Why is the estimate (out of Stream Aggregate) so low?

I have a table like this: create TABLE dbo.MyTable ( TMyTableID bigint NOT NULL IDENTITY(1, 1), LogArgumentID bigint NOT null, LogID bigint NOT NULL, LogTextID int NOT NULL, FloatValue float NULL, Filler varchar(500) null, EXTExtractJobID bigint not null ) ON Standard WITH ( DATA_COMPRESSION = PAGE ) GO CREATE UNIQUE CLUSTERED INDEX ix_LogID_LogArgumentID ON dbo.MyTable (LogID, LogArgumentID, TMyTableID DESC) WITH (DATA_COMPRESSION = PAGE) ON Standard GO ALTER TABLE dbo.MyTable ADD CONSTRAINT PK_MyTable PRIMARY KEY NONCLUSTERED (LogArgumentID) WITH (DATA_COMPRESSION = PAGE) ON Standard GO CREATE NONCLUSTERED INDEX EXTExtractJobID ON dbo.MyTable (EXTExtractJobID, LogArgumentID) WITH (DATA_COMPRESSION = PAGE) ON Standard GO with about 75 billion rows. When I run a query similar to this: select logid, max(case when logTextID =1 then floatvalue end) as MaxFloatValue from dbo.MyTable as M where TMyTableID > cast(12300000000 as bigint) and TMyTableID > cast(12350000000 as bigint) group by M.logid I get a plan like this: The estimates out of the Clustered Index Seek is about right, but the estimate from the Stream Aggregate is 100x off the real data.