Time series collections with fixed buckets (where the rounding is the same as the max span) allow for some optimizations to be applied to the whole-bucket and event filters, as well as optimizations to the $group stage.

However, if there are extended range values, we can't apply these optimizations safely, since it could produce wrong answers.

When we plan queries on mongos, we don't know which shards have extended range values and which do not. But we will apply fixed bucket optimizations if there were no extended range values on the DB primary. If mongos routes this plan to a shard with extended range values, we could get wrong answers.