When Elves Ride Dragons

[Seravy]

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

But I don't want to reduce the casting advantage - early game aside, the AI needs this amount for mid and late game and we can't make it variable. I thought I made that clear.

I know you want to reduce casting advantage but that was already rejected several times. I'm trying to find a different solution to a different problem here.

...ok, let's start over and define the problems first.

Problem 1 : Both the AI and the human player are forced into producing military as soon as possible to avoid getting murdered by others who did produce them.

Problem 2 : Slow strategies don't work - producing military early always yields better results. If not for problem 1 then because it can get you treasure, nodes, neutrals, in other words more value than not producing them.

Both result in :
Reduced choices on both human and AI side, which makes the early game repetitive - you almost always produce 1-3 settlers, then build up to the best realistically reachable military unit and produce those, while you max mana, and summon as much as you can for the first 50 turns of the game.

Pushing research (outside of getting an uncommon summon, Aura of Majesty, or some sort of a combat spell immediately needed in battles) is not a good tactic. Pushing skill is not good either. Building anything other than the settler, early military, and cheap economy buildings (marketplace, granary and co) are not viable. Holding on to gold to hire heroes or mercenaries is usually not viable. Building swordsmen and other low end units is usually not viable as building the tier above that is better - and the swordsmen tier gets mercilessly annihilated by the common summons.

There are exceptions - if you pick Charismatic and rush Aura of Majesty you can get around problem 1. If you build horsebowmen, longbowmen, draconian bowmen you can get around problem 2 and still scout and exploit the map without summoning troops. But on the whole you usually can't get around it.

(by the way this actually makes the game AI friendly - if it was less defined what's the good strategy for the early game, the AI would perform much worse at it - do max mana and summon a lot is boring but at least easy to implement. So it's somewhat questionable if fixing this problem is even a good idea. If there is no "one good strategy" for the early game, that just makes overall AI performance a lot more random and difficulties even more all over the place. We've already concluded early game positions affect the mid and late game the most - having 6 cities on turn 100 will give you an entirely different position than 2 or 12 cities.)

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.

I don't think so. The capital is busy producing settlers and even if it's not, it has to build a fighter's guild and has a high chance to "waste time" on marketplace, granary and other things on top of it. The outposts won't turn productive that early - the settler needs to get produced, reach the tile, build, it needs to turn into a city, then it needs sawmill, 4 swordsmen, a fighter's guild, and then it has a chance to produce units - but might pick a building anyway, it's random. Maybe it's turn 30 on lunatic, but not anywhere else. Large numbers of city troops start to show up much later, probably turn 50-75 on lower difficulties.

[Arnuz]

(by the way this actually makes the game AI friendly - if it was less defined what's the good strategy for the early game, the AI would perform much worse at it - do max mana and summon a lot is boring but at least easy to implement. So it's somewhat questionable if fixing this problem is even a good idea. If there is no "one good strategy" for the early game, that just makes overall AI performance a lot more random and difficulties even more all over the place. We've already concluded early game positions affect the mid and late game the most - having 6 cities on turn 100 will give you an entirely different position than 2 or 12 cities.)

So given this, do you really want to change the game's initial conditions? I think that it's a bad idea too, but for different reasons: first, it might - no, WILL be tough to test. Second, it detracts from the wargame experience: having to choose between economy and military is a staple of it. I actually tend to build economy a lot, relatively speaking.

So, given that, back to step 1: how do we fix the excessive randomness based on geographical boundaries that are the only thing that can block the excessive stacks? I wouldn't change the game drastically, I believe that we can fix the issue.

I am not proposing, and I don't think Nelphine is either, to remove the casting advantage. Let's just reshape it. His idea of only casting summons at the beginning is actually pretty good if you think about it - that's exactly what I do as well. Why shouldn't the AI do the same? You have said yoruself that the buffs are not contributing much at that stage of the game.

And, if it can only summon, who cares if it reaches a mana chokehold? It's the chokehold that we designed and tested to achieve the proper balance at the different difficulty levels. It only means that it has reached a softer limit than the hard-coded number of summons idea. It is NOT a problem because as soon as some troops die and maintenance lowers, or as soon as it finds some mana cache in a lair, it can keep going. And as soon as some city develops a mana building, same...

