I was testing a lot with the maelstrom and lately with the SOA mod about why the (unfair) AI is so bad. I don't play the higher levels cause they cheat too much (complete research after only 20 minutes, no more influence from the planets they control), taking away many strategic elements how to play the game. So I focus on why the "unfair" AI is so bad and how to improve it.
Some reasons the AI sucks are straight forward:
1. Bad resource management (it simply distributes resources over all areas: planets, modules, ships, research)
2. Bad ship management (wastes complete fleet, attacks from bad position)
Now the solution seams to increase the AI resources and ship build rate. But here the "not so" obivous quirks come in the way:
1. The AI will build huge research queues, finally hitting new ship models
2. It creates huge building queues with ships it actually cannot build, cause the required research is number 100+ in research queue
3. Ships in building queues take up fleet support points
4. The AI does not reserve fleet support points for capitals
Result is that at some point in the game the AI cannot build any more ships before it does not complete its 100+ research items. Also when it has maxed out it fleet support points (no more huge jumps creating free capacity) it will practicaly stop building capitals.
Hint for Mods: Ensure research is more expensive than ships, most important cruisers. Be careful with huge resource bonuses. You need to increase research cost the same time or they can shut down the AI ship production.
Example for additional "cheats" heavily improve the AI:
As I learned from the AI discussion that you can create research items that only the AI can activate, but not the player.
You can created following research item "RESEARCHSUBJECT_AI.entity":
TXT
entityType "ResearchSubject"
hudIcon ""
smallHudIcon ""
infoCardIcon ""
NameStringID ""
DescriptionStringID ""
researchWindowLocation
block 2
pos [ 8 , 0 ]
ResearchField "Fleet"
Prerequisites
NumResearchPrerequisites 0
MinimumArtifactLevel -1
BaseUpgradeTime 0.000000
PerLevelUpgradeTime 0.000000
BaseCost
credits 0.000000
metal 0.000000
crystal 0.000000
PerLevelCostIncrease
credits 0.000000
metal 0.000000
crystal 0.000000
Tier 0
onlyWorksIfTierLabsExist FALSE
MaxNumResearchLevels 1
priority 10.00000
researchBoolModifiers 0
researchFloatModifiers 12
researchModifier
modifierType "ResearchCostAdjustment"
baseValue 0.000000
perLevelValue 4.000000
researchModifier
modifierType "IncomePercLost"
baseValue 0.000000
perLevelValue -4.000000
researchModifier
modifierType "ShipBuildRateAdjustment"
baseValue 0.000000
perLevelValue 20.000000
researchModifier
modifierType "CapitalShipMaxSlots"
baseValue 0.000000
perLevelValue 8.00000
researchModifier
modifierType "ExperienceAwardedAdjustment"
baseValue 0.000000
perLevelValue 2.000000
researchModifier
modifierType "PlanetSlotsCivilianIncrease"
baseValue 0.000000
perLevelValue 24.000000
researchModifier
modifierType "PlanetSlotsTacticalIncrease"
baseValue 0.000000
perLevelValue 8.000000
researchModifier
modifierType "ResearchBuildRateAdjustment"
baseValue 0.000000
perLevelValue 0.250000
researchModifier
modifierType "ModuleBuildRateAdjustment"
baseValue 0.000000
perLevelValue 0.250000
researchModifier
modifierType "PlanetUpgradeBuildRateAdjustment"
baseValue 0.000000
perLevelValue 0.250000
linkedPlanetUpgradeType "Invalid"
researchModifier
modifierType "WeaponRangeAdjustment"
baseValue 0.000000
perLevelValue 0.500000
linkedWeaponClass "Invalid"
researchModifier
modifierType "PlanetSlotsShipsAdjustment"
baseValue 0.000000
perLevelValue 1.000000
artifactPicture ""
This will heavily improve the unfair AI with it still being dependent on the planets it has.
A further step is to use the open Capital building slots and created "AI" copies of the most important ships, set slotCount 0.0000 and add:
ResearchPrerequisite
Subject "RESEARCHSUBJECT_AI"
Level 1
You should now face a much much strong "unfair" AI using fleets with 10+ capital ships, but without the quirks from the Cruel or Vicious AI (having completed all research after 20 minutes).
This are just examples, there might be even better possibility to improve the AI. But avoid the resource-research trap.