AI

[Seravy]

Everything regarding how the AI works, what is new, etc, will be posted here. Note that this all of this information is for Caster of Magic only, 1.50 AI generally only has a small portion of these upgrades.

Overland spell selection

The AI first assigns weights to each category and decides on a category of spells to cast randomly based on these weights, then selects a spell from within the category, again by assigning each spell in the category a weight and selecting at random.

Summoning
All overland summoning spells belong to this category, plus summon hero, summon champion, summoning spirits, resurrection, incarnation, enchant item and create artifact. Each of these has a specific priority assigned to it, generally the better the creature the higher the priority. Heroes also check for having a free hero slot.
For spirits, the AI wants exactly 3 spirits on each plane, unless every node on the plane is already theirs. If they have 3 or more, the weight of spirits is zero.
For items and artifacts, the AI will never make any until turn 180 - early expansion is way more important and skill is not high enough to afford such an expensive spell. Past turn 180 the AI has a small chance of using Enchant Item and a somewhat higher but still small chance for Create Artifact. However if they have Artificier, they will never use Enchant Item, and will have a 4 times higher chance of using Create Artifact as other wizards.

The priority of this category is increased if the AI feels they need more units (their army power is inferior to others), and it is significantly increased if a very rare summoning spell is available. Having conjurer also increases the priority of summoning, as is having the summoning circle moved from the fortress, being Militarist or Expansionist.

The category always has a reasonably high base priority, even if the AI actually does not have any summoning spell they can use. For the 2.5 version if the AI selects summoning but has no such spell to cast - only possible for wizards playing heavy Life, as Guardian Spirit is limited to 1 per plane, others will always have at least some generic summoning spells - the AI will instead try the Unit buff category instead of not casting a spell that turn.

City Buff
All positive city enchantments belong to this category, as well as Transmute, Change Terrain, Move Fortress.
For casting any of these, the AI first checks if they can find a valid target, if not the assigned priority is zero.
The AI never casts Move Fortress, because it lacks the comprehension of the consequences, and it's generally impractical: the fortress already has the best defenses due to other mechanics.

The priority of this category is increased by being a Theurgist, Perfectionist, or having better city buff spells available.

Curse
This category can only be used if the AI is hostile towards at least one player. Priority is increased the lower the relation to that player is. Priority is greatly increased by Maniacal and Ruthless personalities and greatly reduced by lawful and peaceful. It's also somewhat increased by Expansionist.

The category contains all the city curses: Pestilece, Evil Presence, Chaos Rift, Famine, as well as Warp Node, Drain Power, Spell Blast, Time Stop, Earthquake, Blizzard, Fire Storm, Stasis, Black Wind, Corruption, Raise Volcano, Call the Void.
Each has its own priority assigned, going from lowest to highest: Stasis(1), Blizzard (3), Corruption (10), Spell Blast (10), Black Wind (10), Fire Storm (15), Warp Node (15), Drain Power (20), Raise Volcano (25), Earthquake (25), Famine (30), Evil Presence (30), Chaos Rift (40), Pestilence (60), Time Stop (150), Call the Void (200)

Casting Time Stop requires the wizard to have at least 5000 mana crystals and ofc not having the enchantment in effect already.
Spell Blast is slightly special. In addition to the default, if the AI notices an enemy wizard casting Spell of Mastery, the priority of the category increases tremendously (pretty much at least 90% chance of using it immediately) and Spell Blast will be selected from the category directly skipping the rest of the process. The same happens if the AI has Detect Magic in play, and an enemy wizard casts any spells on the list of "forbidden" spells. The latter doesn't look at the cost of the spell since it's not known how far the enemy progressed in casting it, it will assume the AI always has enough mana to blast the spell - if not they'll still try but nothing will happen. See the list of spells getting blasted in the last post.

The target for city curses is the "best" city the enemy has, which means the highest total of buildings and population, the fortress has a minor extra priority on top of that. (in vanilla AI the fortress received such a massive priority boost that nothing else could get targeted at all!)

Unit Buff
This category contains all unit enchantments as well as Nature's Cures, Conjure Roads.
The AI never casts Nature's Cures and Conjure Roads, since the former is too situational to just thrown around at random, and the latter may backfire by allowing enemies to move faster.
The AI does not check if they have a valid target for any of these spells, they assume they do - which is usually correct since they produce a lot of units.
As usual, each spell has its own priority, the stronger spells have higher.

The priority of the category depends on the strongest such spell available, having 3 tiers: weak, medium and powerful, with a priority of 25, 60 and 200. Additional large priority is added if the wizard is a perfectionist and some if they're militarist.
Unless the spell has additional conditions, the priority to target a unit is proportional to the unit's calculated military power plus a quarter of the power of other units in the same stack, plus 100 for the fortress and 50 for heroes.

Spell of Mastery
If the wizard has SoM this "category" gets a priority of 500. SoM is the only thing in the "category".

Summoning Circle
The AI has a timer of ~10-30 turns and alternates the location of the circle between their fortress, and their most remove city they can find. This "category" which only contains this one spell has a priority of 200 if the timer expired.

Disjunction
Each global enchantment owned by other wizards adds to the priority of this, each type of spell having a specific priority. The numbers are currently being reworked.
Allied spells are ignored. However, at the time of targeting a spell, allied spells will be considered if Spell Binding was cast - not if Disjunction.
Disjunction and Spell Binding belong to this group.

Disenchant Area
Each city curse or warped node adds to the priority of this.
Contains Disenchant Area only.

Global Enchantment
Each uncast global enchantment available adds to the priority of this. If chosen, each enchantment has its own priority to be the first to get cast, better enchantments have higher. Enchantments already in effect have 0 priority.
Also contains Great Unsummoning and Final Wave, these can be used at any time and have no special conditions.

Divine Order is special. It'll be only cast if the wizard or their allies are having enough Life magic books or enemy wizards have enough Chaos or Death, in particular if the sum of these exceed a certain threshold.

Theurgists have an increased chance to pick this category.

Note : This list is incomplete and outdated since version 4.0 when the AI overland spell selection was completely remade. While the generic concept is the same, there are more spell categories, such as "water walking movement enabling buff" and more details in it.

Overland unit movement, part I - AI turn

Before any AI's turn, all AI players disband their obsolete units simultaneously if, and only if, the global max unit cap of 2000 is getting near. This was updated in 2.4, on top of disbanding units weaker than a certain threshold (I think it's set to one third of the average of the army), they also disband a fixed amount proportional to their army size: 1 unit for each 100 they own. Spirits, settlers, heroes, ships, engineers, etc cannot get disbanded this way, and higher level units and buffed units are worth more. This part of the code is not rocket science but should be enough to slowly replace crappy halberdiers with awesome sky drakes without ramming into the unit cap head first. In most games the cap will not be reached anyway.

The AI starts their turn with their spell being cast. If complete, they start a new spell. In 1.40 and below vanilla, they did not proceed with that spell, and left it for next turn even if they had enough skill to complete it. In CoM and 1.50 they proceed and keep casting while they can. Doing this first ensures anything they cast will be considered when deciding what to do.

(actually they do a few stuff prior, like evaluating continent values, selecting a target continent, updating hostility etc)

Then they decide how to distribute their power base to mana/power/reaseach, and use alchemy if needed.
Then they set tax rates and farmers: 2.5 will bring a major update to this part and makes the AI able to set these up to way that's reasonable and safe. Finally they set production in towns and then proceed to the main part, assigning orders to units.

Once the procedure to assign orders is completed, all units perform their assigned orders (move, attack, whatever) and the AI ends their turn, the next AI's turn starts. This is new for the mod and the 1.50 patch:  In the vanilla game, first all 4 AI players assigned orders, then all 4 executed those moves, so they effectively had their turns simultaneously, without being aware of what the wizard earlier in the turn order did: resulting in attacking each other when not intended or failing to do so when intended, as the other wizard's units were moved.

Also only in the mod or 1.50, the AI starts by cancelling all existing orders on all units - except settlers and engineers, those would not work as intended otherwise - and assigns new orders, most likely to continue whatever they were doing, but if circumstances change they might not. In the original, once the AI assigned an order to a unit it was carried out, and could not be interrupted: A unit sent to attack an enemy city would still do so 20 turns later when reaching it even if the city is no longer there or peace was made.

Overland unit movement, part II - Assigning unit orders

This consists of several steps executed in a set order :

1. Any unit the AI cannot pay maintenance for are disbanded immediately. The AI disbands their worst units first. The AI is not subject to normal maintenance rules: if they fail to actually pay maintenance, they do not lose units - the game assumes this isn't possible in the first place since they disbanded any excess already. Insecticide improved this procedure by allowing the AI to keep their units if their income was negative but they had enough gold/mana to still pay for them.

2. AI stacks that are standing in a tower on the home plane of the AI switch to the other plane if able. The mod adds the ability to the AI to swap planes for units standing on an Astral Gate as well: in this case the city must have at least 6 units remaining, and the capital cannot be left this way. Furthermore the AI can swap planes with Plane Shifting or Planar Travelling units as well (item power not included). In all cases, movement happens from the home plane to the "other" plane. The AI goes with the assumption that they need more forces on the plane where they not started from, which is 99% correct and making it smarter would be close to impossible.

3. Any AI ships on land (usually produced in a city) are ordered to move onto the nearest sea tile.

4. Next all AI units patrolling near a poorly defended city cancel their orders. Since the orders are always cancelled on all units for the mod, this no longer serves any purpose, nor does any harm.

After this the following are done for each continent, one continent at a time :

5. The AI makes a list of stacks they have available on the continent, only idle units are considered in these but since they now reset orders each turn this is generally all of them.

