OK, now that I see the data, I think you should have 2 fact tables, because indeed they are at two different grains.

Recovery is at the treaty level, while total paid is at the claim level.

Also the fact that some claims don't have recoveries makes me think in the direction of two fact tables.

But that's OK. When you analyse the data later on in Excel, you drag the total_paid measure next to the pivot and claims to the rows. When you add the recovery, it will automatically aggregate the data to the claims level.