lairs under the current automatic combat rules? (Phase II)
Ideally, we want a reasonable chance for the human player to get to at least their fair share of lairs on the map, but the AIs should also get their fair share, basically in a 5 player game, the AI should get about 3/5 to 4/5 of the lairs on the map before the human player unless the human player plays too low difficulty or specializes on treasure hunting strategies. (related to the previous issue)

Too hard for them / infrequent currently. In my Expert level games with normal world size, the AI did eventually get underway. I probably got about 50% of all treasure on both planes. It seems to be the initial phase they had trouble with. They get moving with effective doomstacks eventually but before then, they won't even risk stuff like Sorcery nodes with a few nagas. In my more recent Master level games I'm really not detecting a lot of difference, seems like about the same pace.

Part of that may be because the algorithm seems tuned for the AI to be extremely conservative and cautious with its stacks. I literally can't think of an instance where I saw a stack go into a lair or node and lose even a single unit. As a player, that looks awful. The AI slings around stacks of 9 nagas or ghouls for years and does absolutely nothing with them, then at some point suddenly has a seemingly invulnerable new doomstack that mows down everything in its path. If they're not at war they should be taking some risks; they have the resources for it.

There's probably another problem. If all these ruins etc are on the AI's landmass, they do OK. But they don't really go crusading effectively. Not for ruins, anyway. I do always have annoying doomstacks popping up on my island, but they apparently just come to annoy me.

4. AI players should eliminate each other at a reasonable pace. (Phase II)

Basically, if they do not then they'll left behind compared to the human player who grows by conquering other player's territories. In addition, AIs not expanding and destroying each other means balance issues with many late game spells which assume the number of players is generally lower than 4-5 by the time the spells are available.

Not happening even remotely soon enough. Last long game, I saw AIs starting to conquer each other around... 1512? Too late, I have an insurmountable lead by then. I don't know what your target is for first banishments, but probably around 1507?

The AIs seem really friendly with each other. Wars are not frequent. Currently it's 1508 and there are two AI to AI wars taking place. The wars also don't seem to last a very long time. Pacts and alliances, on the other hand, are very frequent. Some AIs seem to pact or ally everyone else and just stay that way the entire game. On Arcanus, there are 6 wizards with at least 5 treaties each, and that includes some maniacal and aggressive wizards.  

I can actually see some potential there; what if an entire network of pact holders could collectively declare war on you at once after some point in time? That would probably provide a comparable challenge to the single big empire you said you want. Of course it happens over time now -- getting in wars with a wizard makes their ally hate you. But it takes quite a while before they'll get around to also declaring and by then you can declare peace with the first wizard. Anyway, they could also knock out each other in that way -- one network of 3-4 wizards picking apart a single other wizard.

[Seravy]

One possible idea to fix the "AI's not declaring war" issue is to create a new war declaration type for this role, that functions slightly similar to militarist but not personality based.

When the target's historian graph strength is below 40% of the AIs, trigger an "Insignificance" based war declaration. Allies are never targeted with this. The human player is never targeted with this if they have at least a Wizard's Pact with the AI.

Would that work?

[Seravy]

AI lair hunting :
One thing seems obvious, the first 2-3 years they can't take any because all the units they build end up pulled into defending outposts.
This makes the 3 easy starting lairs wasted on the AI players as they aren't clearing them early and even allows the human player to steal them if they are close enough.

Looking at Advanced difficulty (1 AI).
In 1403 June they 5 cities, all properly garrisoned, but haven't built any stack or attacked any lair.
In 1504 February they finally managed to take out the lair that had 1 guardian spirit using a war bear and a hero.
Meanwhile one of their outposts have 4 war bears stuck defending, of which 2 have wraithform.
One turn later they finally started assembling a doomstack. 2 Sprites, 2 heroes, a bear with wraithform and a lizard halberdier. However the two bears with wraithform in the city weren't pulled - the city isn't yet full of units to allow sending any out.
Before the stack could complete, the 3 units already together went and conquered a barbarian neutral (5 swordsmen).
Rampaging monsters spawned (a phantom beats, didn't check what else) and attacked the city with 4 bears, city was successfully defended.

One of the AI cities is trying to pull a unit of Sprites to garrison but fails because the shortest path has ocean tiles. The sprites turn back on the ocean tiles because they are being sent to the main action continent (which is the same continent). The only way to fix this would be to swap the order of actions - pull defenders before sending units to the main action continent and building new stacks. That might be worth trying?

