Typed CSV Specification


The purpose of the typed csv specification is to build on the common csv (comma separated value) specification with a standard unambigious format.

The key issues with most csv files at present are:

  • Character encodings are not defined
  • The data has no type attributes, so 1 could be considered an integer, floating point number, or a string. File loaders often have to guess the type, and modelling software needs to be written to explicitly cast types.


An example of typed csv:

# comment lines
@ author: name@domain.com
@ write_date: 2020_03_50

Key Rules

  • Typed CSV files are always encoded as UTF-8.
  • All header names, header types and data rows must have the same length.
  • The first character in the line determines the purpose of the line.
  • The first character must be followed by a comma if it is !,? or *.
  • the separator defaults to comma ,.
  • @ can have a space between it and the metadata key.
  • All meta data must be above the header row
  • Rows must be in the following order: meta > header > types > data
  • Comments can be placed anywhere and will be ignored
  • Rows must end with a new line character \n
Character Purpose Notes
# comment ignored
@ metadata for storing individual values, key and value are separated by a colon :
! header names of the columns in the data
? data types the type of data in the column
* data row a row of data values

Data Types

  • int: Integer (..., -3, -2, -1, 0, 1, 2, 3, ...)
  • float: Floating Point Number (13523.524), only decimal notation is supported
  • str: String/Text
  • bool: Boolean, using the following (case insensitive)
    • T, 1, Y, true evaluate to True
    • F, 0, N, false, evaluate to False
  • dec: Decimal for dealing with currency
  • yyyy_mm_dd: date
  • hh_mm_ss: time
  • u_: a user defined type

Data for number types (int, float, dec) can optionally have underscore characters (_) as thousand separators. These will be ignored on processing.


A single whitespace can be added before the @. Thus the following are valid metadata and mean the same. Any trailing whitespace will be considered part of the key.

Any characters after the colon (:) will be considered part of the value, up until the new line.

The type of the value is not documented.

@ key:value

Reserved keys

Reserved keys are optional, but can enhance the stability of the data

  • @length: The number of rows of data, as an integer, if this is supplied and the values does not match the number of data rows, an error will occur.
  • @separator: The separator character(s)
  • @md5-checksum: An 128bit MD5 checksum, presented as 32 hexadecimal digits (0-9a-f), this hash is based on a string containing header, types and data in the order they appear. Metadata and comments are ignored. See https://en.wikipedia.org/wiki/MD5 for details of MD5.

Custom Separator

If the data is likely to contain commas, a custom separator can be specified by @separator metadata item. The separator can consist of one or more characters.


Application specific types

A type beginning with u_ is left to the application to process.

An example would be u_yyyy_mm which would store the year and month.

Caution should be taken to avoid name conflicts.