CSV format repair

Fix the file format without rewriting the values

Choose the bank's original file rather than a copy already saved by a spreadsheet. For files, the tool reads bytes and lets you choose UTF-8, UTF-16LE, UTF-16BE or Windows-1252. A byte order mark supplies evidence for a Unicode encoding. Without one, UTF-8 is only a suggestion, and you must confirm the choice. Changing the encoding always decodes the original bytes again.

Choose comma, semicolon, tab or pipe as the source separator and say whether the first record is a header. Quoted separators and line breaks remain inside their cells; this is not a global search-and-replace. The output uses commas, UTF-8 with a byte order mark, and CRLF record endings. It does not add a sep= row.

Pasted text has already been decoded somewhere else, so this tool cannot inspect its original encoding. If you do not know the encoding, ask the bank or obtain a new native export. Successful decoding alone does not establish that José, a currency symbol or any other character is correct.

Two references that should keep their zeros

The fictional example starts with three semicolon-separated columns: Reference, Name and Amount. Its two records contain 001234;José;-12.50 and 000007;Mina;100.00. After choosing semicolon and reading the table, mark Reference as an identifier. Mark Amount numeric and choose decimal point, grouping comma, with two decimal places.

The protected comma CSV retains 001234, 000007, José and both amount strings exactly. It reports two identifier-risk cells and no formula-protection changes. The amounts add to 87.50 as an arithmetic cross-check, but the tool does not calculate a financial total or infer a currency. With Amount left as text, the protected copy deliberately prefixes -12.50 with an apostrophe instead.

The original and output tables include every source record. Extra quotes needed to write a comma CSV are file syntax, not changes to the cell text.

A repaired CSV still needs a careful Excel import

A CSV does not carry spreadsheet column types. Microsoft documents that Excel can remove leading zeros, truncate numerical text to 15 digits of precision, and interpret certain codes as dates or scientific notation. Quoting a CSV field does not solve that problem. The tool flags digit-only values with leading zeros or more than 15 digits, including surrounding whitespace, and E-notation values such as 123E5. It inspects strings without converting them to JavaScript numbers or trimming the stored cell text. Date-like codes need your review too; they are not detected automatically.

Use Data > From Text/CSV, select UTF-8 and comma, then open the query editor. Set reference and account columns to Text before loading. If prompted, replace the current type conversion. Compare the preview to the original export: a prior numeric-conversion step can already have lost digits. The copyable note lists column positions, not account values. See Microsoft's instructions for keeping leading zeros and large numbers (content date February 10, 2026).

Excel for Microsoft 365 and Excel 2024 on Windows and Mac also have automatic data conversion controls (content date October 15, 2025). Microsoft notes that those controls do not directly govern Power Query imports. Its UTF-8 guidance (content date May 1, 2026) explains the BOM and import paths. A BOM helps identify an encoding; it does not tell Excel which columns are identifiers.

For the separate spreadsheet steps, read why a bank CSV opens in one column and how to preserve bank reference numbers.

What the repair changes, and what it cannot know

The protected copy prefixes formula-like text with an apostrophe and reports the changed-cell count. Explicit numeric columns bypass that protection only for plain signed numbers successfully parsed with your number format and precision. No dates or amounts are normalized. Invalid numeric cells stay text and are listed for review; identifiers are never converted to numbers. The numeric setting is only a protection decision, not a declaration that a spreadsheet will preserve the value.

You can deliberately choose faithful cell text with no formula protection, after acknowledging the risk. Neither download is byte-for-byte identical to the source: encoding, quoting and record endings are rewritten. OWASP's CSV injection guidance warns that spreadsheet applications can execute formula-like cells and that protection may be lost after saving and reopening. No CSV protection is universal. Do not open an untrusted faithful copy in a spreadsheet.

Broken quotes and ragged records block export rather than being guessed, padded or truncated. Replacement characters also block it: a literal replacement character and one introduced by an earlier decoder are indistinguishable here. Blank records, repeated headers, subtotal-like labels and exact duplicates remain visible and are retained. A valid zero is not blank. This is a format repair, so pending rows, dates, accounts and currencies stay unchanged; there is no date-range filtering, pooling or financial calculation.

The limit is one file up to 5 MB (5,000,000 bytes), with decoded text also limited to 5 MB in UTF-8, at most 10,000 data records and 100 columns. Output tables page through at most 200 records. PDF, workbook, binary and UTF-32 inputs are not supported. The repair cannot recover lost zeros or digits, establish the original encoding from plausible text, authenticate a statement, reconcile a ledger or guarantee acceptance by Excel or another importer.

The identifier-risk CSV includes flagged values and source lines, with an apostrophe prefix on formula-like text. Import its identifier column as Text too. The separate repair report records the source name, encoding choice, column settings, retained records, zero exclusions and reported issues. Both can contain sensitive information. Keep them with the original rather than treating a repaired CSV as bank-issued evidence.

Your original stays where it is

Your file and the values you enter are processed in this browser tab. They are not uploaded or sent to analytics. We do not save them between visits. Downloads you choose to save remain on your device.

Clear removes the tool's active input, choices and results and cancels an in-flight file read. It is not a secure memory wipe. Navigating away is not a clearing action because Back can restore uncleared input. Downloads create separate files and never overwrite the bank export. Copying the import note uses your device's clipboard only when you choose it.

The boundaries behind these checks

The quoting rules follow RFC 4180, an informational document from October 2005, extended here to selected source separators. File decoding uses the browser's TextDecoder in fatal mode rather than silently inserting replacements for decoding errors.

We opened the Microsoft, RFC, MDN and OWASP sources on October 17, 2026. Microsoft's content dates are given beside the links; all three pages also exposed updated metadata dated June 25, 2026. No update date was shown in the retrieved MDN or OWASP text. The fictional arithmetic was checked with exact decimals. These downloads were not tested in a live Excel workbook, and no compatibility certification is claimed.

If the original is a PDF instead

First look for a native CSV from the bank, or prepare a small table manually. This tool cannot extract a PDF or repair a workbook. If only PDF statements remain, AnyStatement can convert them into reviewable transaction data linked to source lines. That is a separate upload workflow, not part of this local repair; no file is handed over automatically.