6. Any melding, road building or outpost creating unit is sent out of any city.

7. (updated in 2.9) Any unmelded node occupied by own or no troops pulls the nearest idle spirit towards itself.

8. Any settlers are immediately ordered to build a town if able, otherwise they are sent towards a settling location of preference.

9. Then any units that can purify are sent to do so if needed.

10. Any engineers available are ordered to either move to the closest city, or to build a road from that tile to the nearest unconnected city. The engineers are now able to build around obstacles instead of trying to build through them.

11. Next, the AI builds a list of all available targets on the continent:  enemy fortresses, cities, units,  nodes, lairs, towers. in this order. Unfortunately this list is limited to 25 things, on very large continents the AI will be unable to add to the list and will overlook anything that would be near the end of the list: towers, lairs, etc. Fortunately, if their is enough enemy presence to fill the list, the AI should most likely ignore nodes and lairs and should not send their troops away to the other plane in towers, so this is not particularly bad. The only possible problem is when there are more than 25 stacks of enemy units and the AI might not notice, and fail to attack something that would be better as a target, still with 25 targets, it's very likely something will be attacked and killed, reducing the count and allowing other stuff to get on the list. Unfortunately, extending the size of the list is impossible.

12. The AI goes through any remaining stack that's still idle,
(new in 2.9) and checks if there is Jihad hostility towards anyone. If yes, the stack is at least as strong as 9 halberdiers, and the capital is on the continent and reachable, they are ordered to move there. During Jihad hostility, enemy troops are not considered roadblocks and the AI will try to move through them - though considers these tiles more expensive to enter than mountains. During any other level of hostility, enemy troops are considered roadblocks.
...and sends each to the highest (priority/distance) target that is weaker than 1.33 times the power of the AI's stack to attack it then erases the target from the list by setting its priority to -1. In the mod, the target is instead retained but set to the lowest priority of 1, also targets with (priority/distance)<1 will be ignored unless distance is less than 5 to avoid sending multiple stacks from far away locations to the same target, however multiple stacks from a nearby location are fine. This not only makes the AI able to attack the human player's city with multiple stacks which it wasn't able to do originally, as well as speeding up transport through towers, but if the units manage to actually reach the tile in the same turn, the two stacks merge together as a side effect, helping the AI to build up larger stacks in multiple locations without having to rewrite the entire stack building process.

13. Any stacks still left over will be sent towards the "main action continent" if they are able to move through water, from any continent that is not "under contest with an enemy". In the mod, directly prior to doing this, the stack will check if there are any viable military targets nearby: while "12." covered targets on the continent itself, doing this allows the AI to perform intercontinental or land to sea attacks which were impossible without this addition. This is probably the largest addition to the AI in terms of code size (and maybe even importance). In the 2.5 update this was changed to not only consider units as targets but cities as well, which does not matter here, but does if called from "18." instead. In 2.9 the stack first checks for Jihad hostility, if the capital of any player whose assigned hostility level is Jihad is on the same plane and the stack is at least as strong as 5 halberdiers, the stack will move to attack that capital. In 5.0 and later, this now uses identical target ratings as land at based attacks, including lairs.

14. Any remaining unit on a continent rated as "no target to attack" or "entirely owned" will go to shore and attempt to board a ship. If no ship is present, it'll mark the location as a "ship required point".

15. The AI will now look through any remaining, still idle stacks and makes a list of units that can be used for building a new stack. The best 9 such units are selected. This was originally done between 11 and 12, which is a major flaw, since the list is used in 16, so units on the list end up with conflicting orders: they'll both be sent to the stack building point and to attack targets/board ships/whatever. This can tear already built stacks apart by pulling some of them back to build a new one, too.
The AI will be able to select any amount of units from any stack not standing on a strategic point. If they're on a node, at least 3 units need to remain but the rest can be used. In cities, at least 6 units need to be present and at most 2 can be sent for the first 50 game turns. After that however, at most one unit can leave the city per turn and only if it has 9 units, also units in the capital can never leave. While this leaves the units on the way vulnerable, it ensures the city is not going to get lost, which is far more important. Always the 9 most valuable units (highest military rating) will be added to the list in order of value.
It's important to mention here that although it's not an actual stack building mechanism, when the AI produces units in a city that's full, always the weakest unit will be pushed outside to ensure the city defenses remain appropriate. However, if these units are not chosen later for stack building due to better units being available, they'll stay there and accumulate, effectively forming a stack on the spot automatically without AI interference, and these stacks are much less vulnerable to getting picked off one unit at a time prior to forming, since no actual movement is needed to be done by units.

16. The AI counts the number of units at the intended stack building tile, and sends as many there to fill up the count to 9 (or the intended lower amount). The mod also changed this, the original send units in excess of 9, which is a problem: if the two stacks cannot merge together due to having over 9 total units, they'll not move, and no unit will be added at all. The stack building tile is the center of their empire on the continent, or the shore if the AI intends to leave the continent.

17. Finally, the AI checks their own cities and nodes, and if there aren't enough defenders, sends any remaining, still idle units there. The AI considers nodes as needing more defenders up to 9 units even though they are also allowed to leave no more than 3 in their if those units are needed to build stacks: this ensures no valuable unit like a hero is stuck on eternal node guard duty, instead new units will be added to the guarding stack, and the valuable units will be able to leave it to build a new stack.

18. After all the continents are done, the AI checks for units over sea. If they are not capable of moving on water, they are disbanded (drown), since the original game did not contain a real drowning check. This is now unnecessary since they would drown on their own anyway but this way they cannot move and do stuff like attacking someone before drowning. Then, any remaining unit is sent to the main action continent. However, in the mod, units first check for a potential target, using the same procedure as "13.", allowing the AI to launch a sea to land attack on a nearby target.

19. Finally, any still remaining stacks on water are checked for ships. Stacks containing ships and units head towards the selected shore tile of the main action continent. Ships already on shore of that continent send the units to land as well as ships on shore of any continent marked "unoccupied by anyone/interesting" so passing by another interesting continent on the route might make the ship stop and unload the troops to explore a new continent. Stacks containing ships and settlers are sent to the "settling continent" instead the same way, if such is not available, they head to the main action continent instead. Stacks containing only ships look for the list of "ship request points" and send one ship to each such point. This procedure had a whole ton of bugs, safe to say at least two dozens in the vanilla game. Finally which the vanilla game did not do but the mod does: any remaining, still idle ship stacks will look for any valid naval target on see to attack. Those that can't find any, will more semi-randomly one tile to a direction based on turn order, which keeps the ships roughly at the same spot but without staying on the same tile. This is critically important since the AI insists on sending their ships to their selected unload points, but never recall them, blocking any further attempts to move ships there to unload. The ships have to be moved away randomly to ensure this does not happen. Another side effect of this is, if a shore is nearby the movement of the ships, held up by the land tiles, will despite being synchronized always moving the same way, result in the ships gathering up in a larger stack. This lets the AI "build" a massive stack of ships that can be used for naval battles very effectively.

20. Starting in 5.0, there is an additional step at the end - the AI will pull their best 9 water movement capable units together, excluding units that already have an attack order. This allows the AI to build a powerful intercontinental doomstack. More details in the thread where this feature was announced.

Other AI overland actions

Buying in cities

Design concept: Rush buying any specific building or unit over others is rarely a relevant difference - it merely allows the town to produce the next item (spent gold/2) production earlier. Since the AI selects what to build on weighted random priorities, speeding up a particular city over others is not providing a significant benefit. Even speeding up a major unit production center does not, as the AI lacks the ability to comprehend "I need more of these units" or "I don't need more of these units". For every case when the massive hordes of incoming Minotaurs or Paladins is helping the AI push ahead and win, there will be a case when it merely helps prolong a war between two AI of even strength, wasting both their economic potential, or worse, the units might be made in a long period of peace when they have nothing to do at all - and might not have even as long as the final war over Spell of Mastery.
Due to this, the AI should use the "BUY" button in two cases: 1. when they have an exorbitant amount of money they have absolutely no reason to keep around, or 2. if the bought building pays returns the investment by produced resources, or accelerates the production of the first few quality troops (either for the city or in general).


Currently defined rules :
-do not buy if can't pay the cost (obvious)
-If gold>10000 then buy
-Is sawmill then buy
-if gold>=50*cost to buy then buy
-if Sage's Guild, Amplifier Tower, Wizard's Guild, Alchemist's Guild or Fighter's Guild and gold>=2*the cost then buy.

This procedure was completely remade in 4.0 and is now more detailed. The AI will generally focus on military spending in adamantium/mithril cities, and cities present on mostly enemy controlled continents.

Alchemy
Design concept: Alchemy is a tool to convert resources in emergencies, unless the wizard has the Alchemy retort, in which case it can be used for generic purpose as well.

Rules:
-If casting the Spell of Return, convert all gold into mana every turn
-If alchemist, before turn 30, convert into having a ratio of 7 gold to 3 mana.
-If alchemist, after turn 30, convert into a ratio of 8 mana to 2 gold. If this would result in more than 24000 mana, make 24000 mana and all the rest as gold. If this would result in more than 30000 gold, put the leftover into mana again.
-if not alchemist and mana<gold/16, convert half of the available gold.
-If not alchemist, 10% chance to convert half of the gold if gold>4*mana. Skip if gold<100.
-If not alchemist, 10% chance to convert half of the mana if mana>8*gold. Skip if mana<100.