Wait, no that breaks stackbuilding. It can't be done. Maybe the AI should force these stacks to move without leaving the continent?

Or maybe we could move thee "go to main action continent" and "board ships" functions further down after stackbuilding and pulling defenders?
At first sight it seems smarter to leave the continent after the stacks are large and cities are well garrisoned, but that might delay boarding ships significantly for no reason? Assembling a land stack on a semi-large continent can take an extra 4-5 turns, or even 10-15 if one of the units is move 1.
Eh no, there is another problem with this, if the stackbuilding finds a stack of 9 that's already assembled, it orders them to stay where they are as it assumes if the units had anything else to do like move to another continent, they already did that. Well I guess I can try removing that...

Let's try to see how that works but I'm going to need an AI that's not lizardmen for this one.

...ehh cities keep pulling units in like there is no tomorrow.
So yeah, the AI pulling defenders now works very well but as a consequence the AI literally has no troops to attack anything before 1404. Which is definitely bad because that makes them sitting ducks, if there is no threat of retaliation, the human player will not hesitate to attack. So from that viewpoint this is even worse than when they didn't defend their cities at all.
...maybe that's fine for Advanced and lower difficulty though. Successfully invading an AI before 1404 isn't trivial.

o_O
I managed to observe the AI take out a gnoll city full of halberdiers with 3 sprites by repeatedly attacking them and killing 2-3 units a time. Nice to see that feature works at least.
This however makes it even more important to somehow make the AI keep some of their early units outside and active instead of being used for garrisoning...

Looks like even on Expert difficulty the AI only starts to have free roaming units around 1503. Before that they are all used to fill garrisons.

So we should probably make a decision here. Which is more important for the AI, the garrisons, or having stacks that can attack things (including players who dared to attack their cities.).
If garrisons are more important, we have nothing left to do. If having units is more important, we most likely should reduce the required garrison amount in smaller cities drastically. (for example, require 2 units in outpost and pop 1, 3 units at pop 2, 4 units at pop 4 and only one more unit for each pop beyond that. Or maybe make it turn based, use the current amounts later on but the first 50-60 turns, use lower values?)

[jhsidi]

So we should probably make a decision here. Which is more important for the AI, the garrisons, or having stacks that can attack things (including players who dared to attack their cities.).
If garrisons are more important, we have nothing left to do. If having units is more important, we most likely should reduce the required garrison amount in smaller cities drastically. (for example, require 2 units in outpost and pop 1, 3 units at pop 2, 4 units at pop 4 and only one more unit for each pop beyond that. Or maybe make it turn based, use the current amounts later on but the first 50-60 turns, use lower values?)

To me the garrisons are really important and I don't want to see them get weaker. As the player it's going to be way too easy to me to exploit weaknesses and it's not a given that the AI will be able to strike back effectively -- sometimes they can, sometimes they have no answer for the player's strategy.

Since this is just an early game issue, could it be partially solved by giving a larger resource bonus to AIs in the early game and then scaling it back as the game moves on? As far as I know the resource bonuses remain static throughout the game. Or perhaps by adding a new mechanic like a pre-existing escort of units for the starting settler units, so that they don't have to spend as much time garrisoning.

[Seravy]

Since this is just an early game issue, could it be partially solved by giving a larger resource bonus to AIs in the early game and then scaling it back as the game moves on?

Not really.

This is the current build order :

1. If (gold>=180), buy marketplace
2. If (gold>=180) and (difficulty>=Expert), buy magic market
3. If Pop<Housingrequirement, build Housing
4. Build Sawmill
(5. Build/Buy Smithy - outposts get one for free so they don't need to)
6. If settlers<2 or Garrison<4 skip ahead to 11
7. If no water move spell and on the other plane, but have settlers and 4 garrison, build ship yard
8. If gold<100 build Marketplace
9. If in contact with human player, build Monument
10. If (conditions) build mandatory military buildings
11. Build normally, but if garrisons size<4 then must build unit, otherwise if settler required, must build settler.

Okay, this has some issues. First of all the buying of markets should only be done if the AI is expecting to do long term housing. Otherwise they are not needed before units. So let's fix that.

Next, Sawmill. It has a "buy priority" of 4. That means the AI will only buy it if it has at least 4 times as much gold.
I think this should be in the same group as marketplace - buy at 1x gold if city is low pop.

