When to repair a CSV versus re-export from the source

Tell structural delimiter and encoding problems from values that are already rewritten.

Repair changes structure; re-export restores meaning

CSV check and repair tools are good at delimiters, encodings, quoting and uneven widths. They are not a second source system. When product codes, prices or names are already wrong in the bytes you hold, another local pass cannot invent the missing truth. The useful question is therefore not “which button looks like fix,” but “is the injury structural, or has the value itself already been rewritten?”

Structural injuries leave the original characters intact while hiding them: a semicolon file opened as commas, a UTF-8 name decoded as Windows-1252, a quoted note that spans two physical lines. Value injuries replace those characters: 00127 saved as 127, a seventeen-digit tracking ID shortened to an exponent, Café saved as Café after a double encode. Structure can be re-parsed. Meaning needs the exporter that still knows the record.

product_id;warehouse;note
00127;North;"Shelf A; cold"
00904;South;Café tray

This teaching extract is semicolon-delimited with a quoted note that contains a semicolon. The product codes still have their leading zeros. Nothing here is permanently lost. Choosing Semicolon in CSV check & repair, previewing three columns, and exporting UTF-8 with commas is a repair. Double-clicking the file in Excel, watching the zeros disappear, and saving over the CSV is not a repair—it is a new export of damaged values.

A decision sequence you can repeat

1. Keep the original download. Copy it before any “Save As,” find-and-replace, or second tool. The first overwrite is often the moment recovery becomes impossible.
2. Ask whether the characters you care about are still present when you open the file in a plain text editor or in TableMender’s preview with an explicit delimiter and encoding. If 00127, Café and the full tracking ID still appear, the problem is likely structural.
3. If the preview already shows 127, Café, 1.23457E+16, or replacement characters (�), stop editing. Return to the system that produced the extract and request a fresh export with text-preserving settings.
4. When structure is the only issue, fix one setting at a time: delimiter, then encoding, then quoting. Preview after each change. Do not combine a global find-and-replace with an encoding guess.
5. After a successful structural repair, convert to a text workbook with CSV to Excel if the next consumer is Excel. A repaired CSV can still lose zeros on the next double-click open.

Inspect structure first →

Repairable cases versus re-export cases

What “re-export” should ask for

When you go back to the source system, be specific. Ask for the original field as text, not as a spreadsheet cell that already applied General format. Prefer a direct CSV or UTF-8 download from the application over “open in Excel, then Save As CSV.” If the system only offers Excel, ask for identifier columns formatted as text before export, or for a report that quotes those fields. Record the delimiter and encoding the vendor claims to use; regional list separators are a common reason a “CSV” arrives with semicolons.

If the only available file already lost zeros, say so in the ticket. Support teams often re-send the same damaged extract. Attach a one-line sample that shows 127 versus the expected 00127, without customer names or secrets. Synthetic examples are enough.

Edge cases and boundary cases

A file can be half repairable. Delimiters may be correct while one column is already scientific notation. Repair the structure, then stop. Do not “fix” the exponent by padding zeros you cannot verify. Keep the repaired file and the source ticket side by side until a fresh export arrives.

Auto delimiter detection is a hint, not a certificate. A score that prefers Semicolon on a locale price file can still be wrong for a mixed extract. Choose the delimiter yourself when quoted fields look wrong. The delimiter and locale guide shows how a decimal comma is not a field separator.

Cleaning is not recovery. CSV clean can remove empty records or duplicates after you decide what a key means. It will not rebuild truncated IDs. Running deduplication on a file that already lost warehouse distinctions can discard the only remaining quantity. Preview with every rule off first.

Formula-like cells that begin with =, + or @ are a separate risk on CSV reimport. TableMender’s formula-protection option prefixes an apostrophe on export when you enable it; that is an intentional character change, not a silent repair. Use it only when the receiving spreadsheet would otherwise execute the field.

Size limits matter at the boundary. This site caps input at 5 MiB, 20,000 records including the header, 256 columns and 32,767 characters per field. A failure at those limits is not guidance about your business key. Split a large extract in the source system rather than deleting columns at random to squeeze under the cap.

Multiple encodings in one file—for example a UTF-8 header glued to a legacy body—are outside what a single encoding menu can honestly claim. Treat that as a source defect and ask for one encoding throughout.

Where the other tools fit

Use CSV repair when the preview can still show the characters you need. Use CSV to Excel when those characters must survive an Excel handoff. Use CSV clean only after the fields mean what you think they mean. Guides for leading zeros, dates, scientific notation and mojibake exist because those failures look like “broken CSV” while the bytes are already settled.

If Excel’s own Save As path created the damage, read Excel Save As CSV pitfalls next. That page covers round-trip habits that turn a good export into a second, worse export.

Before and after: a structural repair beside a value injury

Use two downloads. They are not before-and-after versions of one export.

The semicolon inside Shelf A; cold is content. A global replace of semicolons would split that note. Parse it. The rewritten symptom file has nothing left to parse back into the missing zeros.

Step-by-step troubleshooting

1. Auto delimiter looks wrong. Choose Semicolon, Tab or Pipe and preview again. Detection is a hint. Download stays blocked while the separator is uncertain.
2. Preview shows the characters you need, but Excel will be next. Repair the structure, then convert with text columns. A repaired CSV can still lose zeros on a later double-click open.
3. Preview already shows 127, Café, 1.23457E+16 or �. Do not keep editing. The value is rewritten or the bytes were replaced. Ask for a new export. See Fix encoding only when Auto rejects the original bytes and a known encoding still shows the right letters.
4. One column is fine and another is an exponent. Repair what is structural. Leave the exponent as text. Do not pad zeros you cannot verify.
5. You were about to dedupe before this check. Open CSV clean only after the fields mean what you think. A truncated id is a bad duplicate key.

Honest limits

TableMender does not connect to your ERP, warehouse or CRM. It cannot request a re-export for you, merge two disagreeing systems, or certify that a repaired file matches a destination schema. It does not recover replacement characters, truncated digits or date serials. When the original characters are gone, the honest next step is the source system—not another local transformation.

← All guides