This also makes the various mana retorts more significant for AIs. I start to think that we might be onto a good solution!

[Nelphine]

But I don't want to reduce the casting advantage - early game aside, the AI needs this amount for mid and late game and we can't make it variable. I thought I made that clear.

You've never shown math for why the AI still needs such a huge casting advantage at any stage of the game. I've tried to suggest reasons and explanations for why I don't think the math supports needing the levels the AI currently has (doomstacks, reduced spell costs and now the noticeable unit advantage the AI has at the start), and you have always rejected them - but not with calculated reasons, rather with your experience and opinions. Those are just as valid as mine (heck, even more so, you're doing the coding), but they aren't solid reasons.

So yes, I'll continue to bring it up, every time a new conversation (such as this one) again seems to support my idea - and especially if its a straightforward enough situation I can provide clear math asocuated with it.

And yes, you're free to continue saying no. All I ask is you keep listening, and making informed decisions. I may disagree, but that's all I ask.

Also, I'm enjoying my research focused death game (I'm ignoring ghouls) on master. But that's rushed uncommons as you've mentioned.

[Seravy]

So given this, do you really want to change the game's initial conditions?

I don't know, I'm now leaning towards not changing it. I suppose it's fine if the first 30 turns offer less choices than the turns later in the game.

His idea of only casting summons at the beginning is actually pretty good if you think about it - that's exactly what I do as well.

I thought about it but the AI still has to be able to cast certain other spells (Water Walking!) so if we want to do this we have to go through all the common spells, like how Focus Magic is already disabled before turn 10. We can't just say "all noncreature spells". (and I wouldn't be surprised if we found that 90% of the things that can be disabled already is like Focus Magic, but we have to actually check the code to know, I can't remember all 5000 lines of it...)

However wouldn't this have the opposite of the desired result? Yes, there will be a more predictable output of creatures for the first 15 turns - but it's not just more predictable but also a higher amount - the AI is casting creatures instead of other spells.

So if the goal is only to make sure the number of creatures is less random then it'll work but the price is, the AI will have at least 50% more creatures than before. Considering this discussion started by "AI has too many creatures", I don't think that's healthy for the game at all.

...maybe the earliest turn the AI can attack the player should be pushed further back? 25 turns are definitely not enough to get a few settlements going and building decent troops for all strategies - only the fast ones. If the AI is not using those stacks of Nagas to actively attack the player, it's not a problem?

But where do we want to draw the line?
How much should the player be allowed to build up before getting attacked? If they can reach, say, Paladins, that's definitely too much, right? Isn't it a problem if we give the player a free shot at attacking an AI fortress (or even two!) before the AI has a chance to try the same against the player? Wouldn't this make early strategies even stronger instead of helping slower ones?

(Note that this actually had a bug prior to 5.0 - earliest turn was meant to be 25 but was set at 37 for hostility. So fixing it probably made it even worse. (War turn was correct, the intended 30))

[Nelphine]

The problem isn't too many units on turn 15, its too many units on turn 40. Ban non summons until turn 10 (yes even water walking; the AI can afford to wait 10 turns for anything like that, just like the human almost always has to, if for different reasons). It will get 50% more creatures for 10 turns. But reduce overland casting by ~10% for every difficulty (this will net the first 10 turns only getting 35% more creatures). Assuming all 40 turns gets it the same amount of creatures base, then by turn 40 it will have the same number of units as it does now; but it won't ever be the same - the more turns in, the more units can be summoned. The result will be noticeably less units by turn 40. You'll end up with barracks rush still not working, but the vast difference between AI and human (even on advanced) will be reduced.

[Seravy]

You've never shown math for why the AI still needs such a huge casting advantage at any stage of the game.

Okay, maybe I should be even more clear about my reasons then - I thought I was, but I guess not.

I don't need math when the fundamentals are on my side and we are talking about something we can't calculate precisely - any calculations on something that has over 100% variance based on random variables is meaningless when we try to determine values with 5% precision - if the debate was about needing 350 or 620% advantage, math would have a use). The AI still picks spells randomly based on personality instead of following a coherent strategy, and still wastes its casting skill on putting creatures where they aren't relevant - there was no change in that and it is the basis for needing the casting advantage.

