When Elves Ride Dragons

[Seravy]

We do some testing and assess what's the cheating needed for these numbers, and make that the base K of the formula. Then, any further summoning is based on number of cities or passed turns

Aside from the "casting advantage" the others have no direct relation to casting skill in the first 50 turns of the game. The AI isn't spending power on raising the casting skill during these turns, and it isn't aware of its resources when picking spells to cast. So gold, mana and power modifiers don't matter. It'll keep casting creatures even if it does not have the resources to do so and in this case it self-defeats - stuck on 0 mana with no combat spells and in worse cases also failing to fuel overland casting skill for further spells, or converting important gold into mana without alchemy, so it can't buy sawmills or fighter's guilds.
So I have to assume you are proposing the casting skill advantage to be variable because the others are unrelated.
That's the only one I already said is impossible to do though.

Wizard's Pact : yes it counts. However if that city was sharing that tile with another city which belongs to an enemy of the caster, it'll still get targeted. No they cannot attack your undefended cities so you didn't have a wizard's pact then. Peace Treaty also prevents attacks but it has a limited duration and the remaining turns are not visible to the player.

Attacking and not killing anything does not cause a relation loss unless you attacked a city (which causes a big one). However it does enable the AI to be hostile and attack you back or curse your things.

Elemental attacks are based on the attack type, not the unit. Older versions used to have spells that also cared about the unit but not anymore, now it's only the attack type.

the time between turns takes FOREVER.

Edit your dosbox settings to core=Dynamic and cycles=100000 (or whatever number your CPU can handle), I think this should be in the documentation.

Is it possible/would it help at all, if the AI units/spells had higher upkeep?

As explained previously this can't be done because the AI can't comprehend maintenance costs at all. We have to be sure the AI can always pay the cost and still have some left over even if it goes all out on producing things.
The AI would still have those big armies but would not have mana to cast combat spells or gold to buy anything that needs to be bought.
The AI is vulnerable to mana wasting tactics so it has to store a large amount of that. Gold is necessary to be stocked because you can't hire heroes, mercenaries or items if you don't.

decreased based on number of units NOT garrisoning cities/nodes?

That would have roughly identical results to raising maintenance which is bad.
The AI having lots of units mid/late game is intended anyway. We specifically only have a problem with the overwhelming amount of early summons which are needed to keep early military strategies in check.
Just because the player isn't always attacking the AI on turn 20, the AI still has to be prepared for it and have forces to retaliate. Which means they'll have those forces to attack the player even if it's not retaliation, which is bad for slower strategies.

[Seravy]

but it does count as -them- breaking the pact, which reduces relationship with them but no permanent damage to hidden relation

Unless I fixed that already, it counts as both people breaking the pact simultaneously, so you do get the hidden relation penalty.

So... if it has enough summons/enchants to be at -20 mana income, it will continue to summon more critters, which kills the ones it currently has?

Exactly. Well, there is a bit of a limitation, if income is far in negative, it might skip turns for casting spells but we don't want that to happen. If the AI already reached the point where it can't cast spells, it's as bad as if it was banished or worse.

[Tlavoc]

I probably just misunderstood or only remembered one half of the information.

I just want to say, Thank you everyone for being so helpful and patient! :D

Edit:
What settings do you guys run on Dosbox? I tried 100000 cycles and it's still a good while between turns when there are lots of units. Higher than that the music and movement of the mouse gets choppy.
Edit2:
Closing out some projects I was working on, setting the core to dynamic and max all made the turn wait go from about 1 minute to around 10 seconds. Much more tolerable.

[Arnuz]

We do some testing and assess what's the cheating needed for these numbers, and make that the base K of the formula. Then, any further summoning is based on number of cities or passed turns

Aside from the "casting advantage" the others have no direct relation to casting skill in the first 50 turns of the game. The AI isn't spending power on raising the casting skill during these turns, and it isn't aware of its resources when picking spells to cast. So gold, mana and power modifiers don't matter. It'll keep casting creatures even if it does not have the resources to do so and in this case it self-defeats - stuck on 0 mana with no combat spells and in worse cases also failing to fuel overland casting skill for further spells, or converting important gold into mana without alchemy, so it can't buy sawmills or fighter's guilds.
So I have to assume you are proposing the casting skill advantage to be variable because the others are unrelated.
That's the only one I already said is impossible to do though.

Gah. No, I meant mana/gold. But, I was assuming that you had coded in a limitation to stop summoning/casting when mana gets below a certain point - the thing it does in combat, I guess it's combat only then? So the fact that AIs always seem to have a mana buffer... Is that because their mana is ensured to be much higher than the casting skill lets them use?

