CSV Null Standardizer

About this tool

Open enough real CSV exports and you will find the same column expressing "no data" four different ways: an empty cell here, NA there, #N/A from a spreadsheet, null from a database dump, and a lone - typed by whoever maintained the sheet by hand. Every downstream step then needs its own list of exceptions — a na_values= argument, a CASE WHEN in the load script, a find-and-replace before the import.

This tool collapses all of that into one rule. Paste the table, pick the single representation you want, and every cell that means "missing" is rewritten to it. Everything else is left exactly as it was: real values are copied byte-for-byte (padding included), the header row is never touched, quoted fields keep their quoting, ragged rows keep their length, and the field separator round-trips unchanged.

It standardizes the token, not the data. Nothing is guessed, averaged, or filled in — if you need means, medians, or nearest-neighbour fills, that is imputation and a different job. If you first want to know how much is missing and where, profile the table before you normalize it.

Worked example

Input (delimiter comma, replace with NULL, defaults everywhere else):

id,score,notes
1,42,ok
2,NA,
3,null,-
4,n/a,NaN

Output:

id,score,notes
1,42,ok
2,NULL,NULL
3,NULL,NULL
4,NULL,NULL

Note what did not change: the header stays id,score,notes even though a header cell could match a token, and 42/ok are untouched. n/a matched N/A because matching is case-insensitive by default, and the empty cell on row 2 matched because blank cells count as missing.

Controls

• Delimiter — comma (default), tab, semicolon, pipe, any single character, or auto to sniff the separator from the first line. The output always uses the same separator as the input.
• Tokens that count as missing — the comma-separated vocabulary, pre-filled with NA, N/A, N.A., #N/A, #N/A N/A, #NA, NULL, NIL, NaN, None, <NA>, -, --, and ?. Edit it in place: add a project sentinel like -999, or delete entries you use as real values. Clear it entirely to standardize only blank cells.
• Replace missing cells with — the one representation you want. Leave it blank for an empty cell; type NULL, NA, NaN, or \N (the sentinel a Postgres COPY … WITH (FORMAT csv, NULL '\N') load expects).
• Also standardize blank / whitespace-only cells — on by default. Turn it off to convert the listed tokens and leave already-empty cells empty.
• Match tokens case-sensitively — off by default, so a single NULL entry also catches null and Null. Turn it on when case carries meaning.
• Ignore whitespace around a cell when matching — on by default, so NA is recognised. This only affects matching; a cell that is not missing keeps its padding.
• First row is a header — on by default. The header is copied through untouched and supplies the names used by the column filter.
• Only these columns — comma-separated column names (needs a header) or 1-based positions, e.g. score,notes or 2,4. Blank means every column, and you never have to split the file up to normalize one field.
• Output quoting — minimal quotes only fields that require it, always quotes every field including the replacement token, never writes bare fields. Keep minimal when your replacement is a loader sentinel like \N, because a quoted "\N" is read as the literal two-character string.

Limits and edge cases

The table is capped at 5,000,000 bytes. A row shorter than the header stays short — missing trailing cells are not invented, because writing a value into a column the row never had would change its shape. Because the token list itself is comma-separated, a token that literally contains a comma cannot be expressed. Choosing never for output quoting can produce ambiguous CSV when a value contains the separator or a newline, so keep minimal unless a downstream reader demands otherwise. Output rows are terminated with a single newline (\n), and the result always ends with one. Everything runs locally in your browser — the table is never uploaded.

FAQ