Then there is 6. It isn't really necessary to build that in more than one city.
7. is fine because the marketplace will be bought as it was already set to "buy at 1x is low on population" although if the AI has no gold, it can cause a delay. Still worth it, as it fixes the AIs gold issues so it contributes to other cities getting their swordsmen faster.

So then we get this instead :
1. If Pop<Housingrequirement, then do the following three
1a. If (gold>=180), buy marketplace
1b. If (gold>=180) and (difficulty>=Expert), buy magic market
1c. Build housing
4. Buy Sawmill if city pop<7 otherwise build it normally
(5. Build/Buy Smithy - outposts get one for free so they don't need to)
6. If settlers<2 or Garrison<4 skip ahead to 11
7. If no ship yard in any city, no water move spell and on the other plane, build ship yard
8. If gold<100 build Marketplace
9. If in contact with human player, build Monument
10. If (conditions) build mandatory military buildings
11. Build normally, but if garrisons size<4 then must build unit, otherwise if settler required, must build settler.

So gold only helps with the sawmill which is now changed to require 1x gold which should be readily available on all but the lowest difficulty levels? The AI does get more starting gold as difficulty level increases already. Raising income is pointless as they don't have much to get income from yet.

Or perhaps by adding a new mechanic like a pre-existing escort of units for the starting settler units, so that they don't have to spend as much time garrisoning.

That too already exists - The AI gets +1,3,5 or 7 swordsmen units starting at Expert difficulty.

I'm not worried about the higher difficulty levels, but I was testing Advanced yesterday. I don't think low levels of difficulty should be cheating with free units and large sums of starting gold.

[Seravy]

Started doing tests to see how AIs eliminate each other. All tests will use the maximal 13 players.


Test 1, Expert, default land size, continent size.

1502
No treaties yet or wars yet. (It's before the first allowed turn for AI diplomacy)

Leading wizard is about 2x as powerful as the bottom (which are the majority) - this wizard is chaos draconian cult leader and already owns a chaos node. The advantage comes entirely from power income.

1503 February
First wizard's pact and war declaration happens.

...before reaching 1504 I already spotted a bug. AI wizards were able to make a wizard's pact after they already had an alliance, overwriting the alliance.

Restarting. (same map from turn 1)

1504
3 ongoing wars, 2 wizard's pacts, 1 alliance.
Looks normal so far, and the assumption that the AI's don't meet each other too early to get too many pacts going works.

1505
The gap between the strongest and weakest wizard is now about 150%. Myrran wizards have no advantage yet (they are yet to expand far enough to outclass Arcanus on territory size)
3 wars, 4 pacts, 2 alliances.

1506
The gap is now about 200%.
3 wars, 9 pacts, 2 alliances.
Everyone is still alive but this is early enough so it's fine. 1506 is roughly the transition phase between common and uncommon spells.

1507
Gap roughly unchanged.
4 wars, 12 pacts, 5 alliances.

It's starting to be clearly visible that the prohibition of long distance war declaration along with enough time for the AI to discover more players leads to a bias towards positive diplomacy.

1508
Gap still around 150-200%.
3 wars, lots of pacts and alliances.
Not looking good.

1510
Everyone still alive. Gap around 200%.
There is a clear leader on each plane, about 3 more wizards with relevant historian strength, and the rest are the "all other weak players".
~7 wars ongoing and an insane amount of pacts and alliances.
AI turn times start to take longer but it's still ridiculously fast for a 13 player game considering I'm using the newly added debug feature that plays 12 turns at a time. Each player only spends about 1-2.5 seconds on their turns.

1512
Everyone skill alive. One person weakened significantly from war, but still not banished. I guess the fortress with 7 magicians will be hard to take out without at least rare creatures and their other 3 cities have similar defenses.
Alliances far outweight wars and pacts by now.

1514
The gap between the strong and the leftover wizards is widening. The weakest wizard is down to 2 cities but still alive, now upgraded to Shadow Demons. Still at war with the strongest person on the plane but that defense is probably unbeatable until very rare creatures.

   
Overall, half the wizards are relevant with the other half being not relevant so that works as I was hoping for, except, the not relevant wizards didn't get eliminated. Most probably reason is the diplomacy unbalance, but fortresses being too hard to crack until at least a rare creature doomstack is available is another suspect.

The myrran wizards were pink, green and orange by the way.