Power Distribution
-If time stop is being cast, set 100% mana, otherwise
-if sorcery conjuction and sufficient mana then set 90% skill 10% mana, otherwise 50% skill 50% mana, otherwise
-if spell of mastery is being researched, neither of the below conditions that would null research priority are true, and we could research it in fewer than 50 turns by that setting, set 80% research, 20% mana if sufficient mana, otherwise set 60% mana 40% research, and improve relations and peace interest with every other wizard. Otherwise
-if insufficient mana, set 100% mana, otherwise
-if casting spell of mastery already, set 90% mana, 10% skill
-if below turn 60, and researching Giant Spiders, Gargoyles, or Lycanthropy, set 80% research 20% mana
-if below turn 30, we have fewer than 250 mana, set 100% mana but if fewer than 26 casting skill, and over 100 mana, set 25% skill 75% mana instead.

Default priorities: 4 research, 2 mana, 4 skill
-if spell of mastery is being researched and someone else who is not banished has spell blast, lower research priority to 0
-if spell of mastery is being researched and we are at warlike hostility with the human player and we aren't in the top 25% of the historian chart, research priority to 0
-if theurgist, research priority +3
-if perfectionist, skill priority +3
-if militarist, mana priority +2
-if peaceful personality, research priority +3
-If time stop is known, mana priority +10
-if research behind the human player by 25% number of spells, or fewer than 15 spells known: research priority +3
-if casting skill does not exceed the human player by 50% or is below 80, skill priority +2, but only if the above wasn't true
-during war, both of the above are ignored, as they are overridden by the "more summoning needed" global tactic
-if sage master, research priority +3
-if archmage, skill priority +3
-if mana focusing, mana priority +2
-if chaneller, mana priority -1
-if alchemist, mana priority -2 (but can't go below zero)
-if own enlightenment in effect, halve research priority
-if spell of mastery is being researched and no one knows has blast, we are in the top 1/16 of the historian chart, it's past turn 200, and human player is below 10/16 height on the historian graph, research priority +5
-if casting skill>500, halve skill priority
-if not researching a spell, set research priority to zero
-if over 20000 mana crystals, set mana priority to zero
-set 20% mana, and distribute the remaining 80% randomly weighted by the above priorities.

Taxes

Design concept: The AI should aim for the highest tax rate they can get without losing too much production to rebels. Only the overall empire is considered, even if a single city or two starves, it's acceptable if the others can pay the high tax without rebels.

Rules :
-Overall rebel rate of empire population<10%: Increase tax rate by 1.
-Overall rebel rate of empire population>30%: Reduce tax rate by 1.

Food
Design concept: The AI does not actually have a desertion check - instead, they voluntarily disband any units they can not afford to maintain during their own turns. This means, the AI can never lose units due to conquered cities, and does not need to produce additional food, only the amount actually consumed.

Rules:
-go through each city one at a time. If total food is too low, add a farmer, assuming it produces 1 food. If total food is too high, remove a farmer and assume it has taken away 3 food. Change only once per city. If all cities are done, start from the first one again. Keep going until produced food equals needed amount+2.

Diplomacy
See the Diplomacy section in the wiki for MoM 1.50. CoM's diplomacy system is mostly identical. (some numbers, especially difficulty modifiers, have been tweaked)
There is one new addition: when the human player uses disjunction, spell binding or spell blast against an AI, it will trigger a Diplomatic Reaction with the appropriate warning message and penalty, proportional to the cost of the spell lost.

[Seravy]

Combat Spells
When the AI has the chance to cast a spell, it'll calculate a priority score for each and every spell they are able to cast.
Afterwards, they add a random D20 value to each of them and select the highest. This means that any spell that has at least 20 fewer priority that the best will never be chosen and even at a difference of 10+, choosing the weaker spell is very unlikely (as it has to roll 10 higher than the other spell), while in case of spells with near equal priority, the chance is still close to 50-50 for either. This ensures the AI will be not too predictable, but will never select dumb spells to cast (unless the priority calculation for said spell is dumb).

When calculating priority, first the AI checks the combat situation, my calculating the total power of armies on both sides. Based on the numbers, it rates the combat (hopeless, very bad, slight disadvatage, neutral, advantage, easy victory).
Most spells are assigned to a group which receives priority based on combat situation, according to this table (CoM (and 1.50) only, vanilla table is quite dumber) :

Code:

     L3    L2    L1    N    W2    W3
A    50    0    10    20    25    30
B    -100    -20    0    15    25    35
C    0    10    20    25    20    10
D    50    30    20    10    0    0
E    20    25    20    25    10    -10

Where categories mean
A : Direct damage spells or equivalent (like fire bolt)
B : Buffs, Curses, stuff that is risky or better to use when having some advantage already (like warp creature or web)
C : Powerful universal spells that help decide a combat in the AI's favor (like prayer or most combat global enchantments)
D : Spells that can turn around a lost battle but are meaningless if already winning (flame strike)
E : Summoning or equivalent (phantom warriors, possession)

Individual spells will be explained below. A spell automatically receives "invalid" priority if no target is found, so it cannot be chosen.
If only one own unit is in the battle, row E gains +20 to all values. If only two own units, it gains +10.
If there is only one enemy units, row A gains +18 priority. If only two enemy units, it gains +12.


Unit action selection
Units that have a ranged attack, spell ability or caster will always use that while able. Units that don't have any of these will use a melee attack. Units that have both a ranged attack and caster will decide based on a random chance, the higher the ranged attack strength the more it is used first. Spell abilities (including spell charges in items) always take priority over both ranged attacks and caster ability.
Vanilla AI is quite dumb regarding this, and will try using caster and item charges even if the spell has no valid target or is bad for the AI, resulting in either stupid things (like magic vortex on own city from an item) or the unit freezing and not doing anything (since it has no way to spend the remaining mp.). One of the 2.6x patches fixed both of these problems.

2.6x also added the ability for the AI to move their units prior to casting or shooting, in this case the unit will always exactly move as much as it can to still have a movement point to use the intended spell or ranged attack. Units will decide depending on their stats and abilities whether to 1. charge forward or 2. move backward as much as possible, or 3. try to maintain a distance of 6 from the nearest enemy.

Combat attack target selection (melee and ranged)

The AI calculates the priority of each target, and will attack (or move towards if it's too far) the one with the highest calculated priority. No random factor is involved in this decision. Targets with a total priority below 0 cannot be selected. In case of targets with equal priority, further conditions exist, usually the more damaged unit will be attacked first.

Priority calculation :
1. Add priority equal to the attacking unit's attack.
2. Subtract effective enemy armor against the attacker's attack type (this considers immunities, resist elements etc)
3. If the target has a ranged attack, add 3+(ranged power)/4 priority.
4. If the target is confused, subtract 10
5. If the target has meld, and the combat is at a node, add 20
6. If attacker's defense is over twice as much as the target's attack, priority -10 (why? that should be a great target to attack! Might be a bug? Maybe this should be +10?)
7. If we attacked this target the previous turn, +2
8. If there is a city wall, +5 for the unit at the gate. -100 for targets standing behind still existing wall squares. Wall crusher, merging, teleport, noncorporeal can ignore this.
9. If there is a wall of fire in between and a simulated hit from it would likely result in the death of the entire unit, the target is -100.
10. +12 if the target is black sleeping.
11. If the attack is a melee attack, and the target can be reached and hit this turn, +6. Additional +5 if it's also already wounded. Additional +12 if we are equipped with thrown, gaze or breath attacks. Starting at 2.6x, this actually measures the real distance (considering terrain) and if the target is not reachable at all (like surrounded by others) it will not get selected anymore.

Flying units are invalid if using melee on a unit that has no ranged, thrown, breath or gaze attack to enable the combat.
Invisibile units are not considered at all unless they are already seen.

[Seravy]

List of spells and the AI's ability to use it.
Unable : The AI never casts this spell.
Dumb : The AI can use the spell but isn't aware of what it is for and does it in a very inefficient way.
Average : The AI uses the spell for what it is meant to do but doesn't really know any extra strategies with it.
Semi-Smart : The AI knows how to use the spell well in some cases, but overlooks other possible uses.
Smart : The AI uses the spell as well as an experienced human player would.

Overland Spells
Nature's Eye : Average - See city spell category
War Bears: Average - See summoning spell category for details
Water Walking : Smart - Belongs to medium priority tier. Prefers settlers, avoids units that can already move on water. Casting it on a settler allows it to reconsider its destination.  Unfortunately no connection with the unit movement systems so advanced logic such as "this unit should go there but water is in the way, can I cast water walking know" is impossible.
Sprites: Average - See summoning spell category for details
Earth Lore : Unable - AI doesn't have a scouting system this spell can work on.
Giant Spiders: Average - See summoning spell category for details
Change Terrain: Semi-Smart - AI will only cast this spell if a target worth transforming exists. AI transforms deserts and swamps to grasslands, but doesn't target volcanoes, mountains, hills and forests.
Lank Linking : Smart - Belongs to low priority tier. Prefers to target fantastic units. If a target is not fantastic, it won't be targeted if it already has pathfinding, wind walking or a movement of 1.
Cockatrices: Average - See summoning spell category for details
Transmute : Semi-Smart - AI will only cast this spell if a target worth transforming exists. AI will transform iron and silver into mithril, coal and gold into adamantium. AI will not use the spell for other transformations, and might end up with multiple adamantiums on the same city.
Nature's Cures : Unable - This spell is pointless to cast without a connection to military control systems that react to a damaged stack immediately.
Great Lizard: Average - See summoning spell category for details
Elemental Armor :  Semi-Smart - Medium priority tier - See unit buff category for targeting.
Stone Giant: Average - See summoning spell category for details
Iron Skin :  Semi-Smart - High priority tier - See unit buff category for targeting.
Blizzard :  Semi-Smart - AI will target the largest possible enemy stack that is closes to the most amount of owned units to increase the chance of attacking the damaged stack. However no direct connection exists between the spell and the military systems.
Earthquake :  Semi-Smart - See "curse" category for targeting. AI is now aware of the spell bypassing Nature ward and being blocked by flying fortress, and will target accordingly.
Gorgons: Average - See summoning spell category for details
Survivor Instinct: Average - AI will always cast global enchantments at a high chance if they don't have them in effect yet.
Gaia's Blessing : Average - See city spell category
Regeneration :  Smart - High priority tier - Will prefer stacks that already have the most possible regenerating units. Will not target a unit that can regenerate already (items are an exception as those can be removed from heroes later).
Behemoth: Average - See summoning spell category for details
Awareness: Average - AI will always cast global enchantments at a high chance if they don't have them in effect yet. This spell enables the AI to target any city with a city curse, so it does have a purpose, although pointless after it was already cast.
Colossus: Average - See summoning spell category for details
Earth Gate : Semi-Smart - AI can cast the spell again and can now use the gate to reorganize the units. AI will move the 9 best units to the fortress and divides all other units evenly to have roughly equal garrisons in all other cities. Units are swapped every turn so lost units or units leaving for stackbuilding will be replaced next turn.
Herb Mastery: Average - AI will always cast global enchantments at a high chance if they don't have them in effect yet.
Great Wyrm: Average - See summoning spell category for details
Nature's Wrath: Average - AI will always cast global enchantments at a high chance if they don't have them in effect yet.

