From: Puget Sound, Wash, US
Ran Turn 1-16 again to gather what information might be useful. This is really "black box analysis" without any tools or insight of the internals. Nonetheless kept Task Manager/Performance open to track available memory as well as Performance Monitor to track private bytes consumption. What I found:
A. Consumed private bytes dropped back to ~300-400MB at start of each player Turn. Growth was slow but steady during Turn.
B. Although consumed private bytes looked like (on the System Monitor graph) to be approaching the ceiling many times, Win10 increased allocation each time. The memory consumption chart here can be deceiving since the vertical axis is apparently computed automatically, and as the size of the allocated private bytes is increased, the entire graph is immediately rescaled. See, e.g. the Allied progression in Turn 16 from 809MB to 827MB, from 966 to 989MB, Perhaps significant is the available memory during Turn 16 only decreased from 11GB to 10.3GB at program crash. Not sure memory leak can be confirmed as cause of the crash.
C. Since 3 of us have now had crashes with the identical scenario, chances of this being caused by a specific PC configuration, considering how many variations are possible, must be most unlikely.
D. When I saved this game run for analysis, I set PO random seed to -0, to eliminate any stochastic variability during the analysis. TOAW IV blew up 7 of 7 times identically before I started the thread. Nonetheless, last night while gathering my data, TOAW completed Turn 16 and continued onwards. [I was tempted to add an exclamation mark to the prior sentence or even one of those omicons, but recovered my senses.] Nonetheless, either this software is subject to the Heisenberg Uncertainty Principle, or we don't fully understand the system design or I've had too much Cabernet, but when software produces different results while repeating then the end of the world is not far off.