To add focus to the discussion, I'll offer up my recent PBEM experience with the wonderfully capable FelixZ. A force under Beauregard was attacked in Fredrecksburg by Grant (coming from Culpeper with 1 corps, while the rest of the Union force remained in Culpeper). Beauregard decides to withdraw after a few rounds, and retreats to Culpeper....a region with the maximum number of enemy troops in which I have 0 MC. Upon arriving in Culpeper, he switches to attack mode (because I have no MC in the region), looses of course, and retreats back to Fredericksburg! He then is trapped in a loop for the rest of the resolving turn, bouncing between those two regions. For your enjoyment, I am including the .hst and replay file. (you will note that in the new turn, Beauregard STILL wants to retreat into Grant's main body).
Now I know why Beauregard did what he did, a quick at the retreat logic makes it pretty clear:
// ********************************************************
// ***** CONTROL & RETREAT *****
// ********************************************************
ctlContested = 5 // Minimum control gained upon entering a region (if not passive)
ctlAllowRetreat = 0 // Minimum control to have in a region to allow a retreat into it
ctlRetreatAdjCity = 5 // Interest in retreating toward a region with a city (per level)
ctlRetreatAdjFort = 30 // Interest in retreating toward a region with a fort (per level)
ctlRetreatAdjDepot = 25 // Interest in retreating toward a region with a depot (per level)
ctlRetreatLandLink = 10 // Interest in retreating toward a region, value per land link
ctlRetreatPrevSubSpaceCoeffH = 250 // Coefficient applied to the interest if the region is the one where we are coming from
ctlNoBeachHead = 10 // Minimum control to have so that a region is not a beach/riverhead if you have to cross a body of water before attacking
// Added September 22, 2012
cltRetPenaltyPerNmySU = 4 // retreat penalty (in interest pts) for each nmy SU in retreating region
In the basic retreat calculation the presence of enemy forces IS NOT A FACTOR. Military control is the proxy for enemy forces, but you can see that as coded originally, this value is set to 0. So a unit is allowed to retreat into a province in which its side has no Military Control. After that, the draw is determined by what is in the province. In this case Culpeper was the only town with a depot next to Fredericksburg. When he lost his battle in Culpeper, he could have gone south to Charlottesville, 100% friendly MC with a depot in town, but he went back to the enemy held Fredricksburg because of the whopping 250 coefficient that tries to get a stack to go back to where it comes from.
So the result is a retreat loop. A defending force is dislodged and retreats into an enemy held town/fort/depot province (which is logically the last place they should want to go), retreats back to where it just came from (now an enemy held province) looses, and retreats "back" to the previous province.
Clearly the code is designed for the "Minimum control to have in a region to allow a retreat into it" value to prevent this kind of thing from happening. A certain % of MC is needed for a region to be a valid retreat destination, then the retreat algorithm kicks in to direct forces to friendly towns, forts, and depots. (Frankly, it is quite elegant.) But with with the minimum control set to 0%, stacks can't distinguish between friendly and enemy held depots.
Why is minimum MC set to 0? According to the game manual, it shouldn't be. Page 63 of the manual (describing the effects of Military Control):
Finally, you cannot retreat from battle into completely hostile regions (i.e. less than 5% Military Control). [color="#FF0000"]Note: Engaging in a battle deep within enemy territory is a very risky proposition, as losing will result in complete destruction for lack of a retreat path! [/color]
According to Pocus the decision was made quite along time ago during the patching process of ACW1, based on feedback from beta testers. Apparently, the justification was for historical flavor. Encirclements were "difficult" to achieve and the 5% rule made it too simple to do. That would explain why a new coding was added in September of 2012, that adds a retreat penalty for each enemy sub-unit in a possible retreat province. Clearly though this isn't enough, because Beauregard still went to the province with the most enemy units.
So to sum up, retreats in ACWII need to be improved, because retreating forces all to often decided to retreat into strongly held enemy territory. (Which never happened historically, and makes no sense logically).
There seem to be two options. Either raise the minimum military control necessary for a retreat to 5% (to the level stated in the manual, and what I suspect the AGE engine default level normally is) OR change the retreat coefficient values so that retreating AWAY from enemy units becomes THE major factor in in choosing a retreat destination.
I would argue for the minimum MC value to be re-established, because it is simple, quick, and what the thing was designed for. Any other solution requires re-writing and testing of the whole retreat logic in a way that would only be applicable to this title.