The Results

During a run, as each battle is finished, a line is written to a logfile describing the results. The files appear in a directopry called "ABlogs" either in your run directory or C:\My Documents\Empire Earth II\ folder. The format of these files is as follows:

Name, TypeA, TypeB, NumA, SpentA, SurvivedA, LostA, RatioA,
NumB, SpentB, SurvivedB, LostB, RatioB....

There are more fields but it is just rehashes of the settings of the runfile.

• Name is the name of the runfile line being tested.
• TypeA are the units in the battle.
• NumA is how many units were created total.
• SpentA is the total spent (note that because of rounding this will probably be lower than 12x resources specified in the runfile).
• SurvivedA is the number of A units that survived.
• LostA is the number of A units that were lost.
• RatioA is the most important value. This number respresents the cost of units killed divided by the costs of the units lost. A number higher then 1 means that the unit is more cost effective. A number less then 1 means that the unit is less cost effective.

The meaning for the corresponding fields for group B is exactly the same as for A, except that it will reflect an inverse relationship in cost effectiveness. That is RatioB will generally be the inverse of 1/x relationship of RatioA since it simply is the costs of units killed by group B divided by the costs of units lost by B. Sometimes a unit that is very expensive yet kills a lot of units, will be proven to be more cost effective, as the weaker but cheap units didn't manage to kill anything.

Testing Methodology

We always tried to use a generic civilization at Mad Doc when testing units, so that civ bonuses wouldn't color the results. This was done by editing "db\techtree\upgrade_factorset.csv" and removing the greek light infantry bonus. Search for CivUpgrade_Greek02, then change the value in column H to 0.

In general we at Mad Doc tried to balance the units in 2 ways. One was to balance within an epoch according to the RPS, then pit that epoch against the epochs around it, adjusting as necessary. The second method was to balance by unit type. For example Light Infantry against itself though the epochs, that is epoch 4 vs epoch 2, epoch 6 vs epoch 4 and so on, then use those units as benchmarks for balancing the other epochs. In the end both techniques had to be used at the same time. A resource value of 4000 per group (or about 48000 per battle) gave fairly good results, but you need a pretty powerful machine to run the tests.

You can put wall towers, outposts, fortresses and anti-air on the unit lists just as easily as regular units, and the battle will run just fine, but since the placement of these static defences are somewhat artificial our experience at Mad Doc, was that the data gathered from these tests were less reliable then plain old play testing.

Finally, there is a known bug that you should be aware of in using the autobalancer. On some machinese the autobalancer may crash the machine at the end of a test run, due to improper release of sprite memory. So if it happens to you try to use the autobalancer on another machine.