The .csv file.
- Specification
- None. RFC 4180 documents a convention.
- Delimiter
- Comma, semicolon, tab or pipe, depending on who wrote it
- Encoding
- Undeclared. UTF-8, UTF-16 and Windows-1252 all occur.
- Types
- None. Everything is text until something guesses.
- Size ceiling
- None in the format itself
- Read here as
- Direct. The delimiter is worked out from the file; the encoding is assumed to be UTF-8.
- Status
- Ready
- Fidelity
- Read directly
- Largest file
- 25 MB free, 250 MB once a project is bought
What it actually is
Delimiter, quoting and encoding detected rather than assumed.
A CSV is a text file where something separates the values and something else, sometimes, quotes them. That is the whole format. RFC 4180 describes one common dialect and explicitly does not claim to be a standard anybody implements, which is why two programs can both be correct and disagree about the same file.
Everything that makes a data format safe is absent: there are no types, so a leading zero is indistinguishable from a number; there is no declared encoding, so the bytes have to be guessed, and there is no schema, so a column that changes meaning halfway down the file is perfectly legal.
It survives anyway, because every system on earth can produce one. In practice a CSV is the lowest common denominator between two things that were never designed to talk, and the damage happens at that boundary rather than inside either program.
Upload one
The analysis runs before there is anything to pay for.
You see what it found, and the evidence behind it, first.
No account needed to start. You only pay when you like what you see.
What goes wrong, and what is done about it
4 named failures, each with the repair.
Excel writes semicolons in half of Europe
Excel's CSV export uses the list separator from the operating system's regional settings. On a machine set to most European locales that is a semicolon, because the comma is the decimal mark. The file is still called .csv and opens as one column everywhere else.
What is done: Comma, semicolon, tab and pipe are tried in that order, and the first one that yields more than a single column is used. A semicolon file needs no repair before it is sent.
Leading zeros are gone before you open it
A postcode, a product code or an account number beginning with zero is a number as far as any type-guessing importer is concerned. 01234 becomes 1234, and the join against another file quietly matches nothing.
What is done: Columns whose values are fixed-width and zero-padded are read as text, so the code that arrives is the code that was written rather than a number that used to be one.
A quoted newline turns one row into three
An address field or a free-text comment containing a line break is legal inside quotes. A parser that splits on newlines first, which many hand-rolled ones do, produces extra rows with columns shifted sideways, and the row count is the only clue.
What is done: Quoting is honoured, so an embedded newline stays inside its field. A row that still comes out wider than the header is truncated to the header's width rather than dropped.
Encoding damage that only shows up in names
A file written as Windows-1252 and read as UTF-8 turns é into é. It affects text columns only, so every number reconciles perfectly and the file looks fine until somebody searches for a customer.
What is done: The file is read as UTF-8, including the byte order mark Excel writes in front of it. Nothing detects any other encoding, so a Windows-1252 file has to be re-saved as UTF-8 before it is uploaded.
If you have the choice, send something else
If the source can produce Parquet, take it instead. It carries real types, so leading zeros, dates and long identifiers survive, and it is typically a fifth of the size.
Questions people ask
About this format, not about the product.
My CSV opens as one column in Excel. Is it broken?
Almost certainly not. It is delimited with something your copy of Excel is not expecting, usually a semicolon or a tab. Upload it as it is: comma, semicolon, tab and pipe are each tried, so the file does not need fixing first.
How large a CSV can I upload?
25 MB without a purchase, and 250 MB once a project is bought. Rows are read columnwise rather than loaded into a chat window, so a wide file with a few thousand rows and a narrow one with two million behave the same way.
Can I upload several CSVs at once?
Yes. Every file is read and converted, and each one is downloadable afterwards. The analysis itself runs on a single table, the largest one, so nothing is joined across two files. Where a second file looks related to the first, that candidate relationship is reported with a confidence, and no figure is ever computed across both.
Will it change my file?
Never. Repairs are applied to a new copy and the original stays exactly as uploaded. Both are downloadable, and every change is listed with a before and after.
Keep reading
The jobs people do with this file, and the formats beside it.