Retryable writes work by verifying if a single CRUD operation has already been applied. This is done by checking the statement id in the oplog via TransactionParticipant::checkStatementExecutedNoOplogEntryFetch in write_commands.cpp.

For FLE2, we just need to ensure the statement id from the user transaction is propagated to the first write (typically either ESC or ECC). If a CRUD operation is retried, you can check retriedStmtIds in the reply.