Castor is an unilingual phrases database. Its role is not to be used during a translation but before :

1. Segment your documents and send them to the database
2. You will receive a report saying whenever some documents have been found with a significant number of similar segments

Actually the tool is based on exact similarity, meaning that you find only phrases which exist in multiple documents, i.e. 100 % matches (except for tags, which are not retrieved). Such segments are auto-inserted in most CAT tools. In the future fuzzy matches may also be considered, then the scoring rules may change.

Do not forget to delete documents which have already been translated : those documents should already be present in a long-term translation memory!