Guaranteed early rare

[Seravy]

One more thing I realized, the above numbers are wrong, there are 4 rare arcane spells in the game, not 3 so the total number of rares is 16.

AI early rare spell choices.
In general, the AI should only ever pick spells that are at the very beginning of the tier on their research priority list.
As there are 16 rares and only 8 book slots, the amount of unresearched commons and uncommons is not relevant because the AI can benefit from this feature even if all 8 of their slots are used by rares with no leftover lower tier spells blocking them. So I will only look at the rares on the AI's research order.

Nature
Spells appear in the order of Elemental Armor, Gorgons or Stone Giant, Survival Instinct, Petrify, Gaia's Blessing, Abundance, etc
Earthquake comes before Gorgons for Maniacal and Ruthless wizards.
Petrify is added to the beginning of the list if and only if the human player is a Death wizard.

One possible way to set up AI early choices :

If Maniacal, pick Earthquake.
If Peaceful, pick Gaia's Blessing and Abundance
If Narcissist or Channeller, pick Petrify.
Pick Stone Giant or Gorgons at random (50% chance each)
Pick Survival Instinct.

Sorcery

AI research order here is Storm Giant, Flying Fortress, Invisibility, Banish, Uranus' Blessing, Mind Storm, Spell Blast, etc.
Narcissist wizards put Uranus' Blessing to the front of the list. No other modifiers seem to apply here.

Considering to implement it this way :
Pick Storm Giant
If Narcissist : pick Uranus' Blessing
if Peaceful or Maniacal : pick Flying Fortress
Otherwise 1/3 chance for Invisibility, Banish or Uranus' Blessing each.

Chaos

AI research order is Fire Storm, Flame Strike, Doom Mastery, Doom Bat, Efreet, Chaos Spawn, Chaos Rift, Doom Bolt, Warp Lightning, etc
Inner Power is added to the front if enough relevant creatures are in play buffed by it. There are no other modifiers.

This actually is less straightforward than I thought because there are pretty much four spells the AI will want (Fire Storm, Flame Strike, Doom Mastery and a summon) but only two picks.

There are three creatures here and Chaos works just fine without a rare creature, Chimeras and combat spells can do the job well. So we can get away with not guaranteeing the early creature.

Fire Storm or Chaos Rift is essential for Maniacal and Ruthless wizards who cast curses a lot, not so much for other personalities. They are mostly equivalent spells as both deal damage to garrisons so we only need one.

So I think we could use this set of rules :

If maniacal or ruthless, pick Chaos Rift or Fire Storm at random (50% chance for each).
pick Flame Strike
pick Doom Mastery

Life

Life research order goes like this : Incarnation, Angel, Divine Order, Inspirations, Lionheart, Prosperity, Invulnerability, Holy Arms,...
Divine Order is pushed to the end if at least 2 Chaos or 2 Death books are own. I think I'll change this to 3 books as it feels a bit overly cautious?
Holy Word is pushed to the beginning against Death wizards.
I see no other relevant modifiers.

I'm not so sure about pushing Incarnation early. The AI isn't that good with heroes, but the hero buffs the stack so it greatly raises the threshold of how strong targets the AI can attack and defeat in automatic combat. So it's very powerful but it's only for one stack which might or might not be at a relevant place. It's also expensive to research.

Angel being the only rare summoning spell other than Incarnation makes sense to pick early.
I see no good way to guarantee the economy strategy - we can either pick the two buffs without Divine Order, or pick Divine Order without the buffs and either case we get no Angels which is really the only relevant thing the realm has for combat at this tier beyond unit buffs.
Picking Angel and Divine Order seems the most solid choice, even though Divine Order makes Angel more expensive to cast.
Picking Angel and Inspirations might work better, but wastes a lot of overland skill to not get Divine Order early so it seems a suboptimal play. Still, Inspirations is great as it works both for military and economy as needed.
Divine Order+Inspirations seem best for an economy based approach.

So maybe this ?:
If Peaceful or Lawful or Perfectionist, pick Divine Order+Inspirations.
Otherwise pick Angel+Inspirations.

Death

AI research order is Evil Presence, Drought, Wave of Despair, Wraiths or Vampire, Cloud of Shadows, Warp Node, Dark Rituals, Drain Power, Zombie Mastery, etc.
Gate of Hades is pushed to the top of the list when the human player plays Death. No other relevant modifiers.

I think this might work well :
If maniacal, pick Evil Presence and Drought
If Ruthless, pick Drought
Pick Wave of Despair
Pick Wraiths or Vampire at random (50% chance each)

[Suppanut]

May I suggest?

Nature guaranteed early rare:
Conjurer: Survival Instinct
Chaotic or Theurgist: Call the Wild

Chaos guaranteed early rare:
Warlord: Doom Mastery (warlord wizard would more emphasis on normal unit and bridge it to chaos global buff)

Sorcery guaranteed early rare:
Dark elves or Low resistance races: Reinforced Magic

[hackflow]

And the greatest issue is that there is no combo potential which would require varied choices. At 9 or 10 books you don't have a second realm or at least not relevant amount of books in it. So picking a spell because it's a good combo with another won't happen.
For example, for guaranteed rare the AI might pick Doom Mastery instead of Flame Strike if they have enough Nature books to also get Survival Instinct. There are a few similar choices coded in the AI selection. However at guaranteed early, the AI doesn't have enough books in other realms so they always have to consider what's best for the mono-realm strategy so they would always pick Flame Strike and this is even true for the human player. If I can't build a combo I'll just pick the strongest rare spell(s) every game.

