Generate a Random TSV
Generate a TSV full of plausible fake data — for testing an importer, filling a demo, benchmarking a parser, or reproducing a bug without attaching real data to the ticket. Choose the number of rows and columns and the column types, and get a file with a header row and consistent shape. Nothing here is real: no value describes any actual person.
How to use
- Set the row and column counts.
- Pick the column types — the mix determines what the generated values look like.
- Tick header row to include column names, or leave it off for a bare data file.
- Click generate, then copy or download.
Use fake data, not a redacted copy of real data
The temptation when filing a bug is to take the real export and delete the sensitive columns. That's how personal data ends up in ticket attachments — deleting the name column doesn't stop a rare combination of other fields identifying someone, and once a file is in a tracker it's indexed, backed up, and hard to recall.
Generated data with the same shape reproduces most parser and importer bugs perfectly well, because those depend on structure rather than content. When the bug genuinely depends on the real values, anonymize the identifying columns first and say in the ticket that you did.
When you need a specific edge case
Random data won't contain the pathological rows that break parsers, so build those deliberately: generate a base file, then use fix ragged rows in reverse (or edit in the grid editor) to make a row short, add a quoted field containing a tab with quote columns, or convert to CRLF with line endings. The sample datasets page also has a deliberately ragged file for exactly this purpose, plus larger real-world files for performance testing.
FAQ
How large a file can it generate?
Large enough to be slow in your browser before it's interesting — tens of thousands of rows is comfortable. For a multi-gigabyte load test, generate with a script instead.
Are the values realistic?
Plausible in shape rather than statistically realistic: names look like names and numbers look like numbers, but distributions are uniform and there are no correlations between columns.
Could a generated name match a real person?
The word lists are small and generic, so coincidences are possible. Don't present generated data as though it describes anyone.
Where do I get real sample data?
Sample datasets — real public-domain TSVs of various sizes, including a deliberately broken one for testing error handling.
Privacy
100% client-side. No upload. See the privacy policy.