Table Schema

Author(s)	Paul Walsh, Rufus Pollock
Profile	table-schema.json

A simple format to declare a schema for tabular data. The schema is designed to be expressible in JSON

Language

The key words MUST, MUST NOT, REQUIRED, SHALL, SHALL NOT, SHOULD, SHOULD NOT, RECOMMENDED, MAY, and OPTIONAL in this document are to be interpreted as described in RFC 2119

Introduction

Table Schema is a simple language- and implementation-agnostic way to declare a schema for tabular data. Table Schema is well suited for use cases around handling and validating tabular data in text formats such as CSV, but its utility extends well beyond this core usage, towards a range of applications where data benefits from a portable schema format.

Concepts

Tabular data

Tabular data consists of a set of rows. Each row has a set of fields (columns). We usually expect that each row has the same set of fields and thus we can talk about the fields for the table as a whole.

In case of tables in spreadsheets or CSV files we often interpret the first row as a header row, giving the names of the fields. By contrast, in other situations, e.g. tables in SQL databases, the field names are explicitly designated.

To illustrate, here’s a classic spreadsheet table:

field     field
  |         |
  |         |
  V         V

 A     |    B    |    C    |    D      <--- Row (Header)
 ------------------------------------
 valA  |   valB  |  valC   |   valD    <--- Row
 ...

In JSON, a table would be:

[
  { "A": value, "B": value, ... },
  { "A": value, "B": value, ... },
  ...
]

Physical and logical representation

In order to talk about the representation and processing of tabular data from text-based sources, it is useful to introduce the concepts of the physical and the logical representation of data.

The physical representation of data refers to the representation of data as text on disk, for example, in a CSV or JSON file. This representation can have some type information (JSON, where the primitive types that JSON supports can be used) or not (CSV, where all data is represented in string form).

The logical representation of data refers to the “ideal” representation of the data in terms of primitive types, data structures, and relations, all as defined by the specification. We could say that the specification is about the logical representation of data, as well as about ways in which to handle conversion of a physical representation to a logical one.

In this document, we’ll explicitly refer to either the physical or logical representation in places where it prevents ambiguity for those engaging with the specification, especially implementors.

For example, constraints SHOULD be tested on the logical representation of data, whereas a property like missingValues applies to the physical representation of the data.

Descriptor

A Table Schema is represented by a descriptor. The descriptor MUST be a JSON object (JSON is defined in RFC 4627).

The descriptor MAY have the additional properties set out below and MAY contain any number of other properties not defined in this specification.

The following is an illustration of this structure:

{
  // fields is an ordered list of field descriptors
  // one for each field (column) in the table
  "fields": [
    // a field-descriptor
    {
      "name": "name of field (e.g. column name)",
      "title": "A nicer human readable label or title for the field",
      "type": "A string specifying the type",
      "format": "A string specifying a format",
      "example": "An example value for the field",
      "description": "A description for the field"
      ...
    },
    ... more field descriptors
  ],
  // (optional) specification of missing values
  "missingValues": [ ... ],
  // (optional) specification of the primary key
  "primaryKey": ...
  // (optional) specification of the foreign keys
  "foreignKeys": ...
}

Properties

fields

A Table Schema descriptor MUST contain a property fields. fields MUST be an array where each entry in the array is a field descriptor as defined below.

The way Table Schema fields are mapped onto the data source fields are defined by the fieldsMatch property. By default, the most strict approach is applied, i.e. fields in the data source MUST completely match the elements in the fields array, both in number and order. Using different options below, a data producer can relax requirements for the data source.

fieldsMatch

A Table Schema descriptor MAY contain a property fieldsMatch that MUST be a string with the following possible values and the exact value by default:

• exact (default): The data source MUST have exactly the same fields as defined in the fields array. Fields MUST be mapped by their order.
• equal: The data source MUST have exactly the same fields as defined in the fields array. Fields MUST be mapped by their names.
• subset: The data source MUST have all the fields defined in the fields array, but MAY have more. Fields MUST be mapped by their names.
• superset: The data source MUST only have fields defined in the fields array, but MAY have fewer. Fields MUST be mapped by their names.
• partial: The data source MUST have at least one field defined in the fields array. Fields MUST be mapped by their names.

Field Properties

A field descriptor MUST be a JSON object that describes a single field. The descriptor provides additional human-readable documentation for a field, as well as additional information that can be used to validate the field or create a user interface for data entry.