There is still Spell - Retort synergy possible. I.e. with Conjurer makes sense to take a summon or Survival Instinct, with Warlord - Doom Mastery or Blazing March or Reinforce Magic, with Channeller - a combat spell, with Spell Weaver - some economic city enchantment or curse, Cult Leader - Dark Ritual, etc. So AIs don't need to always choose the same. I would also add that they should not always make perfect choices if that makes game more repetitive, add some randomness or vary on personality basis just for fun.

I still see the issue with 9 books, it feels week. And books 7 and 8 don't add that much either. Also there is a general issue that having fewer books you get to later spells faster, is that intended?

BTW, 6 + 6 combos like Doom Mastery + Survival Instinct looks fun but not any strong. One might simply fall behind at the start not having any retorts. What if we guarantee rare from book 5, and book 6 guarantees very rare?

[hackflow]

@Seravy I see that you pay attention to personalities but less to Retorts and none to a starting race. Is it hard to use complicated rules here to take into consideration all of this? Maybe instead of if-elses it would be easier to put scores, i.e. Warlord adds scores to Chaos Mastery and normal unit enchantments, Conjuerer to summons, Maniacal personality to curses, particular races to economic enchantments or unit enchantments again if that is a military race etc. Then select randomly with probabilities proportional to scores.

Maybe not even linearly proportional, i.e. Warlord adds 1 to Doom Mastery, Ruthless 1 to Chaos Rift and Fire Storm, some military race adds 1 to Doom Mastery again. So it's 2:1:1, 50% for Doom Mastery for linear, but 4:1:1 - 66% for Doom Mastery if we square scores.

We may even take into account other guaranteed or early picks, i.e. pair up Roots of Genesis with Gaias Blessing or Supreme Light with Life summons.

[Seravy]

No, it's not hard to add rules for retorts at all.

However the ones suggested before would have no effect compared to what I posted.
AI spell selection does not use a priority or weighting system. (unlike personality and objective selection which are meant to be random, this is meant to always pick the best option consistently for that player, random chance is only used when two options are equally good and functionally similar.)
It uses a list of rules, and the first N matching rules will be the chosen spells if N picks were available. Once the picks are used, any further rules have no effect.

Conjurer: Survival Instinct

It's already picked under the current rules unless another retort or personality used up the pick in which case they probably should have precedence. Early access to Survival Instinct is much less critical than the other spells.
Same is true for the summoned creatures.

Chaotic or Theurgist: Call the Wild

That spell is way too low on the AI's research priorities. Even if it appears early they won't research it.
In general this spell offers very little benefit to the AI and researching and casting it early is bad strategy, so moving it forward on the research list makes no sense to me.

Warlord: Doom Mastery

Same situation as Conjurer. The AI will always pick this spell unless they picked the city curse for Maniacal.

[hackflow]

AI spell selection does not use a priority or weighting system. (unlike personality and objective selection which are meant to be random, this is meant to always pick the best option consistently for that player, random chance is only used when two options are equally good and functionally similar.)

So opponents end up with the same spells at the same time in the game. Maybe adding randomness here would be a good thing to maximize fun?

[Seravy]

Fortunately, AIs that pick 10 books are rare.
Mono-realm wizards roll books between 7 and 10 and dual will never have 10 so only 1 in ~10-12 wizards will be able to pick guaranteed early rares. Without that, spells show up in random order and/or not every spell is available so while the AI's preferences are not random, there is enough randomness from the game rules.

[hackflow]

Fortunately, AIs that pick 10 books are rare.

I guess that is probably enough.

What about 7-9 books being weaker peeks? And 6 + 6 combos very weak ones?

[Seravy]

I don't think 8 books is a pick choice at all, it's probably the strongest.
7 books is weaker but being able to spend 5 on retorts or other realms makes that fair.

Only 9 is in a weird spot because the only reason to pick that instead of 8 or 10 is if you absolutely need the 3 retorts/other books or if you absolutely need the second guaranteed rare.

But maybe that's enough as it's a "best of both worlds" kind of a deal. At 10 books, only 2 picks are left for retorts or extra books which is very limiting, and at 8 only one very rare is guaranteed which is rarely enough for a proper endgame strategy.

I don't think there is anything wrong with 6+6. You get a lot of starting spells, good starting power and a reliable mid and endgame. Fewer books with more retorts give a more risky start with higher potential if things go well.

[hackflow]

But taking 6+6 for combo doesn't feel like any strong. To really get Rare + Rare benefit you need to wait a long time and starting spells and some power doesn't cut it.

> Fewer books with more retorts give a more risky start

I would say Retorts make things less risky instead. I.e. having Alchemy gives you a lot flexibility to absorb economical risks, Warlord is also very reliable and many others. They also provide benefit early which snowballs, by the time you get that Doom Mastery + Survival Instinct you'll be cornered already so it won't matter. Besides you only need 2-3 books to get the most beneficial starting spells like Sprites, Healing or Confusion you don't need 6 books for that.

Maybe allow guaranteed rare at book 5, so that Rare + Rare combo will cost 10 picks? leaving 6 + 6 and 6 + 5 for VRare + VRare/Rare - not sure there are any though and sure as hell it's easier to do inside one realm with just 6 books.