It certainly became better in one specific category - doomstacks. But half the AI in the game don't profit from that at all - they summon ground troops, or their "doom" stack consists of creatures the human player can easily counter.
Just because the unit isn't sitting in a city and instead is marching into 3 consecutive flame strikes in a stack of 9 doesn't mean the AI is better at playing the game - it is merely doing something different that sometimes works better, other times does not.

We are still thinking about whether it's better to just let the AI keep some doomstack material in cities or it's better if they are always used on offense - honestly, I don't know. As a player I'm definitely doing both, as needed, but the AI can't comprehend the "as needed" part. But even if we conclude always using the units in the doomstack is better, that is in no way grounds for saying "The AI is 34% better at using overland spells overall thus needs 10% less bonus on it". (what raising from 50 to 60% does make is actually a 34% drop in the amount of spells cast. Instead of 200% only 166%. That's huge!)

(actually there is one more area the AI is better at - since EXP9 it can cast a few spells as "on demand" instead of at random. But this "on demand" casting is on TOP of the random casting which is still there for most of those spells, which increases the demand for casting skill instead of reducing it. More importantly these spells are a minor fraction of the AI's skill budget - they are generally 20-100 cost common and uncommon utility spells. They don't influence the AI's spending on creatures or other tactics significantly if at all - 98% of the AI's casting skill is still spent randomly.)

If there is one change that might have any major influence on this, that's raising the needed amount of army power for an attack to 100% which MIGHT make AI attacks less wasteful, but this is a subjective thing - on one side attacks do result in less losses and more damage to the enemy when an attack is made, but it also results in more stacks that won't be attacking anything because they don't meet the minimal amount needed. While it "feels" like an improvement, I can't say for sure and I definitely can't quantify it and say "yes the AI needs 20% fewer creatures to achieve the same amount of damage to its targets through overland attacks". The only thing certain is it needs 33% more forces to launch the same number of attacks against the same targets. Whether it results in more victories and less dead units is something we can only hope for. Also, this still only affects the "summoned creatures that aren't hard countered by the enemy strategy" use of skill - anything else is unaffected. Yes, summons are the largest portion of skill use but not the only one.

Oh and high end summons are cheaper now. Which means the human has better access to them. The AI needs several stacks of X creature to be able to keep up with a human that has same X creature due to less efficiency at using them (logistics, losses in battle, overland healing the stack, etc) so if anything, this change means the AI needs more advantage. Instead of "AI needs enough bonus to keep up with my doom bats and efreets" it's now "AI needs enough to keep up with my 9 great drake doomstack(s)" as I can now afford using the latter. If I have 9 great drakes and can summon one more each turn, the AI probably needs to summon at least 3 to keep up with it, otherwise I can overwhelm it simply because I use my Great Drakes more effectively. (this is actually a good argument for making the advantage grow based on turns as rarer creatures have a greater "human is better at using this" factor, but unfortunately I can't make the casting skill advantage variable.)

In fact, the mere ability of the human to preserve their creatures in combat while ensuring the AI is taking constant losses is enough to warrant the 2x or higher AI modifier. If anything, it's too low if I only look at creatures, but probably about right if I also consider the AI is better at using stuff like certain curses or global enchantments (or not, change to recast and disjunction priorities to balance this out probably took this AI advantage away so it's only curses nothing else, not to mention AI pays 40% extra for globals now...)).

Fortunately, the power and production advantage ensures the AI's advantage actually grows over time - as it can afford raising skill and building Amplifying Towers more than the player - so the AI's inability to use very rares as well as humans has been accounted for, kinda.

[Seravy]

As a final thought, I consider this an extremely "AI unfriendly" game. It's full of "rock paper scissors" elements where the player knows what beats what and the AI does not. It's like playing RPS where you always get to pick your move after seeing what the enemy picked. On top of that, the complexity is too high to solve by hardcoded algorithms and real thinking intelligence is obviously not possible in a DOS game. I consider it a miracle that the AI can actually function and be a threat when it only gets double the resources as the human and I suspect this is already very close to the natural limits of what can be done. We might be able to shave off 5-10%s on the higher difficulty levels (especially lunatic where it still gets almost 2.5 times the normal), but that's about it. (and note that 10% in the casting advantage category actually equals a ~34% difference in casting capacity, it's not a small change. )

[Nelphine]

I'm aware that a 10% change is huge. And I agree that we can't possibly expect math to tell us exactly what the percentage should be.

