ai-bookkeeper's tests are mostly hand-written files, so the most useful thing to do with a free run is read files someone else wrote. The MIT-licensed csv2ofx package ships about 26 sample CSVs in its source archive (fetched from PyPI; nothing copied into the repo). Every one was read with readStatement.
Twenty-five read as expected. The UBS-style file did not. Its headings are French (Date de valeur, Débit, Crédit, Solde) and the importer did not know "Date de valeur" as a date column. With no date column it decided the header row was a "preamble", skipped it, and guessed the layout from the cells. The amounts and balances came out right, but the first row had an empty description, and the other rows kept only Description 3 (a street address) instead of the payee in Description 2.
Two small fixes, both pinned by tests rewritten with made-up rows:
Date de valeurandDate valeurcount as date headings (after the booking date names, as Valuta already does).- Numbered
Description 1..4columns are joined in header order, so the line reads "Virement postal ASSOCIATION EXEMPLE RUE DU LAC 1, 1201 GENEVE, CH".
This is still one sample file from a project's tests, not a download from a bank. The next real test is still a fresh export.