Resist Magic :  Average - Low priority tier - See unit buff category for targeting. - Unfortunately able to target units immune to magic due to lack of space, same for charmed.
Floating Island :  Average - Can only summon if AI has fewer then 3 ships total. Can only summon after turn 30. Summons next to own city, not where units want to board a ship, the two systems cannot be connected.
Guardian Wind :  Semi-Smart - Low priority tier - See unit buff category for targeting. - Will not target missile immune units
Focus Magic :  Smart - Medium priority tier - Prefer caster units. Avoid units with no ranged attacks. - See unit buff category for targeting.
Nagas: Average - See summoning spell category for details
Spell Lock :  Smart -Prefer units that have more buffs on them. Can't target nonfantastic units that have no buffs at all.
Conjure Roads : Unable - Enchanted roads can help accelerate the expansion of the human player, and are too risky to use for the AI. Better to rely on engineers for normal roads instead. Targeting would be difficult for this too.
Dispelling Wave :  Smart - Used as a hostile spell against enemies only. (see curses) Priority within the curse category equals the amount of priority the best target has, which is proportional to the number of unit enchantments and city enchantments in the area, with some city buffs being worth more than others.
Flight :  Smart - Medium priority tier - Prefer settlers. Can't target flying units. - See unit buff category for targeting.
AEther Binding: Average - AI will always cast global enchantments at a high chance if they don't have them in effect yet.
Aura of Majesty: Average - AI will always cast global enchantments at a high chance if they don't have them in effect yet.
Spell Blast :  Smart - Used as a hostile spell against enemies only. AI will cast Detect Magic is Spell Blast is known. If Detect Magic is in effect, AI will check what spells others are casting. If any enemy is casting a spell on the spell blast list (see table below), spell blast receives top priority and will almost always be cast as the next spell. Detection of Spell of Mastery does not require Detect Magic.
Uranus Blessing : Average - See city spell category
Invisibility :  Smart - High priority tier - Will prefer stacks that already have the most possible invisible units. Will not target a unit that is invisible already (items are an exception as those can be removed from heroes later).
Wind Walking :  Smart - Medium priority tier - Will prefer faster unit. Will never target units that move 2 or less. Will not target unit if the stack already has a windwalker in it. Will not target a unit that has the wind walking ability.
Storm Giant: Average - See summoning spell category for details
Djinn: Average - See summoning spell category for details
Sky Drake: Average - See summoning spell category for details
Stasis : Dumb - Used as though it was an overland direct damage spell. Might hit city garrisions etc where it's pointless.
Magic Immunity :  Semi-Smart - High priority tier - See unit buff category for targeting. - Will not target magic immune units
Flying Fortress : Average - See city spell category
Spell Ward : Smart  Count total books of all enemy players (the human player will always be considered an enemy even if they're not). Subtract own books. Prefer the realm that has the highest such number. Prefer better cities. Avoid cities if they contain 3 or more fantastic units of that realm. Cancel the ward if 3 or more such fantastic units enter the city or get summoned into it afterwards.
Great Unsummoning : Smart The AI will use it if there are at least 5 fantastic creatures that can be affected by the spell in play. Allied creatures subtract from the total.
Spell Binding : Smart AI uses this as a disjunction type spell. Priority of targeting has its own table separate from disjunction. AI will not start casting this for ally spells, but may target them if they still exist on completion.
Suppress Magic: Average - AI will always cast global enchantments at a high chance if they don't have them in effect yet.
Time Stop : Smart AI uses this as a hostile spell, only if there are enemies to attack. AI will never cast this unless they have at least 5000 mana crystals. Knowing this spell adds a large amount to "curse" aka hostile spell priority.

Hell Hounds: Average - See summoning spell category for details
Corruption : Smart Will first target ores. Will prefer the terrain with highest food output, closest to the best city the enemy has. Will only target tiles in range of an enemy city. Used as a "curse" type spell.
Eldritch Weapon :  Average - Low priority tier - See unit buff category for targeting
Wall of Fire : Average - See city spell category
Fire Giant: Average - See summoning spell category for details
Chaos Channels :  Average - Low priority tier - See unit buff category for targeting. Will no longer target fantastic units or use it twice on the same unit.
Raise Volcano : Smart Will first target ores. Will prefer the terrain with highest food output, closest to the best city the enemy has. Will only target tiles in range of an enemy city. Used as a "curse" type spell.
Flame Blade :  Average - Low priority tier - See unit buff category for targeting
Gargoyles: Average - See summoning spell category for details
Fire Storm :  Semi-Smart - AI will target the largest possible enemy stack that is closes to the most amount of owned units to increase the chance of attacking the damaged stack. However no direct connection exists between the spell and the military systems, and it will not prefer multi-figure units.
Immolation :  Average - Medium priority tier - See unit buff category for targeting
Chimeras: Average - See summoning spell category for details
Chaos Spawn: Average - See summoning spell category for details
Efreet: Average - See summoning spell category for details
Doom Bat: Average - See summoning spell category for details
Doom Mastery: Average - AI will always cast global enchantments at a high chance if they don't have them in effect yet.
Chaos Rift :  Semi-Smart - See "curse" category for targeting. (prefers best enemy city)
Hydra: Average - See summoning spell category for details
Meteor Storm: Average - AI will always cast global enchantments at a high chance if they don't have them in effect yet. (not longer hits own units)
Great Wasting: Average - AI will always cast global enchantments at a high chance if they don't have them in effect yet.
Chaos Surge: Smart The AI will count the number of Chaos creatures they and their enemies own, and cast or not cast the spell accordingly.
Call the Void :  Semi-Smart - See "curse" category for targeting. (prefers best enemy city)
Armageddon: Average - AI will always cast global enchantments at a high chance if they don't have them in effect yet.

Bless :  Average - Low priority tier - See unit buff category for targeting
Endurance :  Average - Medium priority tier - See unit buff category for targeting
Holy Weapon :  Average - Low priority tier - See unit buff category for targeting
Holy Armor :  Average - Low priority tier - See unit buff category for targeting
Just Cause: Average - AI will always cast global enchantments at a high chance if they don't have them in effect yet.
Heavenly Light : Average - See city spell category
Guardian Spirit : Smart  - Will only be summoned if there fewer than 3 spirits on the plane yet. Will be used instead of Magic Spirit. Will not be used for military purpose at all.
Heroism :  Average - Medium priority tier - See unit buff category for targeting
True Sight :  Average - Medium priority tier - See unit buff category for targeting
Resurrection :  Semi-Smart Used as a summoning spell. Will select the hero based on levels, and a table of priorities.
Divine Order: Smart The AI will cast this based on the spell books they, and other wizards have, considering the treaty status. Assumes the spell helps Life wizards, and hurts Chaos and Death.
Unicorns: Average - See summoning spell category for details
Planar Travel :  Average - Medium priority tier - See unit buff category for targeting
Astral Gate : Average - See city spell category. AI can now actually move units through this but will always leave at least 6 units guarding the city, and will not use it in the fortress.
Lionheart :  Average - High priority tier - See unit buff category for targeting
Invulnerability :  Average - High priority tier - See unit buff category for targeting
Incarnation: Average - Used as a summoning spell with priority far exceeding any other summon, only if the hero is not yet in play.
Inspirations : Average - See city spell category.
Prosperity : Average - See city spell category.
Altar of Battle : Average - See city spell category.
Angel: Average - See summoning spell category for details
Stream of Life : Average - See city spell category.
Holy Arms: Average - AI will always cast global enchantments at a high chance if they don't have them in effect yet.
Life Force: Average - AI will always cast global enchantments at a high chance if they don't have them in effect yet.
Enlightenment: Average - AI will always cast global enchantments at a high chance if they don't have them in effect yet.
Crusade: Average - AI will always cast global enchantments at a high chance if they don't have them in effect yet.
Charm of Life: Average - AI will always cast global enchantments at a high chance if they don't have them in effect yet.
Consecration : Dumb - See city spell category. Will be used regardless of any player owning Chaos or Death spellbooks. Original was limited to only game where the human player's primary or secondary realm is Chaos or Death which is a problem because they might have them as the third realm, and other wizards might have it, so this was disabled.
Arch Angel: Average - See summoning spell category for details