Here is an illustration:

{
  "name": "name of field (e.g. column name)",
  "title": "A nicer human readable label or title for the field",
  "type": "A string specifying the type",
  "format": "A string specifying a format",
  "example": "An example value for the field",
  "description": "A description for the field",
  "constraints": {
      // a constraints-descriptor
  }
}

The field descriptor object MAY contain any number of other properties. Some specific properties are defined below. Of these, only the name property is REQUIRED.

name

The field descriptor MUST contain a name property and it MUST be unique amongst other field names in this Table Schema. This property SHOULD correspond to the name of a column in the data file if it has a name.

Backward Compatibility

If the name properties are not unique amongst a Table Schema a data consumer MUST NOT interpret it as an invalid descriptor as duplicate name properties were allowed in the v1.0 of the specification.

title

A human readable label or title for the field

description

A description for this field e.g. “The recipient of the funds”

example

An example value for the field

missingValues

A list of missing values for this field as per Missing Values definition. If this property is defined, it takes precedence over the schema-level property and completely replaces it for the field without combining the values.

For example, for the Table Schema below:

"fields": [
  {
    "name": "column1"
  },
  {
    "name": "column2",
    "missingValues": ["-"]
  }
],
"missingValues": ["", "NA"]

A data consumer MUST:

• interpret "" and NA as missing values for column1
• interpret only - as a missing value for column2

Types and Formats

type and format properties are used to give The type of the field (string, number etc) - see below for more detail. If type is not provided a consumer SHOULD assume a type of “string”.

A field’s type property is a string indicating the type of this field.

A field’s format property is a string, indicating a format for the field type.

Both type and format are optional: in a field descriptor, the absence of a type property indicates that the field is of the type “string”, and the absence of a format property indicates that the field’s type format is “default”.

Types are based on the type set of json-schema with some additions and minor modifications (cf other type lists include those in Elasticsearch types).

The type list with associated formats and other related properties is as follows.

string

The field contains strings, that is, sequences of characters.

format:

• default: any valid string.
• email: A valid email address.
• uri: A valid URI.
• binary: A base64 encoded string representing binary data.
• uuid: A string that is a uuid.

number

The field contains numbers of any kind including decimals.

The lexical formatting follows that of decimal in XMLSchema: a non-empty finite-length sequence of decimal digits separated by a period as a decimal indicator. An optional leading sign is allowed. If the sign is omitted, ”+” is assumed. Leading and trailing zeroes are optional. If the fractional part is zero, the period and following zero(es) can be omitted. For example: ‘-1.23’, ‘12678967.543233’, ‘+100000.00’, ‘210’.

The following special string values are permitted (case need not be respected):

• NaN: not a number
• INF: positive infinity
• -INF: negative infinity

A number MAY also have a trailing:

• exponent: this MUST consist of an E followed by an optional + or - sign followed by one or more decimal digits (0-9)

This lexical formatting MAY be modified using these additional properties:

• decimalChar: A string whose value is used to represent a decimal point within the number. The default value is ”.“.
• groupChar: A string whose value is used to group digits within the number. This property does not have a default value. A common value is ”,” e.g. “100,000”.
• bareNumber: a boolean field with a default of true. If true the physical contents of this field MUST follow the formatting constraints already set out. If false the contents of this field may contain leading and/or trailing non-numeric characters (which implementors MUST therefore strip). The purpose of bareNumber is to allow publishers to publish numeric data that contains trailing characters such as percentages e.g. 95% or leading characters such as currencies e.g. €95 or EUR 95. Note that it is entirely up to implementors what, if anything, they do with stripped text.

format: no options (other than the default).

integer

The field contains integers - that is whole numbers.

Integer values are indicated in the standard way for any valid integer.

This lexical formatting MAY be modified using these additional properties:

• groupChar: A string whose value is used to group digits within the integer. This property does not have a default value. A common value is ”,” e.g. “100,000”.
• bareNumber: a boolean field with a default of true. If true the physical contents of this field MUST follow the formatting constraints already set out. If false the contents of this field may contain leading and/or trailing non-numeric characters (which implementors MUST therefore strip). The purpose of bareNumber is to allow publishers to publish numeric data that contains trailing characters such as percentages e.g. 95% or leading characters such as currencies e.g. €95 or EUR 95. Note that it is entirely up to implementors what, if anything, they do with stripped text.

format: no options (other than the default).

boolean

The field contains boolean (true/false) data.

