# convenient, but we don't have access to the full transaction

    # entry objects. Feels a bit strange that these approaches are so

    # disconnected.

    #

    # beancount.query.query_compile.compile() and

    # beancount.query.query_execute.filter_entries() look useful in this respect,

    # but I'm not clear on how to use compile(). An example would help.

    entries, _, options = loader.load_file(args.beancount_file)

    # String concatenation looks bad, but there's no SQL injection possible here

    # because BQL can't write back to the Beancount files. I hope!

    query = f"""

        SELECT filename,

        META("lineno") AS line,

        META("bank-statement") AS bank_statement,

        date,

        number(cost(position)),

        payee,

        ENTRY_META("entity") as entity,

        ANY_META("check-id") as check_id,

        narration

        WHERE account = "{args.account}"

            AND date >= {begin_date}

            AND date <= {end_date}"""

    _, result_rows = run_query(entries, options, query)

    # Apply two passes of matching, one for standard matches and one

    # for subset matches.

    (

        matches,

        remaining_statement_trans,

        remaining_books_trans,

    ) = match_statement_and_books(statement_trans, books_trans)

    subset_matches, remaining_statement_trans, remaining_books_trans = subset_match(

        remaining_statement_trans, remaining_books_trans

    )

    matches.extend(subset_matches)

    # Add the remaining unmatched to make one big list of matches, successful or not.

    unmatched = process_unmatched(remaining_statement_trans, remaining_books_trans)

    matches.extend(unmatched)

    # Print out results of our matching.

    print(

        format_output(

            matches,

            begin_date,

            end_date,

            args.csv_statement,