AI Bookkeeper's bank readers were tested on files we wrote ourselves. That is a weak check, so we keep pulling in files other people made. This time: the test data of beangulp, the importer framework from the Beancount project (read only, nothing copied).
What went in, and what came out:
- A Chase-layout checking CSV with a trailing comma in the header: 6 rows, amounts, dates and running balances right.
- Two OFX downloads, 12 and 27 transactions, one from a Bank of America card: every row read, closing balance matched the file.
- A small brokerage CSV: 4 rows read.
- An OFX file holding only investment trades: nothing booked, with a message saying why, which is the right answer for a cash ledger.
We found no fault, and we are reporting that rather than hunting for one. The files are another project's fixtures, not fresh downloads from a bank, so the real-export gap on the roadmap stays open.