Painful.

I guess a patch could be to hard code a limit of summons per year, but I have no idea how difficult that would be. On first glance it'd seem easier to change the casting skill advantage.

[Tlavoc]

Of course, that would also be a problem if several summoned units died. Every one summoned unit you kill puts them behind quite a bit if it's a hard coded limit

[Nelphine]

So... if it has enough summons/enchants to be at -20 mana income, it will continue to summon more critters, which kills the ones it currently has?

Exactly. Well, there is a bit of a limitation, if income is far in negative, it might skip turns for casting spells but we don't want that to happen. If the AI already reached the point where it can't cast spells, it's as bad as if it was banished or worse.

Not quite. This is exactly what economy destructuon spells should be able to do to the AI (for instance large amounts of evil presence, warp node, eternal night.) However, that's a slightly different topic so I'll leave it for now.

[Seravy]

So... if it has enough summons/enchants to be at -20 mana income, it will continue to summon more critters, which kills the ones it currently has?

Exactly. Well, there is a bit of a limitation, if income is far in negative, it might skip turns for casting spells but we don't want that to happen. If the AI already reached the point where it can't cast spells, it's as bad as if it was banished or worse.

Not quite. This is exactly what economy destructuon spells should be able to do to the AI (for instance large amounts of evil presence, warp node, eternal night.) However, that's a slightly different topic so I'll leave it for now.

They do it. We were talking about what the AI does on its own, external influence of any sort ignored.

[Nelphine]

Correct (I just don't believe they DO successfully do that - the cheating means their power income from other sources that can't be hurt, is so high that the economy destruction doesn't actually slow them down. However, that may be because I'm playing master/lunatic, thus, topic for another day.)

Back on topic.

Summons are the primary problem early game - mostly because some summons require certain types of units to defeat, and not all players can have that. However, the AI similarly has to assume that the players DO have those units, because otherwise it risks getting wiped out when the human does have them.

This therefore is a twofold problem - one) summons requiring other units to defeat (try to defeat naga as chaos in the early game), two) the AI has a lot of them.

Problem one is a balance issue directly with the units. I think summons in general need to be looked over (at least common/uncommon) as there is some wide variances, which are fine when looking at the unit by itself, but when compared to units of other realms/cities, can be a problem. This often leads to whackamole - you pick a strategy which defeats some AI tactics, but if an AI happens to have something you're weak against, then you die. (For instance, if you want to fight mono chaos, they won't have fliers or strong range at the start; this means almost any strategy can work; but if you plan for that and instead fight ghouls or sprites, you get brutally wrecked if you are trying to use unbuffed barbarians or gnolls.) For some strategies this is fine - we KNOW mono life gnolls are going to have huge problems, and that's ok. But when it affects a fair number of strategies (how many strategies can actually deal with focus magic cockatrices fairly early?) it becomes a bigger problem where you can play 5-10 games without encountering the AI with that strategy; and then get brutally smashed by the AI when you do meet them. (Which makes testing require dozens of games to ensure you've actually tried everything.)

Problem one I would suggest not addressing in this thread.

Problem two, the AMOUNT summoned by AI.
Lets math this out. The BASIC concept of early game AI activity is to ensure that it doesn't get 'barracks rushed'. Seravy, you've thrown out '15 turns' as your goal for this... at what difficulty? I would suggest normal does not need to be able to protect its outposts from a barracks rush in 15 turns, but it should have 9 units for its fortress - including starting swordsmen, and constructed city troops. I would guess offhand that the AI should be able to build 2 city troops in that time, so it needs to summon 5 creatures in 15 turns.

Now problem one comes in: Do we set this to summon 5 ghouls, or 5 hell hounds? Honestly, this is normal. So we're going to say 5 hell hounds. Other summons should be strong enough to defend the fortress without a full 9 units in 15 turns on normal. So on normal the AI should be able to cast 200 skill of summons in 15 turns. Assuming it will cast 2 other spells (average cost of 40) during this time, then it needs to be able to do 280 skill in 15 turns.

That's less than 20 overland casting skill per turn. Assuming a normal AI has an average of 10 books, that means that the normal AI literally needs NO casting advantage in order to meet this goal.