However, we can say that the original had even larger cheating bonuses - and when major improvements to the AI were in the game, those cheating amounts were reduced.

We can say the AI has doomstacks (which is a huge ai improvement) and in the early game (first 100 turns) if you aren't playing certain strategies or get great geographical luck, those doomstacks (plural) can be overwhelmong. And if you DO get that luck or strategy, then you can beat those doomstacks and it doesnt make a difference that the AI has more if them.

We can say that we have worked put the math for what the AI needs to survive a barracks rush. We can say that we've worked out the math for how much casting skill the AI needs to have access to enough very rares to be a threat. Both of those sets of math say the AI should have reduced overland casting skill - and about the same amount of reduction.

We can say that there are certainly bigger problems than overland casting advantage. But overland cadying advantage doesn't stop the lunatic strategies (,you could make lunatic have a 20% modifier and I'd still play my bexerkers, except against mass curses, and if I get consecration early enough, even that wouldn't be a problem)- and it makes a lot of other strarehies brutally hard to play. Reducing the casting advantage won't solve everything - and I hope I never gave the impression that it would. But it will make the game more playable, and will allow for a wider variety of strategies to be used. Which will make it easier to see exactly what else should be improved for the AI.

ANY change at this point is a big deal. But we agree there are problems. This is a straightforward solution, that's also extremely easy to revert if after 100 games its obviously not working. And if it blows up in our face because its a terrible change, it will be easy to see - no doomstacks, undefended cities, maniacal wizards not cursing enough to even be annoying.

[Nelphine]

As to the rock paper scissors comment: reducing casting advantage makes that EASUER to deal with. Right now, the AI has such huge advantages, the human MUST use strategies the AI can't counter. Which means modifying those strategies makes the game unplayable. If the doesn't have the huge numbers of advantage, then you can make the rock paper scissors less noticeable. You can tweak summons and city troops to be more comparable allowing more strategies to work, and the AI won't have to be prepared for the huge variance in strategies.

Basically - numbers force rock paper scissors, rock paper scosdots forces numbers. You HAVE to change one in order to get started on solving both. Numbers us easier to change to start.

[Seravy]

then you can make the rock paper scissors less noticeable.

But that's one of the core part of the game which makes it good and distinct. Most (4X) games don't have powerful abilities such as complete immunity to an attack type which makes them boring and makes this one stand out as the superior one that actually rewards strategy instead of micromanagement and number crunching. Besides, you can't tweak immunity to be "less immune" to something even if you wanted to. (I guess you can give archers and magicians a stronger melee attack but that's kinda...weird.)

You can tweak summons and city troops to be more comparable allowing more strategies to work, and the AI won't have to be prepared for the huge variance in strategies.

One problem here might be that city troops you can get early such as swordsmen are not a huge investment, they are cheap. So they are weak. Summoned creatures on the other hand are huge investments, thus stronger.
That's balanced. Why is it not working out anyway?
Because the starting conditions are uneven.
Your city has ~10 production, but you have ~20 power and casting skill. Despite the inherent assumption that skill is a less available resource than production. So creatures are overweight 4 times compared to how it should be in early troop availability.(2 production = 1 casting skill)
But it's worse - your production is competing with settlers which are essential in most cases, and buildings (which are also essential to unlock troops or to have better economy). Meanwhile your power is not competing with anything - your early spells will be 90% summons (or buffs if you play Life but that's the same deal, it's military spending). There simply aren't economy or similar choices at common spells - at least not enough to invest the majority of your skill into them (and this is actually true for a large part of the game, probably only at the "rare" tier can economic spells outweight creatures in all but the Chaos realm.)

So yeah, we might be onto something here. The problem is we raised starting power (and the original game raised starting skill in one of their patches - it wasn't always this high, 1.0 you had a skill of, like, 5 to start with!) but starting production, even with the sawmill, lags behind while the demand for production is higher than the demand for skill. So production can't compete with skill until like turn 30-40 after which it outweights skill as intended and it stays that way until the end of the game.

But if production gets more available at start, early races just steamroll the map...that merely trades one problem for another. We would need to completely rework the cost of settlers, military buildings and (racial) units.