Revision of the spell selection system

[zitro1987]

Getting +6 spells instead of +4 (but without cost/research bonuses) incentivize books, even to add a realm.

Also, more spells is more exciting than cost reductions.

However, I may be bad with the balance, maybe +5 spells (instead of +4) and leaving just the research bonus might work better, leaving 9 books as the max.

[Catwalk]

Just to be clear, I don't think zitro's proposal achieves the desired goals. So make sure you keep his ideas separate from mine

[Seravy]

I just ran 3 test games, books for AI wizards were distributed as follows:

This is for Impossible where the AI has 2 extra picks.
On hard, I got a 4+4 wizard on the first try.

I can assure you that you will be able to pick a number of books that allows uncommon spells to appear early on. In fact, this system could well lead to them appearing sooner. Would that satisfy you?

Probably.

No, you were saying that going with a high book setup was worth it for the guaranteed spells.

They are, but retorts are even more worth it, which is the problem.

It doesn't need transportation right away (unless you insist on putting them on tiny islands). It also doesn't need a combat spell right away.

I say it needs both. Especially with the free sawmill which allows the player to attack AI settlements even earlier and easier than before. Having a free swordsmen is pointless if you can't cast a spell in the battle.
Island starts are a possibility, but even if there is no island start, being able to go to that good spot with adamant on the adjacent landmass instead of the crappy desert on their own is highly relevant.
I'm glad you at least accept they need the summoning spell.

Guaranteed absolutely is good enough.

Erhm definitely not. That leaves the AI with zero starting spells. Meaning no summons. Meaning no defenses in the capital if the human player does research War Bears or Sprites. (no, the extra swordsmen won't be enough and impossible isn't the only difficulty level anyway) Also meaning no ability to retaliate if the human attacks them, and no ability to take easy neutrals or nodes early. This would put the AI back to where it was a a year ago.

I just had a look at the difficulty settings, and an Impossible AI wizard with 10 books gets 37 power from turn 1. He also gets a 60% discount on casting cost and 22 gold income. He can cast 1 Sprite per turn if he wants to!

Except, you are suggesting they don't get a starting spell right now, so they can't cast it at all. I would need to redesign the power distribution system for the early game and add a ton of special cases for early research picks to make it work, and even then the AI misses the chance of summoning many creatures while they are researching it.

Anyway, of course I'm interested. I believe there is less than 1% chance it'll work out but I have been wrong before. However, keep in mind that we are talking about crazy amounts of work here so that new system has to be not just barely acceptable but massively superior to the current in every possible aspect. If I see even a single detail where things might go wrong, I'm not going to risk it - which is why I'm trying so hard to explain those details right now.

[Catwalk]

Apologies if I was confusing above, my comment about guaranteed absolutely being good enough was in reference to your suggestion that they need 3 starting spells. 1 starting spell and 2 guaranteed spells is sufficient.

As for early AI power distribution and research, can you ultra-briefly describe how it makes decisions about those two things? I think I'll start a game with 1 AI and see what it does early on, to get an idea.

[Seravy]

Apologies if I was confusing above, my comment about guaranteed absolutely being good enough was in reference to your suggestion that they need 3 starting spells. 1 starting spell and 2 guaranteed spells is sufficient.

As for early AI power distribution and research, can you ultra-briefly describe how it makes decisions about those two things? I think I'll start a game with 1 AI and see what it does early on, to get an idea.

Research - it picks the cheapest spell, plus there are a lot of multipliers to the cost based on spell type - for example global enchantments and certain combat spells get more priority. There are no spell specific considerations here unless we add them one at a time manually - I did that for Enlightenment, Gargoyles, Werewolves and Giant Spiders, I think that's everything but there might be a few more.
Ultimately, the system works fairly well but isn't particularly smart and cannot adapt research choices to the game whatsoever except one thing - they add more priority to spell groups they have none of learned - not sure if that applies to all groups or only specific ones though. Also the grouping is really messy and often makes zero sense.
One more detail, the AI adds the "trade value" to the priority - so spells they like to have more will be researched a bit more than those they don't. (and spells the AI is unable to cast will never get picked for research unless nothing else is left)



Power distribution - this is a procedure I have rewritten completely, so I can even share the source code and it's somewhat easier to modify.
The current version does one of these :

If gargoyles/wolves/spiders are being researched before turn 60 - 80% research, 20% mana
Before turn 30, below 250 mana, below 25 skill, above 100 mana - 25% skill 75% mana
Before turn 30, below 100 mana or above 25 skill  - 100% mana