In the physical representations of data where boolean values are represented with strings, the values set in trueValues and falseValues are to be cast to their logical representation as booleans. trueValues and falseValues are arrays which can be customised to user need. The default values for these are in the additional properties section below.

The boolean field can be customised with these additional properties:

• trueValues: [ "true", "True", "TRUE", "1" ]
• falseValues: [ "false", "False", "FALSE", "0" ]

format: no options (other than the default).

object

The field contains a valid JSON object.

format: no options (other than the default).

array

The field contains a valid JSON array.

format: no options (other than the default).

list

The field contains data that is an ordered one-level depth collection of primitive values with a fixed item type. In the lexical representation, the field MUST contain a string with values separated by a delimiter which is , (comma) by default e.g. value1,value2. In comparison to the array type, the list type is directly modelled on the concept of SQL typed collections.

format: no options (other than the default).

The list field can be customised with these additional properties:

• delimiter: specifies the character sequence which separates lexically represented list items. If not present, the default is , (comma).
• itemType: specifies the list item type in terms of existent Table Schema types. If present, it MUST be one of string, integer, boolean, number, datetme, date, and time. If not present, the default is string. A data consumer MUST process list items as it were individual values of the corresponding data type. Note, that on lexical level only default formats are supported, for example, for a list with itemType set to date, items have to be in default form for dates i.e. yyyy-mm-dd.

datetime

The field contains a date with a time.

format:

• default: The lexical representation MUST be in a form defined by XML Schema containing required date and time parts, followed by optional milliseconds and timezone parts, for example, 2024-01-26T15:00:00 or 2024-01-26T15:00:00.300-05:00.
• <PATTERN>: values in this field can be parsed according to <PATTERN>. <PATTERN> MUST follow the syntax of standard Python / C strptime. Values in the this field SHOULD be parsable by Python / C standard strptime using <PATTERN>. Example for "format": ""%d/%m/%Y %H:%M:%S" which would correspond to a date with time like: 12/11/2018 09:15:32.
• any: Any parsable representation of the value. The implementing library can attempt to parse the datetime via a range of strategies. An example is dateutil.parser.parse from the python-dateutils library. It is NOT RECOMMENDED to use any format as it might cause interoperability issues.

date

The field contains a date without a time.

format:

• default: The lexical representation MUST be yyyy-mm-dd e.g. 2024-01-26
• <PATTERN>: The same as for datetime
• any: The same as for datetime

time

The field contains a time without a date.

format:

• default: The lexical representation MUST be hh:mm:ss e.g. 15:00:00
• <PATTERN>: The same as for datetime
• any: The same as for datetime

year

A calendar year as per XMLSchema gYear.

Usual lexical representation is YYYY. There are no format options.

yearmonth

A specific month in a specific year as per XMLSchema gYearMonth.

Usual lexical representation is: YYYY-MM. There are no format options.

duration

A duration of time.

We follow the definition of XML Schema duration datatype directly and that definition is implicitly inlined here.

To summarize: the lexical representation for duration is the ISO 8601 extended format PnYnMnDTnHnMnS, where nY represents the number of years, nM the number of months, nD the number of days, ‘T’ is the date/time separator, nH the number of hours, nM the number of minutes and nS the number of seconds. The number of seconds can include decimal digits to arbitrary precision. Date and time elements including their designator MAY be omitted if their value is zero, and lower order elements MAY also be omitted for reduced precision.

format: no options (other than the default).

geopoint

The field contains data describing a geographic point.

format:

• default: A string of the pattern “lon, lat”, where each value is a number, and lon is the longitude and lat is the latitude (note the space is optional after the ,). E.g. "90.50, 45.50".
• array: A JSON array, or a string parsable as a JSON array, of exactly two items, where each item is a number, and the first item is lon and the second item is lat e.g. [90.50, 45.50]
• object: A JSON object with exactly two keys, lat and lon and each value is a number e.g. {"lon": 90.50, "lat": 45.50}

geojson

The field contains a JSON object according to GeoJSON or TopoJSON spec.

format:

• default: A geojson object as per the GeoJSON spec.
• topojson: A topojson object as per the TopoJSON spec

any

Any type or format is accepted. When converting from physical to logical representation, the behaviour SHOULD be similar to String field type.

Rich Types

A richer, “semantic”, description of the “type” of data in a given column MAY be provided using a rdfType property on a field descriptor.

