This is a suggestion, not a request.
Preliminary matching in general is source-dependent: it must have a source text to translate. However, finding "hits" (guessed fragment matches) is an internal operation within the same TM and does not need such text in theory.
The problem with "hits" searching is that it takes much time.
The entire process would run much faster if "hits" searching can be carried out separately (like the TM Edit workflow) and the results can be stored as a separate TM for future use.