Our existing code to track collscan/ixscan usage (CollectionIndexUsageTracker and CurOp DebugInfo) assumes that any access path (STAGE_IXSCAN, STAGE_COLLSCAN, etc) present in the QuerySolution tree is applied to the main collection of a query. With the plans produced by join re-ordering, this is no longer true. 
 
This bad assumption can result in query failures and incorrect stats tracking.