Test 2 same parameters

1502
Nothing particularly interesting yet.

1504
1 war, 2 pacts.

1506
4 wars, 11 pacts, 6 alliances (yea that looks bad already)

...might as well not waste my time and skip to the conclusion.

So until 1511 no one was eliminated or even significantly weakened.
I think there is no reason to test any longer, the problem seems evident.

I'll consider possible solutions tomorrow including :
-changes to AI treaty forming
idea : divide the chance of forming the pact/alliance by the number of known (contacted) players.
-additional war declaration type
idea : declare war if enemy total historian graph is 40% or below own, army strength is 40% below own, and total number of players in game>desired amount. Peaceful personality, existing alliance or in case of lawful personality, existing wizard's pact prevents this.
-If all the above doesn't work, the probable culprit is cities being too hard to take under the current, more accurate automatic combat rules. Possibly increase relevant of combat spellcasting (researched spell tiers, casting skill amount) to allow the stronger wizard to win.

[zerkeron]

To offer a suggestion regarding the "weak AI Wizards in lategame" topic, I'll draw an Idea from somewhere else:

Total War: Warhammer 2 has a similar problem (as do many other games of the "Total War" series).
There are dozens of factions that are spread over a humongously large map.
In the early and midgame the player takes a few cities and defends against attacks which leads to the normal and cool gameplay I like from these titles.
But most of the time your empire grows faster than any of the AI's empires.
At some point, the lategame usually consists of the player just steamrolling over all of those medium and small AI factions while getting bigger and bigger.

A somewhat ok solution brought a mod for TW:WH2 that allowed weaker AI factions to surrender to stronger AI factions, hereby joining forces.
That means, that all cities and units now join the stronger AI faction.


They only do this:
A.) when their power ranking is too insignificant later in the game, and they can join an allied faction.
B.) when the player's power jumps ahead, to keep up with the player's relative power.


This way the player always has one or two enemies that keep up with his strength, instead of 10 or more small factions that maybe can't even fill a proper doomstack.

Maybe this helps in having no insignificant two city Wizards in the lategame.

[Seravy]

Related : I found a bug that made the AI check for war declaration against other AI only about 1/10 as frequently as intended.

[Seravy]

Implemented the downscaling of AI positive diplomacy chances by contact count and fixed the war declaration bug.

Test game 3.

1504
Most wizards only have contact with 1-2 others.
1 war, 1 pact.

1506
Most wizards have contact with 4-6 others. This shows the AI avoiding to move units too far away without reason isn't very efficient at reducing contact count and the diplomacy scaling based on contact count is probably necessary.
6 wars, 4 pacts, 3 alliances.
Now that's more like what I was hoping to see.

1508
All wizards pretty much know all other wizards already.
5 wars, 8 pacts, 5 alliances.
The alliances haven't had the chance to spread the war declarations around yet, or possibly the wars are already over and waiting for the expiration of the peace treaty to restart?

1510
9 ongoing wars, 13 pacts and 11 alliances.
No one has been eliminated yet.

1512
~11 wars, 20 aliances, only 2 pacts.
By now it mostly stabilized into who managed to made an alliance and who will fight wars whenever their peace expired.
The charismatic life/death wizard has the most allies as expected.

One wizard has been reduced to only having their fortress city and only 2 others are what I'd consider outstandingly weak.

1514
8 wars, 18 alliances, 2 pacts.
This does show the number of alliances is stabilized and won't grow further (in fact is getting lower because chaotic wizards are involved) and wars work fine, but no wizard was eliminated still despite all the wars.
The weakest and strongest wizards are at war but not even the strongest wizard can crack that fortress with 8 magicians in garrison.

Overall, diplomacy seems mostly fixed (although adding the new war declaration type might still be useful) but AI conquest isn't going as smoothly as it did in CoM I due to the more accurate automatic combat rules.

So next, test 4 :

For this one I added logging of city conquest and units died in combat. I'll check the results after the end of the test only to save time.

1514
Compared to the player count of 13, there are very very few units that died in battle and they were almost all common tier - nagas, sprites, ghouls, swordsmen, bowmen.
Cities were conquered 55 times which is not as bad as I thought but of course this also included cases where the same city got conquered back and forth. Overall, it is roughly 4 conquered cities per player which definitely isn't frequent enough to eliminate wizards.

I'll run the next test with the AI's allowed loss ratio raised to 75%. (currently it's 66%)

...to be continued