Skeletons :  Average - See summoning spell category for details
Cloak of Fear :  Average - Medium priority tier - See unit buff category for targeting
Ghouls :  Average - See summoning spell category for details
Wraith Form :  Average - Medium priority tier - See unit buff category for targeting
Drain Power :  Semi-Smart - See "curse" category for targeting.
Lycanthropy : Semi-Smart - Will not target settlers, transports, heroes, or units costing over 100. Lowest cost unit has highest priority. Does not consider mithril and adamantium due to lack of space.
Berserk : Smart - Unit must have 4 or higher melee attack. Prefer highest melee, lowest defense unit. Ignore priority from unit position (fortress, stack size). Prefer regenerating unit. Never target a hero unless it is regenerating - items are not considered, only the Regeneration enchantment and unit ability.
Night Stalker :  Average - See summoning spell category for details
Wall of Darkness : Average - See city spell category
Shadow Demons :  Average - See summoning spell category for details
Wraiths :  Average - See summoning spell category for details
Cloud of Shadows : Average - See city spell category
Evil Presence :  Semi-Smart - See "curse" category for targeting.
Famine :  Semi-Smart - See "curse" category for targeting.
Warp Node :  Semi-Smart - See "curse" category for targeting.
Summon Demon :  Average - See summoning spell category for details
Death Knights :  Average - See summoning spell category for details
Demon Lord :  Average - See summoning spell category for details

Magic Spirit : Smart  - Will only be summoned if there fewer than 3 spirits on the plane yet.  Will not be used for military purpose at all.
Summoning Circle : Semi-Smart -The AI will cast this every turn, selecting one of three choices based on turn number : The continent with the fewest own cities minus enemy cities, the fortress, and the "next" city after the previous one selected in this third option.
Disenchant Area : Smart  - The AI only uses this if he has something cursed, the more curses are around the higher the chance.
Detect Magic : Smart  - The AI only uses this is it knows Spell Blast.
Enchant Item and Create Artifact : Semi-Smart -These are used as a summoning type spell. Limited to only be used after turn 180. Item or Artifact selected can has to obey normal game rules, and can contain Spell Charges. Cost of item is no more than 10x the wizard's casting skill for each casting. Artificier never uses Enchant Item and uses Create Artifact at a higher chance. AI will not use these during Nature's Wrath.
Summon Hero : Smart  - AI will never use this is more than 3 hero slots are already taken.
Summon Champion : Smart  - AI will never use this is if all hero slots are taken. AI will not use this during Nature's Wrath.
Disjunction : Semi-Smart - see post below and above. AI uses Disjunction more frequently if they have AEther Binding in effect or are Runemasters. AI does not dispel allied enchantments.
Spell of Mastery : Semi-Smart - AI will not use this is there is another player who knows it, the AI has Spell Blast, and less than 1.2 times the casting skill of that player, as in this case they have no chance to finish first, but can stop the enemy from casting it. Aside from that it is used regardless of diplomatic situation or chance of surviving in the ensuing war.

Combat Spells

Spell groups - each spell that belongs to any of these categories will have "+group name" in their formula which means, add the priority provided by this spell group.
Generic : If the spell targets and no target can be chosen : priority = Don't use. If the target unit is a hero, priority +7. If realm is warded, do not use. All additions of priority are cumulative unless mentioned explicitly. This group applies to all spells, and won't be explicitly listed.
A, B, C, D, E : See above table and description.
DAMAGE : Only used by direct damage spells. Simulates hitting the target with the direct damage spell 10 times and calculates an average of the measured damage. If damage would exceed 1.5 times the remaining hit points of the target, that amount is used instead. Add priority +3 for each point of damage dealt above the first three. Add +20 if the damage exceeds the remaining health of the unit. Add another +20 if it is also a hero.
RESIST : Only used by resistance based effects. -10 priority if the spell has exactly 10% chance of working, -5 priority if exactly 20%, otherwise +2 priority for each 10% above the first 30%. Add 10 more priority if the chance is 100%. If the chance is over 100%, it is treated as if it was 100%.
SAVEFAIL : the amount of resistance the target has against the spell, from 0 (100% success) to 10 (always fail), essentially calculated as target's effective resistance-save modifier of spell-save modifier of equipped items-spell specific modifiers (like exorcise vs undead).
ANTIFLY : If neither Web nor Flight is known, we have zero units that can attack a flier (melee, breath, thrown, ranged, gaze), and there is either a flying enemy in the gate when that is applicable, or all enemy units fly, then add +35 priority.
FLY : if enemy has zero units capable of attacking a flying unit, add +7.
WP+ : Add +15 if the enemy forces consist entirely or mainly of units having no magical attack types meaning Weapon Immunity is highly beneficial.
WP- : Subtract 30 priority if not WP+.
ANTIINV : If all enemy units are Invisible, and no own unit is immune to illusions, add +40 priority.
ANTIFF : If enemy city has a Flying Fortress enchantment and at least one own unit does not fly, add +30 priority.
REGEN : If the target can regenerate, add +10 priority.
VALUE : If the target costs over 200, add +10 priority. If cost is over 300 add another +10, if over 500 a third +10 priority.
DISPEL : Double priority if having AEther Binding or Runemaster affecting the spell. Add 12 priority if battle is over sea.
ISHERO : true if the spell is targeting a hero.



Earth to Mud : Smart  - +15 Priority if at least 1.5 times more ranged power (figures*ranged attack strength) than enemy. Additional +12 on turn 1. Only consider the position of nonflying,nonteleporting,nonmerging,corporeal enemy units. If the target tile already has mud, priority = Don't use.
Resist Elements  : Smart  - (Percentage of enemy army's attack power that uses the type of attack blocked by the spell/4)+1 if enemy wizard has Nature books + Priority B. If the first two components are zero, never use.
Call Centaurs :  Smart - Do not cast in sea combat. Priority group E. +12 priority if enemy does not have units that move 4 or higher and do not have more than a marginal amount of ranged attack power.
Web : Smart  - B priority. If Crack's Call is known, and web would target a flying unit, which is not yet webbed, priority group A instead. +20 priority if it's turn 11 or later. +10 priority when defending a city with flying fortress before turn 4, if the target is a flying, nonteleporting, melee units. +27 priority if target is a flying unit at the game. +20 priority if there are enemy flying units and we have no unit that can attack them. If enemy has more flying units than the owned units capable to attack fliers, +2 priority for each own unit that cannot attack fliers.
Fairy Dust : Smart - Priority A+7+DAMAGE+ANTIFLY. Do not use if target has less than 3 figures.
Crack's Call : Smart - Don't use in sea combat. Priority A+12+REGEN+VALUE. Additional +20 priority as attacker if there is a wall. Additional +25 priority if the target would be a hero. Do not target wall tiles that have no units standing on them - if no unit can be targeted, it means everything flies : breaking walls does not help attacking enemies.
Construct Catapult : Semi-Smart - Do not cast in sea combat. Priority group E+10+ANTIFLY. Target tiles furthest away from enemies.
Ice Bolt : Smart - Priority A+DAMAGE+ANTIFLY.
Elemental Armor : Smart  - (Percentage of enemy army's attack power that uses the type of attack blocked by the spell/2)+10 if enemy wizard has Nature or Chaos books + Priority B. If the first two components are zero, never use.
Petrify : Smart - Priority +17+REGEN+ANTIFLY+VALUE+2*(SAVEFAIL-3)+10*SAVEFAIL*ISHERO-2*missing figures in the target+ Priority group A if chance>40%, D if <=.
Iron Skin : Semi-Smart - Priority C+8+8 if enemy has no Sorcery and Chaos books.
Earth Elemental : Average - Priority E+20, Do not use in sea combat
Regeneration: Semi-Smart - Priority B. Add 12 priority if target is a hero. Do not cast unless combat situation is favorable or better. (last two columns in the above tables)
Entangle : Smart- Priority C. If more than 2 enemy units, or later than turn 3, priority +41. For each enemy unit that ends up immobile and unable to act, +8 priority.
Call Lightning : [b]Smart - [/b]Priority A+ANTIFLY+ANTIINV+ANTIFF. If not turn 1, priority +40. If combat situation is the worst, do not use unless enemy has no ranged power, no unit faster than our fastest, or we have a flying unit and the enemy cannot attack one.

Resist Magic  : Smart  - (Percentage of enemy army's attack power that uses the type of attack blocked by the spell/2)+5 if enemy wizard has Death books + Priority B. If the first two components are zero, never use.
AEther Sparks : Semi-Smart - Priority A+Damage+target's mana/4+ANTIFLY.
Guardian Wind : Semi-Smart - Sum(enemy missile attack*figures)/4 + priority B, or "Don't use" if no enemy unit has a ranged attack (or the strongest is rated 0 after rounding down). No cap on priority, might outpriortize better spells if the enemy army contains an excessive amount of ranged units.
Phantom Warriors : Average - Priority E. Do not use if attacking and enemy has Wall of Fire.
Confusion :  Average - Priority D+10*SAVEFAIL*ISHERO+ANTIFLY.
Counter Magic : Smart  - If enemy spell realms = none, Do not use. Otherwise, priority B+C+(own units*4)-10. Do not use if caster is a unit with below 50 mana. Do not use if remaining casting skill<40. Do not use if remaining mana crystals<120.
Psionic Blast : Smart - Priority A+DAMAGE+ANTIFLY.
Blur : Average - Priority C+4*Owned units.
Dispelling Wave : Semi-Smart - +8 for each own unit with a major curse. +3 for each major buff on an enemy unit. Ignore both of these if the unit was Confused. See table below for the list of major curses and buffs. +10 if any enemy units had Invisibility on them. +20 for each enemy combat global enchantment. An additional +20 for Wrack and Call Lightning. Do not use if still at zero at this point. +DISPEL+ANTIFF+ANTIINV

...