The value of the rdfType property MUST be the URI of a RDF Class, that is an instance or subclass of RDF Schema Class object

Here is an example using the Schema.org RDF Class http://schema.org/Country:

| Country | Year Date | Value |
| ------- | --------- | ----- |
| US      | 2010      | ...   |

The corresponding Table Schema is:

    {
      fields: [
        {
          "name": "Country",
          "type": "string",
          "rdfType": "http://schema.org/Country"
        }
        ...
      }
    }

Constraints

The constraints property on Table Schema Fields can be used by consumers to list constraints for validating field values. For example, validating the data in a Tabular Data Resource against its Table Schema; or as a means to validate data being collected or updated via a data entry interface.

All constraints MUST be tested against the logical representation of data, and the physical representation of constraint values MAY be primitive types as possible in JSON, or represented as strings that are castable with the type and format rules of the field.

A constraints descriptor MUST be a JSON object and MAY contain one or more of the following properties.

Property	Type	Applies to	Description
required	boolean	All	Indicates whether this field cannot be null. If required is false (the default), then null is allowed. See the section on missingValues for how, in the physical representation of the data, strings can represent null values.
unique	boolean	All	If true, then all values for that field `MUST` be unique within the data file in which it is found.
minLength	integer	collections (string, array, object)	An integer that specifies the minimum length of a value.
maxLength	integer	collections (string, array, object)	An integer that specifies the maximum length of a value.
minimum	integer, number, date, time, datetime, duration, year, yearmonth	integer, number, date, time, datetime, duration, year, yearmonth	Specifies a minimum value for a field. This is different to minLength which checks the number of items in the value. A minimum value constraint checks whether a field value is greater than or equal to the specified value. The range checking depends on the type of the field. E.g. an integer field may have a minimum value of 100; a date field might have a minimum date. If a minimum value constraint is specified then the field descriptor MUST contain a type key.
maximum	integer, number, date, time, datetime, duration, year, yearmonth	integer, number, date, time, datetime, duration, year, yearmonth	As for minimum, but specifies a maximum value for a field.
exclusiveMinimum	integer, number, date, time, datetime, duration, year, yearmonth	integer, number, date, time, datetime, duration, year, yearmonth	As for minimum, but for expressing exclusive range.
exclusiveMaximum	integer, number, date, time, datetime, duration, year, yearmonth	integer, number, date, time, datetime, duration, year, yearmonth	As for maximum, but for expressing exclusive range.
jsonSchema	object	array, object	A valid JSON Schema object to validate field values. If a field value conforms to the provided JSON Schema then this field value is valid.
pattern	string	string	A regular expression that can be used to test field values. If the regular expression matches then the value is valid. The values of this field MUST conform to the standard XML Schema regular expression syntax.
enum	array	All	The value of the field `MUST` exactly match a value in the enum array.

Implementors:

• Implementations SHOULD report an error if an attempt is made to evaluate a value against an unsupported constraint.
• A constraints descriptor MAY contain multiple constraints, in which case implementations MUST apply all the constraints when determining if a field value is valid.
• Constraints MUST be applied on the logical representation of field values and constraint values.

Other Properties

In additional to field descriptors, there are the following “table level” properties.

Missing Values

Many datasets arrive with missing data values, either because a value was not collected or it never existed. Missing values may be indicated simply by the value being empty in other cases a special value may have been used e.g. -, NaN, 0, -9999 etc.

missingValues dictates which string values MUST be treated as null values. This conversion to null is done before any other attempted type-specific string conversion. The default value [ "" ] means that empty strings will be converted to null before any other processing takes place. Providing the empty list [] means that no conversion to null will be done, on any value.

missingValues MUST be an array where each entry is a string.

Why strings: missingValues are strings rather than being the data type of the particular field. This allows for comparison prior to casting and for fields to have missing value which are not of their type, for example a number field to have missing values indicated by -.

Examples:

"missingValues": [""]
"missingValues": ["-"]
"missingValues": ["NaN", "-"]

Primary Key

A primary key is a field or set of fields that uniquely identifies each row in the table. Per SQL standards, the fields cannot be null, so their use in the primary key is equivalent to adding required: true to their constraints.

The primaryKey entry in the schema object is optional. If present it specifies the primary key for this table.

The primaryKey, if present, MUST be an array of strings with each string corresponding to one of the field name values in the fields array (denoting that the primary key is made up of those fields). It is acceptable to have an array with a single value (indicating just one field in the primary key). Strictly, order of values in the array does not matter. However, it is RECOMMENDED that one follow the order the fields in the fields has as client applications MAY utilize the order of the primary key list (e.g. in concatenating values together).