Now, on advanced, the AI is going to need slightly better troops than hell hounds (and if it only has hell hounds, then hurray - it has extra to go out of the fortress), but I don't think it needs yet to defend any other cities. It also starts with extra swordsmen. So, it needs to be able to summon 3 ghouls in 15 turns, plus 3 other spells (average cost of 40). That's 360 skill in 15 turns. This means 23 overland casting skill per turn. If an Advanced AI averages 9 books, then its going to need 75% overland casting costs in order to achieve this. (Currently Advanced has 66%)

On expert, the AI starts with swordsmen, but it doesn't want to have too many of those in its fortress against a barracks rush. So let's say it needs to summon 4 ghouls in 15 turns, plus 3 other spells (average cost of 40). That's 430 skill in 15 turns. Expert will also average 9 books, so it needs ~65% overland casting costs for this. (Currently has 50%)

Master, the AI should get 5 ghouls. That's 510 skill in 15 turns. I also guess that Master averages 10 books. So it needs ~60% overland casting costs for this. (Currently has 50%).

Lunatic the AI should get 6 ghouls and 4 other spells (average of 40). That's 630 skill in 15 turns. Average books here is probably 11. So it needs 50% overland casting costs for this. (Currently has 40%).

[Seravy]

try to defeat naga as chaos in the early game

That's actually fairly easy if you rush gargoyles, or if you have wall of fire. But against bears with sorcery, you lose. (well actually you can rush flight and build a wall so...idk, almost everything can be countered but if two wizards with different things attack you, it's over.)

Seravy, you've thrown out '15 turns' as your goal for this... at what difficulty?

Let's see....

on Normal I think it's fine if the (first) AI doesn't have enough troops to fully protect the capital on turn 15, but probably should somewhere around turn 30-45. On turn 15 if it has 1-2 summoned units that's good enough. Normal is meant to be easy to beat.

On Advanced, it's say it's still acceptable if the (first) AI is weak to the top early strategies but should not fall to "hey I have 9 klackon swordsmen with no buffs, can I have my win please?". It should have at least a full capital by turn 20-25 and a few units around it to retaliate if outposts get attacked, so I'd say, it needs like, 0.75 units per turn? So on turn 20 it has 15 units - 9 in capital, 6 to counterattack...

On expert and above the AI should have 9 units in their capital on turn 15, and probably at least 9 more on turn 30 to retaliate with a powerful stack to deter players from stealing pop 1 cities.

Assuming it will cast 2 other spells (average cost of 40) during this time,

Unfortunately this is far from the truth, it's safer to assume the AI will spend roughly as much on other spells as summoning. Magic Spirit, Floating Island, Water Walking/Wraith Form on settlers, buffs on units (No a Holy Weapon or Endurance is not worth as much as a bear or ghoul or even a hound), wall of fire on the city (which won't help if the player sends bowmen or ghouls) etc. there are plenty of common spells the AI can cast, and quite a few of those are not contributing to military or on a much lower scale than creatures.
(I feel you were trying to set the goals that match your preferred casting advantage settings - it's too much of a coincidence you got the result you previously already suggested, especially since you were trying to calculate something that has so much variance and random elements that it cannot be calculated at all.)

...and this has answered why we can't address this issue through changing AI advantages. How much the AI spends on noncreature spells is random. It might be 200% of the cost spent on creatures, or 0%, depending on what spells they have, what objective and what is rolled for the actual choices. If we are only talking about 20 spells total, we can't say "it evens itself out over time as many rolls are done", results far from average can and will happen. Assuming the average is 10, rolling 3 fewer compared to 3 more means double the amount of creatures (7 vs 13!) and these are fairly likely outcomes. Especially as the presence of spells affects the chances - if the AI has more noncreature options available, it will cast fewer creatures.

I feel this problem can only be solved through changing starting conditions, but don't know how that can help without making other things even worse. More starting units or larger cities just puts even more pressure on everyone to make sure they don't fall behind in their military, especially if those starting units are uneven - and they have to be, every race and realm gets different units.

Basically the game is like starting chess on an empty board (king only) with putting a piece on the board costing one move. Obviously, for the first moves, everyone will want to put pieces on the board instead of anything something else because once the other player has enough advantage in the number of pieces, they won the game. This is true for both AI and player which isn't good - it would be better if different choices would also be viable (like going economy or research first instead of military+summons+settlers).

[Nelphine]

So ban the AI from casting non summons (if it knows any summon other than magic spirit/floating island) for the first 15 turns. (Could pdobably jusy do 10 turns). Then you can reduce the casting advantage while not falling prey to barracks rush.

After that time, they'll be building city troops in numbers - they'll have 18+ units by turn 30 even on advanced with the numbers I've shown.