We should perform some simplifications/dominance tests for conditional effects
in the translator. In the example of msg1958, almost all effects are completely
redundant and should be dropped (to reduce task representation and hence search
overhead, and also to avoid confusing the relevance analysis, heuristics etc.).

Task representation size before/after in something like the BAMS example from
that task would be a good measure of effectiveness for such reductions, assuming
that we can fix the issues that currently prevent us from handling the BAMS tasks.