That tag has nothing to do with how the AI chooses to USE its siege weapons and everything to do with a final dial percentage modifier to the value of building another siege weapon after all the complex evaluations take place that do utilize the BBAI global defines. At least, it does when evaluating for the sake of building city attack stacks. When implementing that tag, it was implemented after seeing where all the other global values BBAI has in the mix take place, therefore it was made in the code with full visibility of the code. It does nothing to manipulate how siege weapons behave once in play. It ended up being a negative because the tag was made to attempt to help to overcome the problem we were having that NO siege was being trained. Once the REAL problem with siege not being trained was discovered and fixed, we suddenly had a major overflow of siege being built and so far, it appears -20 seems to be the spot where we get pretty much the right balance of siege and other types in the stack so without the tag we still have an overbuild. The -20 means "build 20% less than you would normally think you'd want according to all the complex coding that has come before this final end of the evaluation point". The way the stacks will arrange themselves will be changing fundamentally next version so this tag and others may disappear or take on new meaning. Koshling has suggested, and he's entirely correct in my opinion, that many problems we may be having with city attack stacks not finishing the job of taking the city is coming down to units that cannot finalize the attack themselves becoming the 'stack leaders' guiding the rest of the stack to behave as they believe themselves possible to behave. AKA, the way this is currently arranged allows siege units that have combat limits to become the leaders of the stacks because they are CITY_ATTACK AI just the same as the rest of the units and they end up taking charge of the stack when the leading City attack AI perishes. Therefore, for this and to make it easier and more time efficient for the code, splitting such siege units out into an entirely different unit AI and its own stack that compliments rather than completely melds with the core city attack stack is the step that's going to be necessary to solve these problems.