We've had multiple bugs recently about handling extended range data in time-series collections. It would be helpful to document how the buckets are different when the data is extended range and what engineers should consider when developing query rewrites for extended range data.