The goal is to fuel the early summoning spells to have forces defending the capital, and expand. For that mana and skill are necessary.
It's important to mention the link between this system and alchemy - if the AI is below 1/8 of their gold in mana, they'll convert to mana, which we don't want to trigger in the early game - it's a waste of resources. Considering the high starting gold amounts we have now, as long as the AI has ANY overland spell available, mana production has to be kept at or close to 100% to keep up with the casting skill. Even with that, Archmage wizards tend to run dry on mana anyway, as the income can't keep up with a casting skill of 30. (on all except impossible, the AI probably won't have a power base of 30 on turn 1)

1 starting spell and 2 guaranteed spells is sufficient.

I don't think that's going to be doable, unless the AI cheats - as I said the interface is meant for one type of spell selection. I can most likely make it set "researchable" or "known" whichever we want, but I can't make a copy of the entire interface and have both - and it would also require disabling whichever the player picked in the former form the choices. So that leaves the player with no starting spells while the AI gets 1. Considering how much influence the summoning spell has in the early game, that would be way too unfair I think.
(this is all guessing, though)

[Nelphine]

The other problem with this entire concept is sprites. You simply don't need web or call centaur or earth lore to make sprites work - they make things a little faster, but if you take all the applicable retorts, you're spamming out one sprite per turn, from turn 1 (or 1 every two turns I guess). You can afford to risk looking at earth elemental nodes even if it turns out a giant spider is there and kills your sprite, because it won't happen often enough to slow you down enough to make a difference.

It absolutely is retorts that make sprites work, not sprites by themselves - which means trying to fi x the problem with out touching retorts is going to be a huge amount if work that might not succeed anyway.

[Catwalk]

I disagree completely Nelphine. Retorts can't make anything other than Sprites "work". Fixing all retorts in order to fix one broken creature is a bad priority.

Also, Sprites don't work nearly as well without Web (to a small extent) and Earth Lore (to a large extent). Your snowball is delayed considerably without the map info from Earth Lore, it's much harder to locate the neutral cities before the AI takes them and it's much harder to locate that crucial first node which can boost your power base.

Sprites themselves should be fixed first, and then we should see which retorts function as intended and which ones are overpowered.

[Nelphine]

Have you done your test with 10 nature books? I don't believe you can make sprites snowball that fast when you can't cas t that many of them in the first two years.

And yes I agree that retorts aren't as big s problem with other spells, but just looking at what sersvy did with 9 war bears (and what I've seen the AI do with early summons) indicates rhat while sprites are the best, they aren't the only summon that cab make it work. Ghouls are another one that should do well with those retorts.

Similarly, do the test with 2 nature books and all your retorts. You should still be able to push sprites almost as well even without web and earth lore.

[Seravy]

Conjurer is clearly broken as is when stacked with specialist - It doesn't matter if you summon sprites, bears, or spiders ghouls and gargoyles at 50% the normal price. And if you can last that long which is not too hard with half price summoned armies, half price sky drakes or colossus are scary.
...and then there is the half price summoning in combat as a bonus feature.

Spellweaver costs 2 picks, adds 50% to your casting capacity in overland only, and does not entirely fuel this skill with mana - and it's a good retort.
Conjurer+Specialist doubles your casting skill on summons only - the most relevant spell type as the majority of casting skill is usually spent on them in overland and at this price even in combat - and fuels the entire 100% boost with mana - reduced costs are something you don't need to pay for. On top of it, makes research faster, prevents dispelling and reduces upkeep. And it also costs 2 picks.

Similarly if we reduce starting commons, Sage Master will need change, as it will be the "new Mana Focusing", those free 9RP will suddenly be massively relevant to the game.

Then there is picking warlord+tactician+alchemy for unstoppable early military units - not as potent as conjurer+specialist, but the addition of the starting sawmill might even make it so.
Archmage obviously boosts early game as well, not much more than spellweaver, but for half as many picks.

the other retorts are most likely "safe" as they don't provide an early military or economic bonus - oh wait, Inquisitor does nothing but that! The rest of them are fine though, so we "only" need to test these 9.

[Nelphine]

Having tried the alchemy warlord tactician draconian Bowman, (before free sawmill), I couldn't defeat the ai with it. It took too long to get going, and even witha miners guild I couldn't push one out Every turn.