List of spells the AI considers as necessary to blast :

Code:

db 01Ch ;Survival Instinct
db 01Ah ;Earthquake
db 01Fh ;Regeneration
db 020h ;Behemoth
db 022h ;Awareness
db 024h ;Colossus
db 027h ;Great Wyrm
db 028h ;Nature's Wrath
db 039h ;Aether Binding
db 03Bh ;Aura of Majesty
db 045h ;Magic Immunity
db 046h ;Flying Fortress
db 048h ;Spell Ward
db 04Ch ;Spell Binding
db 04Eh ;Sky Drake
db 04Fh ;Suppress Magic
db 050h ;Time Stop
db 06Bh ;Doom Bat
db 06Ch ;Doom Mastery
db 06Fh ;Hydra
db 071h ;Meteor Storm
db 072h ;Great Wasting
db 076h ;Great Drake
db 077h ;Call of the Void
db 078h ;Armageddon
db 085h ;Resurrection
db 08Eh ;Incarnation
db 09Eh ;Crusade
db 09Fh ;Arch Angel
db 0A0h ;Charm of Life
db 0BCh ;Zombie Mastery
db 0C1h ;Death Knights
db 0C4h ;Pestilence
db 0C5h ;Eternal Night
db 0C6h ;Evil Omens
db 0C7h ;Death Wish
db 0C8h ;Demon Lord
db 0D3h ;Create Artifact
db 0D4h ;Summon Champion

List of overland spells the AI currently is unable to cast :

Code:

Nature's Cures
Earth Lore
Conjure Roads
Move Fortress

List of combat spells the AI currently is unable to cast :

Code:

Stasis

AI's preference of picking common spells at the start for each realm :

Code:

Nature
1. Sprites
2. Water Walking
3. War Bears
4. Web
5. Call Centaurs
----------------------------------------this many books guarantees all 10 commons for research
6. Fairy Dust
7. Nature's Eye
8. Resist Elements
9. Earth to Mud
10. Earth Lore

Sorcery
1. Confusion
2. Floating Island
3. Nagas
4. Focus Magic
5. Psionic Blast
----------------------------------------this many books guarantees all 10 commons for research
6. Resist Magic
7. Guardian Wind
8. Phantom Warriors
9. AEther Sparks
10. Counter Magic

Chaos
1. Fire Bolt
2. Wall of Fire
3. Hell Hounds
4. Fire Elemental
5. Corruption
----------------------------------------this many books guarantees all 10 commons for research
6. Shatter
7. Warp Wood
8. Eldritch Weapon
9. Warp Creature
10. Disrupt

Life
1. Heroism
2. Just Cause
3. Healing
4. Heavenly Light
5. Endurance
----------------------------------------this many books guarantees all 10 commons for research
6. Holy Armor
7. Holy Weapon
8. Star Fires
9. Guardian Spirit
10. Bless

Death
1. Ghouls
2. Wraith Form
3. Black Sleep
4. Summon Zombie
5. Darkness
----------------------------------------this many books guarantees all 10 commons for research
6. Life Drain
7. Skeletons
8. Weakness
9. Cloak of Fear
10. Mana Leak  

Minor curses and buffs the AI ignores when dispelling :

Code:

Resist Elements
Holy Weapon
Bless
Orihalcon (cannot dispel)
Weakness
Haste (indistinguishable from the item power)
Stasis

[Seravy]

Regarding AI priorities to use Disjunction.

The obvious best strategy to make the AI win is to set these high, but that's not fun. If the AI dispels all good enchantments immediately, they'll stop having a role and instead are reduced to "dispel bait". The other problem is, if they're high the AI dispels other AI as well, which will actually help the player.
The typical priority of stuff the AI can have ranges 100-300 for summoning, and a few other categories, the generic total being ~500-1000.
So a dispel priority of 50 means every turn there is 5-10% chance to cast disjunction PER wizard. Which is still rather high.
Each spell has a specific priority to dispel and these are added together. For example if I have crusade (25) and someone else has AEther Binding (3) then the priority of casting a disjunction is 28.

One more factor, the AI does not spend max on their disjunctions, so more expensive enchanments will take more tries to get rid of, thus they might need a higher dispelling priority - or not, since they are also more expensive to recast?

In 2.5 the following feature will be added : Runemasters, wizards with AEther Binding, and those who can steal with Spell Binding instead of simple dispelling will have a multiplier on their dispelling priority (3x, 2x and 2x). The aim of this is to make games more varied : some game a lot of dispelling happens, in others, not much. Eliminating the wizard with extra dispelling preference (or getting rid of their AEther Binding) might become part of a strategy that relies on enchantments. In exchange, generic priority of dispelling should get lower.

Eternal Night
This is a relatively low threat to the AI - yes they get halved power from religious buildings but halved is still quite a lot and other sources are unaffected. However, this also is very beneficial to the human player : they can take advantage of lowered resistance scores while the AI - in strategic combat - cannot. So the threat level of this enchantment overall is medium.
If cast by a third party AI, the human player is hurt quite badly : not only are their units less resistant to enemy spells, but they also lose power that they need much more than the AI, so in this case not dispelling the enchantment is almost always more beneficial to the AI.
Death wizards never dispel this.

Proposed dispel priority : 15 (low-medium)

Evil Omens
This hurts the human player far more than the AI, since they typically have less casting skill, and mana crystals. The AI also has an overland casting bonus multiplier for difficulty. However, this spell is the least fun of all enchantments and should probably not stay around for the entire game - or very long.

Proposed dispel priority : 50 (medium)

Zombie Mastery
This spell doesn't directly hurt anyone, and poses a relatively low threat to the AI, unless the human player happens to use those zombies to expand into their territory.
Although zombies are not very strong, they have a lot of combo potential making this a fun enchantment to use, and the AI has some effective few ways to stop them - Great Unsummoning and Holy Word mainly.
AI players with this enchantment add a lot of spice into the game as anything defeated by zombies will be added to the AI's army as undead!

Proposed dispel priority : 8 (low)

Aura of Majesty
Strong effect on diplomacy which might or might not be a threat to the AI, but is generally helpful for them. Even if used by the human player, maintaining peaceful relations while the AI gets ahead can backfire, and among AI players, better relation obviously helps them.

Proposed dispel priority : 3 (low)

AEther Binding
A subtle but powerful enchantment that wins games in the background by boosting casting and dispelling power. Probably much larger threat than it looks like, and should at least sometimes get removed to protect own enchantments from the added dispel power.

Proposed dispel priority : 10 (low-medium)

Suppress Magic
The spell that used to have the highest dispelling priority for the AI up till now, which made it pretty worthless to use for the human player.
Threat level is high : it stops the AI from spamming curses, melding nodes, buffing units and cities, moving the summoning circle but they can at least summon their expensive creatures and use their enchantments without much interference.
The current priority of 2000 for dispelling is way too much if we want this spell to be worth casting.

Proposed dispel priority : 100 (high)

Time Stop
This cannot be dispelled, since you need to get a turn to do that.

Awareness
Not much of a threat, once cast even if dispelled the map stays revealed, and scouting can be done easily using stacks of units if desired. Currently the AI doesn't have a priority for dispelling this added.

Proposed dispel priority : 0 (none)

Nature's Wrath
High threat, the AI loses units and buildings at random. Not much threat to the human player at the same time, they can avoid casting the spells that trigger the effect.

Proposed dispel priority : 150 (high)

Herb Mastery
Pretty much no threat unless at war, and rarely worth attention even then.

Proposed dispel priority : 2 (low)

Chaos Surge
Very powerful boost for chaos wizards that helps everyone playing chaos and hurts everyone else.
Chaos wizards cannot dispel this and others most likely should.

Proposed dispel priority : 30 (medium)

Doom Mastery
Powerful combo enabler which is not that good on its own and makes units vulnerable to anti-fantastic unit spells.
As such, dispelling it is questionable.

Proposed dispel priority : 8 (low)

Great Wasting
A major hit to economy, which the AI can handle reasonably well but the human player cannot. Now that the AI can react to this by lowering taxes and raising farmers as needed, it is much less of a threat, although it still is. The AI probably should dispel it anyway since it's a very unfun spell when used by other AI and a much larger threat in the hands of the human player who can stack other economic attacks with it (like Armageddon, Evil Presence, etc) and completely destroy empires with it.

Proposed dispel priority : 75 (high)

Meteor Storm
A huge threat to the AI since they keep massive amounts of units on the map and are unable to provide them shelter. Also stops them from expansions as it destroys outposts. Hurts the human player quite a bit by destroying buildings, but nowhere near as badly.

Proposed dispel priority : 100 (high)

Armageddon
Takes quite a long time to properly raise power and become a great threat and the AI can handle the economic hit much better since it's not as much as Great Wasting. Still, it wins the game over time and should be removed.

Proposed dispel priority : 60 (high)

Tranquility
Not particularly threatening directly but can be over a long period of time through getting Spell of Mastery. If a third party AI does that, it actually helps the AI to win the game.

Proposed dispel priority : 10 (low-medium)

Life Force
An immediate major boost in power comparable to the long term effect of Armageddon, but situational. If cities are taken, power is lost, so it can be dealt with by conventional war methods.

Proposed dispel priority : 20 (medium)

Crusade
Major boost to military power, but not game threatening on its own, and it desirable to stay in effect if cast by other AI players.

