Random Start Deployments?

Replayability Improvement:

It's unavoidable that players remember enemy unit positions and no longer need recon for the opening moves. This unbalances replayed games against the defender even in human-opponent games and especially against the AI which is newborn and lacks past-life recon memory. It would be better if each scenario randomly selected a unit mix/deployment from at least 3 different variations.

Is the AA on the left flank or the right flank this time?

Update 8/17/06

A possible way to simulate variable AI deployments during campaigns is to allow the AI to go first and react to any spotted deployments of the player's core, so a player's changed deployment might cause an AI changed deployment before the player's first move. The AI should have adequate spotting (frontline strongpoints?) so that it's not too easy to trick the AI.

An interesting experiment would be to allow the AI to go first and to start the AI with minimum units but plenty of points to purchase its own custom force and make its own custom deployments on its first turn. See if it buys the same thing in the same place every time, or if at least it varies purchases based on a player's core deployment . If the AI tends to spend everything on planes, you either could see if this achieves variable air deployments, or you could make the AI's planes unpurchasable so it buys a custom ground force.

Update 8/18/06

Limitations of the purchase method include restrictions on location (near a town), experience (being new units), strength (start at 10, not 5 or 15), and entrenchment (starting with none).

Update 8/18/06

A small test with PG's Poland scenario which deleted all Polish units except the 3 cavalry and set Prestige Points to the exact value of the deleted units indicates:
  • The AI replaced the number of units
  • The AI bought units that differed from the original, removed units
  • The AI bought cheaper units overall so it did not spend all of its allocation (only 1 unit cost less but this lowered the total sum)
  • The AI bought units that were "more effective" than the original units for the same price (remove an infantry and it buys a 37mm gun which has much higher cumulative initiative/spotting/attack/defense) (See this update comparing infantry and ATG value)
  • The AI arguably deployed better than the original design (instead of a vulnerable, forward artillery and a tank in the last city, the AI placed an anti-tank gun in the last city and forward-deployed its tank outside another city)
  • The AI did not vary its purchases in multiple iterations of its turn
  • The AI did vary its deployment in multiple iterations of its turn by alternating between 2 positions (only 1 non-city unit, the artillery, was removed and it was the new non-city unit, the tank, which alternated between 2 other locations, although the tank's appearance also shifted the local cavalry unit by 1 hex)
  • The AI did not vary its purchases or deployments based on the forward cavalry unit being changed to Guards cavalry with 3-hex spotting range to see an additional German infantry unit
Promising Results:

The results are slightly promising in that providing the AI with a skeleton force and money to customize its forces showed that it automatically varied its deployment among 2 choices, thereby requiring more realistic recon by the returning player. Larger and more extensive testing is necessary to determine if the AI would use any purchase variance or more deployment variance in other situations (1) where more non-city units were removed, and (2) where the AI is presented with a much greater difference in spotted units than tested here.