Here’s an example:

"schema": {
  "fields": [
    {
      "name": "a"
    },
    {
      "name": "b"
    },
    {
      "name": "c"
    },
    ...
  ],
  "primaryKey": ["a", "c"]
}

Backward Compatibility

Data consumer MUST support the primaryKey property in a form of a single string e.g. primaryKey: a which was a part of the v1.0 of the specification.

Unique Keys

A unique key is a field or a set of fields that are required to have unique logical values in each row in the table. It is directly modeled on the concept of unique constraint in SQL.

The uniqueKeys property, if present, MUST be a non-empty array. Each entry in the array MUST be a uniqueKey. A uniqueKey MUST be an array of strings with each string corresponding to one of the field name values in the fields array, denoting that the unique key is made up of those fields. It is acceptable to have an array with a single value, indicating just one field in the unique key.

An example of using the uniqueKeys property:

"fields": [
  {
    "name": "a"
  },
  {
    "name": "b"
  },
  {
    "name": "c"
  }
],
"uniqueKeys": [
  ["a"],
  ["a", "b"],
  ["a", "c"]
]

In the case of the definition above, the data in the table has to be considered valid only if:

• each row has a unique logical value in the field a
• each row has a unique set of logical values in the fields a and b
• each row has a unique set of logical values in the fields a and c

Handling null values

All the field values that are on the logical level are considered to be null values MUST be excluded from the uniqueness check, as the uniqueKeys property is modeled on the concept of unique constraint in SQL.

Relation to constraints.unique

In contrast with field.constraints.unique, uniqueKeys allows to define uniqueness as a combination of fields. Both properties SHOULD be assessed separately.

Foreign Keys

A foreign key is a reference where values in a field (or fields) on the table (‘resource’ in data package terminology) described by this Table Schema connect to values a field (or fields) on this or a separate table (resource). They are directly modelled on the concept of foreign keys in SQL.

The foreignKeys property, if present, MUST be an Array. Each entry in the array MUST be a foreignKey. A foreignKey MUST be a object and MUST have the following properties:

• fields - fields is an array of strings specifying the field or fields on this resource that form the source part of the foreign key. The structure of the array is as per primaryKey above.
• reference - reference MUST be a object. The object
  • MUST have a property fields which is an array of strings of the same length as the outer fields, describing the field (or fields) references on the destination resource. The structure of the array is as per primaryKey above.
  • MAY have a property resource which is the name of the resource within the current data package, i.e. the data package within which this Table Schema is located. For referencing another data resource the resource property MUST be provided. For self-referencing, i.e. references between fields in this Table Schema, the resource property MUST be omitted.

Here’s an example:

"resources": [
  {
    "name": "state-codes",
    "schema": {
      "fields": [
        {"name": "code"}
      ]
    }
  },
  {
    "name": "population-by-state",
    "schema": {
      "fields": [
        {"name": "state-code"}
      ],
      "foreignKeys": [
        {
          "fields": ["state-code"],
          "reference": {
            "resource": "state-codes",
            "fields": ["code"]
          }
        }
      ]
    }
  }
]

An example of a self-referencing foreign key:

"resources": [
  {
    "name": "xxx",
    "schema": {
      "fields": [
        {"name": "parent"},
        {"name": "id"}
      ],
      "foreignKeys": [
        {
          "fields": ["parent"],
          "reference": {
            "fields": ["id"]
          }
        }
      ]
    }
  }
]

Foreign Keys create links between one Table Schema and another Table Schema, and implicitly between the data tables described by those Table Schemas. If the foreign key is referring to another Table Schema how is that other Table Schema discovered? The answer is that a Table Schema will usually be embedded inside some larger descriptor for a dataset, in particular as the schema for a resource in the resources array of a Data Package. It is the use of Table Schema in this way that permits a meaningful use of a non-empty resource property on the foreign key.

Backward Compatibility

If the value of the foreignKey.reference.resource property is an empty string "" a data consumer MUST interpret it as an omited property as an empty string for self-referencing was a part of the v1.0 of the specification.

Backward Compatibility

Data consumer MUST support the foreignKey.fields and foreignKey.reference.fields properties in a form of a single string e.g. "fields": "a" which was a part of the v1.0 of the specification.

Appendix: Related Work

Table Schema draws content and/or inspiration from, among others, the following specifications and implementations:

• XML Schema
• Google BigQuery
• JSON Schema
• DSPL
• HTML5 Forms
• Elasticsearch