Proposed dispel priority : 15 (low-medium)

Just Cause
A decent economic boost but due to cheap casting cost, dispelling is not worth it.

Proposed dispel priority : 0 (None)

Holy Arms
Slightly weaker than Crusade.

Proposed dispel priority : 9 (low)

Divine Order
Dispelled only if deemed harmful based on spellbooks and diplomacy. It should be removed frequently to keep the game more varied - recasting it is cheap.

Proposed dispel priority : 40 (high)

Charm of Life
Better than even Crusade, probably the best global army buff in the game.

Proposed dispel priority : 25 (medium)

Detect Magic
Now that the AI is aware of the Spell Blast tactic, and even uses it against other AIs, this is an actual threat that should no longer be zero priority.

Proposed dispel priority : 5 (low)

Survival Instinct
Similar to crusade or charm of life but benefits fantastic units, which is a more limited use. Also nice combo potential with other enchantments, making it a fun spell that should be possible to stay in play, but not get ignored completely.

Proposed dispel priority : 20 (medium)

Once the AI decided to cast the disjunction and completed it, they will not use any random chance to select the target.
Instead they'll select spells in this order of priority - if multiple players have the same, the one with the lowest player ID :

Suppress Magic, Nature's Wrath, Armageddon, Evil Omens, Great Wasting, Meteor Storm, Eternal Night, Chaos Surge, Doom Mastery, Life Force, Charm of Life, Crusade, Holy Arms, Zombie Mastery, Herb Mastery, Tranquility, Survival Instinct, Aura of Majesty, AEther Binding, Just Cause, Awareness, Detect Magic.

I believe this should also change to :

Nature's Wrath, Suppress Magic, Meteor Storm, Great Wasting, Armageddon, Evil Omens, Eternal Night, Life Force, Chaos Surge, Charm of Life, Crusade, Survival Instinct, Doom Mastery, Holy Arms, Zombie Mastery, Tranquility, Herb Mastery, AEther Binding, Aura of Majesty, Detect Magic, Awareness, Just Cause.

Spell binding has a similar list which is slightly different - spells having the same effect for all players are lower for example on that list.

[namad]

Isn't divine order most harmful to nature realm? I get that chaos and death are the opposite of life but do death/chaos mages actually rely that heavily on summons?

Maybe make it all three? or maybe make it purely based on life? then again sorcery can benefit a lot from reduced enchantment cost because of things like magical immunity and stasis counting...

Maybe make it based on 1*lifebooks+0.5*sorcery books with no opposed books at all? or let nature count as an opposed book as nature relies pretty heavily on summoning while both life and sorcery are the least reliant on summoning?

[Seravy]

Isn't divine order most harmful to nature realm? I get that chaos and death are the opposite of life but do death/chaos mages actually rely that heavily on summons?

Maybe make it all three? or maybe make it purely based on life? then again sorcery can benefit a lot from reduced enchantment cost because of things like magical immunity and stasis counting...

Maybe make it based on 1*lifebooks+0.5*sorcery books with no opposed books at all? or let nature count as an opposed book as nature relies pretty heavily on summoning while both life and sorcery are the least reliant on summoning?

Summons are only 10% more expensive, but combat spells are 25%. Chaos and Death uses the most combat spells and relies heavily on their availability, and the fewest enchantments (which get cheaper). On the other hand Nature has quite a few enchantments they can use to balance out the 10% more expensive summoning and their combat spells are quite limited (only petrify, ice bolt, fairy dust) so they don't lose nearly as much as the other two.
I'd say Chaos gets hurt the most (they have close to no enchantments and have lots of combat spells and summoning)
then Death (lots of combat spells, quite some summoning, but some tricks actually count as enchantments like possession and lycanthropy. City curses get cheaper too. Still, Death relies very heavily on combat casting to produce undead units, or just to kill stuff.)
Nature is somewhere in the middle (more expensive summons, but cheaper buffs, and the enemy can cast much fewer nasty spells in combat to kill their otherwise pretty tough units)
Sorcery is also in the middle (lots of useful enchantments, but they usually need to summon in combat quite a lot and have some decent combat spells too...but cheaper confusion!)
Life is a clear winner (more than half of the life spells in the game are enchantments, and life is horrible at summoning, their combat spells are also limited to only hurt fantastic units)

[namad]

Sorcery usually relies on phantom warriors but they don't cost that much mana. If two life mages stack divine order though things like statis and confusion start seeming free.

Feel free to leave it like it is. I just wanted to chime in on it.

[Seravy]

Current AI starting common spell preferences, from first pick to last :
- I believe a lot of changes are needed here -

Nature

1. Water Walking (2 books)
Allows the AI to send settlers to other continents without having to rely on ships, and allows other units to expand and conquer neutral towns on other continents. Guarantees the AI will never get stuck if starting on an island.

2. Sprites (3 books)
Although low military power, it can move across water and is very effective against the human player who might not yet have a way to deal with flying and ranged creatures in the early game.

3. Web (4 books)
Being a combat spell, this does nothing for the AI except when fighting the human player, however it's very essential in that case to deal with flying units.

4. Nature's Eye (5 books)
Boost early research to reach uncommons faster. Questionable if it's worth being this high in the list but can help if the AI plays barbarians who have no research buildings at all.

5. War Bears (6 books)
The staple common nature creature that ensures the AI has a decent early game army to attack with and isn't too vulnerable to surprise attacks. Sprites actually do a better job because the AI can afford large quantities of those unlike the human player. Might need to be moved slightly up?

6. Call Centaurs (7 books)
Probably the best combat spell option in common nature, this helps the AI fight the human player better.

7. Resist Elements (8 books)
This can be used overland, but it doesn't do anything for strategic combat so it's only effective against the human player, if they use magical ranged or breath attacks. Can be pretty powerful in that case.

8. Fairy Dust (9 books)
Like all combat spells this also only works against the human player. Being the only direct damage in common nature, it might be worth pushing higher on the list?

9. Earth to mud (10 books)
Highly situational combat spell that might do something but usually won't help the AI.

10. Earth Lore (never)
the AI cannot even use this so last slot is no brainer.

Sorcery

1. Nagas (2 books)
Can move through water and is the only overland summon in sorcery in c/uc. Poison can be devastating against the human player if they don't have ranged or high resist troops. Creature has a quite high power in AI combat, and is efficient in taking out lairs and neutral cities.

2. Psionic Blast (3 books)
Armor ignoring direct damage which is very effective against early game heroes. Not sure if it's worth being ranked this high, most likely not at all.

3. Phantom Warriors (4 books)
One of the better spells the AI can use in combat, also ignores armor like psionic blast and costs less to summon, although it is quite vulnerable to ranged damage and the AI isn't good at avoiding that. Might not be worth 3rd place.

4. Confusion (5 books)
Far the best thing the AI can do to stall and damage the human player's armies, probably should be ranked higher.

5. Counter Magic (6 books)
More of a late game spell than early game, but can be quite effective as such which deserves a high ranking - if the AI gets even more books they'll be able to research it by the time it's needed anyway so later slots would be meaningless. In the 2.5 version the AI learned not to use this when low on skill/mana so it's not gonna waste their chance to cast more important spells on a non-boosted Counter Magic that barely does anything.

6. Guardian Wind (7 books)
Situational but can turn a battle around if the entire army is made of archers. Does nothing for the AI in strategic combat even if cast overland, missile immunity has no effect there.

7. Focus Magic (8 books)
A decent spell that actually powers up AI units in a way that helps in AI only combat as well as against the human player. Not that outstanding but one of the very few things in sorcery that can add to the overland military power of the wizard, and should most likely be picked way earlier.

8. Resist Magic (9 books)
Another spell that does nothing against other AI and neutrals, but can be very powerful against the human player, sometimes. It's an amazing counter to gaze/touch creatures, and death wizards in general. Might worth moved higher than guardian wind which is less versatile?

9. Floating Island (10 books)
For something that enables the AI to move settlers across water, this is ranked surprisingly low and should probably be much higher.

10. AEther Sparks (never)
A decent direct damage spell but psionic blast is generally more valuable for the AI due to defense ignoring (they can afford the higher cost). Probably should stay in the last slot.

Chaos

1. Fire Elemental (2 books)
Guaranteed victory against a human player who does not bother building an alchemist guild with the weapon immunity, but pretty useless otherwise. Probably ranked too high?

2. Hell Hounds (3 books)
A staple summoning spell and the only such at common for Chaos. It's essential for the AI to have an early game summoning spell so this has to be ranked high even if it lacks the flight/water walking advantage of nature and sorcery.

3. Wall of Fire (4 books)
Protects the AI from early elimination by basic units such as swordsmen, halberdiers. Also effective against werewolves and most other frequently used rush tactics. I have been stopped by this quite often and it is definitely worth being an early pick.

4. Fire Bolt (5 books)
A very cost effective direct damage spell that can only be used against the human player but is pretty good at making them pay for their aggression. As it's the only such in common chaos, a high rank is worth it.

5. Warp Creature (6 books)
A very unpredictable spell that only works against the human player. It's versatile and ensures the AI always has something to cast (since it has a high penalty and works on fantastic units), but aside from a lucky 0 resist+instant kill spell it's not essential. Maybe it should be ranked lower?

6. Shatter (7 books)
Same as above but also more situational as it's normal unit only. However it's devastating against the correct targets which could make it worth this slot.

7. Eldritch Weapon (8 books)
Actually boost attack power in strategic combat, this is helpful for the AI even if they do not meet the human player. It also counters weapon immunity (although chaos does that pretty well anyway), and is the only unit buff at common they can use. Might be worth a slightly earlier slot?

