Yesterday I had the case where the AI positioned units in a suicidal position.
The incident was related to the "new" spotter-feature. The problem was that the entire logic for the spotter was based on the position of where my unit had last been spotted and not where it could have gone in the meantime. In this case I accidentally found this flaw but after seeing in what situation that can happen, I realized how exploitable it is.
With the recent fix of line-of-sight-checking and another fix related to caching and auto-play, which made the AI play worse in auto-play than when playing against the player it does a much better job when playing without spotters.
I think I can remove the spotter-mechanic without regression. In one of my benchmarks there even was an improvement to losing only 1 sectoid to friendly-fire and none to X-Com.
I have to do some play-testing of what it feels like to play against then and whether this creates new weaknesses.
