Well, that's what he said. But my point's still the same - you can't code the AI to make calls on strategy. That's telepathy. So don't code something that makes calls on an attack "failing"... simply record each attack and the outcome of each. You wouldn't even need to display each attack; if a country is attacked sequentially you could aggregate the results. I'm agreeing with the basic premise.
Although now that I think about it, my original scenario might be solveable by putting an "under attack" flag, or whatever, on a defending country. The flag lasts the round, and you could then put up a report saying, "Country X resisted an attack from Y. X lost 1 and Y lost 3." And have one of those for each resisted attack. So the reporting isn't related to attacks made, which have a human element in them, but defenses, which are passive.
[EDIT] Blah, I'm making this too complicated. Look, all I'm saying is, under language currently proposed, in the example scenario I gave you would resolve two failed attacks and one capture, when what was actually happening was a single successful two-pronged attack. So you can't talk about "failure," only record the occurence of attacks and the armies won/lost.
Actually, the scenario is a bit flawed. Say Country Z lost some troops in the second prong of the attack. You could resolve the whole thing after the fact to a simple successful attack from Y to X, but then you wonder where the troops from Country Z went. As it stands now, you make an intelligent guess. The original proposal doesn't want that uncertainty, as I understand it. So you'd have to record each attack.
Another issue is that this version of Risk has occassionally a real-time element to it. The current language proposes reporting of halted attacks at the end of a turn, which would be fine if we were moving sequentially. But if two or more players are moving at once you'd get a log with a very muddled chronological line. You might be able to resolve this by time stamping events and logging them after the fact, but do you stamp the beginning of the attack or the end?
I still say better to log all attacks. The log would be longer but it would be simpler to understand.