>Because these tables come from a mainframe, the unique clustered index comes from NEWID(). I'm still looking for a real key.

Very often who writes the software it doesn't plan it with the fundamental principle:
the coherence and the invariance of the rules.
Probably MSSQL it has different rules in to assign the costs of batch and sp.

However

-- we try to deceive it
   
SELECT FLAIRTR2.* 
FROM FLAIRData.dbo.FlairTR2 
WHERE pk IN 
 (select FLAIRTR2.pk   -- pk has to be or the key of the clustered index or it has to belong to the index that we want that SQL uses
   FROM FLAIRData.dbo.FlairTR2 FLAIRTR2
    JOIN #Temp TMP
ON FLAIRTR2.FiscalYear = TMP.FY 
AND FLAIRTR2.Account = TMP.Account 
AND FLAIRTR2.CFI = TMP.CFI
AND FLAIRTR2.SWDN = TMP.SWDN
AND FLAIRTR2.PrimaryDoc7 = TMP.Prima)

-- we force the master JOIN table to #Temp 
SELECT FLAIRTR2.* 
FROM #Temp TMP
  JOIN FLAIRData.dbo.FlairTR2 FLAIRTR2
ON FLAIRTR2.FiscalYear = TMP.FY 
AND FLAIRTR2.Account = TMP.Account 
AND FLAIRTR2.CFI = TMP.CFI
AND FLAIRTR2.SWDN = TMP.SWDN
AND FLAIRTR2.PrimaryDoc7 = TMP.Prima
OPTION ( FORCE ORDER )


-- we force the FLAIRTR2 index
SELECT FLAIRTR2.* 
FROM #Temp TMP
  JOIN FLAIRData.dbo.FlairTR2 FLAIRTR2 WITH ( INDEX (indexname) )
ON FLAIRTR2.FiscalYear = TMP.FY 
AND FLAIRTR2.Account = TMP.Account 
AND FLAIRTR2.CFI = TMP.CFI
AND FLAIRTR2.SWDN = TMP.SWDN
AND FLAIRTR2.PrimaryDoc7 = TMP.Prima

The force solutions is static,
and therefore it is a desperate solution.