8. Warp Wood (9 book)
A very situational spell against the human player only, it deserves to be ranked low.

9. Disrupt (10 books)
Only useful against the human player, who rarely builds a city walls in the early game. Should be ranked low for sure.

10. Corruption (never)
Now that the AI learned how to use this spell effectively to disable ores and highest food yield terrain, as well as having a high chance of using it given the right personalities, this should be ranked far higher, especially as it works vs other AI as well.

Life

1. Just Cause (2 books)
More fame for more heroes and mercenaries, and less unrest so higher taxes are possible. Great economic boost that helps the AI a lot.

2. Heroism (3 books)
Very powerful spell in the early game which can provide a nice resource advantage if cast on a sage/aether/ritual master hero, or just turn common swordmen into an unstoppable force. Heroes buffed by this can be really effective in combat as well. Definitely worth a high rank.

3. Healing (4 books)
A very strong and essential combat spell that helps the Life AI to be able to keep their buffed units alive in combat. Does not work in strategic combat but still a must have.

4. Heavenly Light (5 books)
Generates power for the AI as well as making their cities resistant to early attacks. Essential.

5. Holy Armor (6 books)
A pretty decent buff that helps both in strategic and normal combat and is most effective against the human player who might not have troops being able to damage a heavily armored creature.

6. Holy Weapon (7 books)
A pretty decent buff that helps both in strategic and normal combat but is hardly essential.

7. Guardian Spirit (8 books)
Since the AI can only use this for melding, it's not important to have early or at all.

8. Endurance (9 books)
Although To Defend has no effect on the outcome of strategic combat, the boosted movement can help the AI's expansion somewhat, and units buffed by this, heroism and holy armor at the same time can be a major threat to the human player. Probably should be ranked higher?

9. Bless (10 books)
A situational spell that can protect the buffed normal units from two of the 3 realms most dangerous to them. Does not really affect how well the AI works more like how well the human player can do when paying Death or Chaos so probably fair to keep it near the last slot.

10. Star Fires (never)
Now that it works on all realms, this is a pretty decent early game spell for the AI to defend against strong fantastic creatures : the only thing that can be a threat to their very buffed heroes or units. Should probably picked way earlier.

Death

1. Skeletons (2 books)
A cheap staple summed unit but the AI can afford stronger units and this is pretty weak against a human player. Having these in the capital for the early game exposes it to risk. This spell should probably be ranked much lower and ghouls should be 1st. The AI generally has enough unrest reduction even without these from normal units.

2. Ghouls (3 books)
One of the best things the AI can get, these not only are effective units against the human player, but also against neutrals or other AI, and they create undead, further increasing the AI player's armies. Should be first.

3. Wraith Form (4 books)
Water walking in Death AND weapon immunity. Should be first or second.

4. Summon Zombie (5 books)
A very effective spell to cast against the human player in combat, but does nothing in strategic combat at all.

5. Black Sleep (6 books)
An excellent spell against the human player, probably even better than summoning zombies. No effect in strategic combat.

6. Life Drain (7 books)
Does nothing in strategic combat and generally hurts less than black sleep against the human player. However it's in an opposite spell group (use when cornered while black sleep is "use when at advantage") so both are valuable nonetheless.

7. Darkness (8 books)
Only helps against the human player and only if many death creatures are in combat. Such situation is only likely if many death books are taken, so this should be ranked low.

8. Weakness (9 books)
Only works against the human player and is generally inferior to life drain or black sleep. Should be last?

9. Mana Leak (10 books)
Again, only works against the human player but it's a pretty decent counter against spellcasters, ranged units, and even hurts the player's mana pool. Great vs walled cities too. Should be picked earlier.

10. Cloak of Fear (never)
Since the AI casts this randomly instead of when knowing the enemy has low resistance, this is not a very valuable spell for them and it doesn't even work in strategic combat. Should be last to preserve overland casting skill in the early game to summon ghouls or use wraith form.

[namad]

I might actually move war bears down if you don't buff them, just to get the ai to build even more sprites (they already build a lot) but if you're buffing it moving it up instead might make sense. Nature is my second least played realm in your mod so far.

IMO floating island has to be #1 sorcery pick for any and all reasons water walking is.
I'd consider making focus magic either the first or two second for sorcery (floating island being the other). Why?
If the mage has 10 blue books it doesn't matter at all. If the mage has 2 or 3 sorcery books focus magic combo's the most interestingly with other colors.
I'd put phantom warriors at third then, but I could see it being higher, if the game is set to large landmass floating island should probably drop below phantom warriors but the ai cannot know that can they?
Phantom warriors is probably the best spell in this slot in terms of raw player power but the AI cannot be relied on to find a way off their starting island and later on in the game focus magic yields a lot of very fun surprises ghouls+focus magic is nifty but so is any very rare caster+focus magic which unlocks things like holy word and flame strike.
In some ways a floating island is even better than water walking (in others it's worse).

IMO ghouls should be first, but the rest looks fine. That said if you only have 2 death books maybe wraith form should be number 1, so maybe it should be number 1 always.

I'd move heavenly light down the list, yes it's important but the AI needs holy weapon(armor) more. The ai can usually get plenty of mana and units what it struggles with is making any of them any good.
I'd set guardian spirit to last because in theory knowing guardian spirit reduces the ai's ability to scout with magic spirits.

I just realized I haven't picked a single chaos book yet in your mod.

Also maybe I should look at specific ai opponents, do more people use your wizards or the default ones? I use the default ones because I have fond childhood memories of them, I hear your optional replacements are more well tuned though? For example if there is no AI that gets 1 black common pick (without getting 2) then it doesn't really matter at all what the order between #1 and #2 on the list are?

[zitro1987]

Nature

1. Sprites (-1)
2. Water Walking (+1)
3. War Bears (-2) - you plan to boost these, very different tactics compared to sprites
4. Web (+1)
5. Call Centaurs (-1)
6. Fairy Dust (-2) - I assume you are applying the +1 damage boost to it
7. Resist elements
8. Nature's Eye (+4) - half of spell's benefit is anticipating a battle, more beneficial to humans than AI which builds research buildings more quickly anyways.
9. Earth to Mud - agree, AI uses web better
10. Earth Lore (never)

Sorcery

1. Confusion (-3) - AI is good at stalling you with this spell. Not the best spell, but we're rating AI effectiveness.
2. Floating Island (-7) - should be higher on list, it can help AI expansion
3. Nagas (+2) - Agree on unit being early - water walking, high damage output. Many weaknesses can be exploited by human player.
4. Psionic Blast (+2) - ok spell better in hands of AI than player. Can nuke naga weaknesses (ranged units)
5. Focus Magic (-2) - good boost to units, including ranged poison naga attacks. AI can better affort expensive spells.
6. Resist Magic (-2) - can make ~10-11 resistance units against almost everything ... I guess I can't use nightstalkers, cockactrices, shatter, confusion, etc.
7. Counter Magic (+2) - AI's intelligence with this spell is hit-or-miss. The spell is not as useful early on.
8. Phantom Warriors (+5) - AI not smart with this spell, often casting them away from adjacent melee units then getting killed by ranged units or spells
9. AEther Sparks (-1) - shaman-killer, magician-killer, sprites-killer, not as bad as you make it out to be and generally more cost effective than psionic blast even against non-magic units assuming they have low armor
10. Guardian Wind (never +4) - good spell, but situational and only against human.

Chaos

1. Fire Bolt (-3) - A useful early direct damage spell that AI uses well
2. Wall of Fire (-1) - update
3. Fire Elemental (+2)- update
4. Hell Hounds (+3)
5. Shatter (-1) - I disagree, this spell could be more prioritized. AI can disable multiple units early on due to low resistance.
6. Corruption (-4) - annoying but effective spell at slowing you down early on. Predominantly chaos wizards should start with it.
7. Eldritch Weapon - casting this on most of their swordsmen/halberdiers/bowmen can greatly increase damage output.
8. Warp Wood
9. Warp Creature (+3) - AI may be lucky halving your cheap summonable unit melee. More useful if you quickly research an uncommon unit. Not essential spell for AI.
10. Disrupt (+1 never) - not essential at all.


Life

1. Heroism (-1) - essential spell. Non-life wizards with 2 books should enjoy it and it allows a universal combat spell.
2. Just Cause (+1) - slightly less essential but should remain high and be there for half-life or full-life wizards
3. Healing - AI needs a combat spell, so let's do this one
4. Heavenly Light - another must-have
5. Endurance (-3) - good combo potential with holy armor, but not worth the cost for low-armor units and AI may not pick higher armor ones like guardian spirits.
6. Star Fires (-4) - when AI does not need to heal units, it should have something else to use in combat. May be more cost-effective than casting heroism in every combat.
7. Holy Armor (+2)
8. Guardian Spirit (+1) - protects nodes from being stolen early, so not useless
9. Holy Weapon (+3) - we already have a few buffs, let's push some down.
10. Bless (never +1) - sometimes used against me well, sometimes pointless

Death

1. Ghouls (-1)
2. Black Sleep (-3) - terrifying spell early to mid game. Should be AI's standard combat spell.
3. Wraith Form - water walking, pathfinding, weapon immunity, etc. Very useful in the hands of AI which can afford spell.
4. Life Drain (-2) - another good spell
5. Darkness (-2) update
6. Summon Zombie (+2)
7. Mana Leak (-2) - situational but good enough to move earlier
8. Skeletons (+7) - ghouls are generally better in the hands of AI
9. Cloak of Fear (-1) - overpowered spell at -3 resist, but AI doesn't use it as well
10. Weakness (+2 never)