AI

[Tiltowait]

Sounds pretty cheap, getting to know where invisible units are for free. It's there for the player because it would otherwise break the game but there's no reason for the CP to have the same possibility of breaking the game thrust at it.

[Nelphine]

No. Teleport in and merging are hard for human players to get. Great wyrms, unicorns, djinn are all common in lairs, and invisibility is one of the few tactics for clearing those early.

And, an easier way to know where the invisible units are (for humans) is to know where they started, their speed, and their movement style. Teleport ing merging just let you confirm that - it's not needed at all.

[Seravy]

So I guess I have to implement random moving fpr them as well? Was hoping I can avoid that, it's a bit more complicated and a teleporting unit is very likely to find an enemy in 25 turns if the random movement if correctly done instead of the current one tile at a time like if there was no teleportation on the unit.
I guess I might as well...hope there is enough space.

[Nelphine]

Given that other units only get to detect at the end of their movement, and mergers already randomly pick a place to move to, I'm not sure what else you want?

[Seravy]

Given that other units only get to detect at the end of their movement,

Not since 2.8.
If a unit moves near an invisible unit it will attack if it can still move.
If the unit can teleport, they can usually do it 3-4 times a turn, the chance of finding the invisible unit in 25 turns is very high, even more so if there is more than one unit.

and mergers already randomly pick a place to move to

Also not since 2.8.
At the moment, they move only to the next tile like nonteleporting units, I have to implement moving anywhere separately because being able to stop between moves meant I had to replace the old system which used up all movement points in one go.

[Nelphine]

Ah, I missed that update. Hmm... how many tiles on the map? Each move covers 9 tiles, 4 moves means 36 tiles plus the original 9 you knew plus more based on allied units.

Maybe instead of moving them randomly each move, give them say, a 30% chance of simply knowing before their movement, in which case act as if they can see invisible, otherwise, randomly move once.

Would need to make patch notes very clear on this, but it should have roughly the same affect, and should be a lot less space.

Edit: I'd actually try to give them 10% per movement. So move 3 is 30, move 4 is 40, etc. Might need to be 8 per movement.

[namad]

Given that other units only get to detect at the end of their movement,

Not since 2.8.
If a unit moves near an invisible unit it will attack if it can still move.
If the unit can teleport, they can usually do it 3-4 times a turn, the chance of finding the invisible unit in 25 turns is very high, even more so if there is more than one unit.

and mergers already randomly pick a place to move to

Also not since 2.8.
At the moment, they move only to the next tile like nonteleporting units, I have to implement moving anywhere separately because being able to stop between moves meant I had to replace the old system which used up all movement points in one go.

Really glad this works now. I remember asking for it a few months ago and hearing it was impossible, so congratz.

[Seravy]

The AI used the formula "hero slots in use<=hero ID/4" then hire an offered hero otherwise refuse.
I decided to make this more elaborate as some early heroes are much better and some later ones are not all that good. (Note that fame requirement of these are usually altered but the position in the list is not)
Summoned heroes are unaffected and always accepted, however kyrub also added a condition "do not case Summon hero if number of heroes>=4" so it is of less consequence. Summon Champion can be used anytime but champions are generally good enough to worth the last slot.

I'm adding this table -this is not the actual order it is in the memory, it's sorted for posting - (number is how many already owned heroes means the offer is rejected) :
db 1 ; Dwarf (AI doesn't consider armsmaster when deciding where to move a hero)
db 1 ; Barbarian (Later heroes are just better)
db 1 ; Orc Warrior (Later heroes are just better)
db 1 ; Huntress (All other bow heroes are better)

db 2 ; Dervish (Early gold, aside from that not worth it)
db 2 ; Druid (Solid caster hero with decent spells but everyone else is better)
db 2 ; Rouge (Bard is better for leadership. Legendary is not as important for AI due to maintenance discount)


db 3 ; Bard (Leadership, Holy Bonus, but weak in melee and has bad mana)
db 3 ; Healer (Healers are nice, also has good mp)
db 3 ; Warrior Mage (more versatile than Druid with better MP)
db 3 ; Ranger (Better than warrior mage but bow slot is bad on a wizard/warrior)

db 4 ; Sage (Dispelling Wave is amazing, plus free research which the AI doesn't get that big advantage on)
db 4 ; Thief (Strongest early game melee hero, can remain powerful even in late game with some levels)
db 4 ; War Monk (super agility makes this almost unkillable, AI keeps it in fortress until midgame)
db 4 ; Assassin (Less durable than Thief but blademaster with first strike is just good)
db 4 ; Golden One (The next tier warrior mage)
db 4 ; Unknown (The best warrior mage, with 5 random abilities. Maybe worth slot 5 too? Nah, need that for champs)

db 5 ; Magician (Flame Strike and Destruction ranged attacks...)
db 5 ; Wind Mage (Wind Walk. Horrible stats though, maybe could be lower? Probably worth it, it'll be sent to battle and die anyway, not likely to keep the slot in use too long)
db 5 ; Draconian (Good melee hero that also flies so no flight trickery can stall it)
db 5 ; Amazon (Same as above but with thrown instead. Also Charmed. Best melee nonchampion hero.)
db 5 ; Witch (Black Prayer is this good if the AI has resistance based spells)
db 5 ; Ninja (Invisibility and strong hero in general)
db 5 ; Warlock (Doom Bolt!)
db 5 ; Elven Archer (There are much more important champions)
db 5 ; Knight (Leadership/Armsmaster/Legendary champion is not useful for AI that much as others)

db 6 ; Beastmaster (Soul Linker makes this top priority as it's likely to be used to protect capital along with very rare summoned units)
db 6 ; Illusionist (Best spellcaster hero in the game. Especially as AI now takes advantage of Quick Casting and prefers to equip her with movement and haste items.)
db 6 ; Swordsman (Raise Dead for other heroes, and a good hero overall)
db 6 ; Priestess (Buffs entire army in the battle a lot)
db 6 ; Paladin (Second strongest melee champion)
db 6 ; Black Knight (Strongest melee champion)
db 6 ; Necromancer (Super Soul Linker....if there was a 7th slot I would reserve it for this...Huge with fantastic units in capital, even more so if she has mp left to animate them when they die.)
db 6 ; Chaos Warrior (Most versatile hero in the game, great in both melee and ranged and spellcasting)

Maybe the AI should rank most heroes 1 lower and have a tier specifically for Necromancer and Illusionist which are that much more effective for it? Not sure but they are worth keeping a slot for. Refusing a champion then summoning one which uses that slot anyway is not very good though...

[Seravy]

It seems the elf plague isn't limited to only neutrals. I just discovered thanks to statistics (only 1 AI was playing elves out of over 50), AI wizards also change their race if it was elves and the tile of the capital is not a forest. No wonder I felt there aren't enough elf wizards in the game...

[Seravy]

I always knew something is off with the AI almost never casting a combat spell the turn they finish Spell of Return but this, this, this is just.....worse then I could have imagined.

Stay tuned for the next update as soon as we decide what to do with Berserk (which might take a while actually)

-Fixed bug : AI wizards lose all their mana crystals every turn while banished due to doing alchemy wrong.

Yep, you read that right. Instead of adding the converted mana to their existing mana, they overwrote it. Next turn when they had no more gold they did it again, turning 0 gold into 0 mana. And kept doing that each turn. The only reason why they could finish casting the SoR is they had some gold income so it wasn't actually 0 but a few hundred...