CSV Validator and Checker

Read CSV errors before choosing a repair

A useful CSV check distinguishes a missing value from a broken record. Start by confirming the delimiter, decoded characters, and first row in the preview. A semicolon file read as comma-separated text can look like a single-column table; selecting the correct delimiter comes before fixing columns. See the delimiter guide for examples.

Five fixture labels and their diagnostic meaning

The CSV Edge Case Corpus uses the following reference labels. These are dataset labels, not error codes returned verbatim by this tool. The interface groups related findings under readable names, and some corpus checks differ from the interactive report.

DUPLICATE_HEADER → Duplicate headers

id,name,name repeats a column name. The check trims whitespace and ignores letter case, so Name and name also conflict. “Make unique” adds suffixes; verify that the renamed columns still match your destination. Inspect theduplicate-header fixtureor download its CSV.

EMPTY_HEADER → Other checks: empty headers

In id,,score,, two columns have no name. The preview supplies fallback labels but retains the warning until an explicit header repair. Unlike the corpus, the interactive duplicate-name check skips blank names. Review the empty-header fixture.

INCONSISTENT_COLUMNS → Inconsistent rows

Data records are compared with the first non-empty record's field count. A missing final value and an unquoted comma can produce different widths, but they need different fixes. Theshort-and-long-record fixturecontains both cases shown below.

UNCLOSED_QUOTE → Quote errors / Parser messages

A missing or invalid closing quote can absorb later lines into one field. Correct the source quoting; column padding does not repair it. Reproduce this with theunclosed-quote fixture.

TRAILING_DELIMITER → Review the final field

The corpus flags a delimiter at every non-empty physical line's end. This is not a separate issue group in the tool. A final delimiter can represent a legitimate empty value. In thetrailing-delimiter fixture, the extra unnamed header is what needs attention; do not delete all trailing delimiters indiscriminately.

Compare short rows, extra fields, and broken quotes

id,name,score
1,Ada,97
2,Grace
3,Katherine,91,unexpected

Grace's record has two fields; Katherine's has four. “Normalize widths” can append empty trailing fields to short records, but refuses the entire repair when any record is wider than the headers. It does not truncate unexpected. Even for a short record, first confirm the missing value belongs at the end: padding cannot identify a missing middle field. Undo reverses the most recent repair.

id,note
1,"this quote never closes
2,another value

Here, adding a comma will not close the quoted field. Compare it with thevalid multiline CSV, where an embedded newline is enclosed in quotes and a literal quotation mark is doubled.

Confirm headers and destination requirements

Auto detection is a guess. Duplicate labels can make the first row look like data and hide header warnings. If it contains column names, choose “Use as header” and parse again. Choose “Treat as data” only when the file really has no header record.

The validator blocks repaired-file export while column-count, duplicate-header, empty-header, quote, or parser errors remain; the error report stays available. Preview-only files also cannot produce a complete export. A clean structural result still does not validate database types, required fields, unique keys, or foreign keys. Excel can display a value that your database schema rejects. Check those destination rules separately, and use the encoding guidewhen the characters themselves look wrong.