European Number, Date and Currency Normalizer

A normalizer for messy European number, date and currency strings such as 1.234,56 €: a parser and converter that returns decimal numbers, ISO 8601 dates and ISO 4217 currency codes as JSON, one item per string, with an ambiguity flag.

Turns messy European number, currency and date strings into clean JSON. You pass a list of strings such as 1.234,56 €, CHF 1'234.50, 3. März 2024 or 03/04/2024; you get one dataset item per string with a numeric value, an ISO 4217 currency code or an ISO 8601 date, plus an ambiguity flag and the other possible readings.

It is a tool over its own input: it reads no website and makes no outbound request.

Also searched as: European number parser, decimal comma to dot, German date to ISO 8601, currency string to amount and ISO 4217 code. British spelling: European number, date and currency normaliser.

Read this first

• Fixed rules, no LLM. The same input and options always give the same output. Nothing is sent to a language model.
• Ambiguous inputs are flagged, and the chosen reading may be wrong. 1.234 can be one thousand two hundred thirty-four or one point two three four; 03/04/2024 can be 3 April or 4 March. The Actor picks one reading by the rules in the table below, sets ambiguous to true and lists the other readings in alternatives. Check flagged items, or pass a locale or dateOrder hint.
• No accuracy guarantee. Unflagged items can be wrong too. Do not use the output for accounting, tax, legal or payment decisions without your own checks.
• Your data. The Actor runs on the Apify platform under your account. Your input and the results stay in your run's storage on Apify, under Apify's retention rules. The author keeps no copy: the code has no server of its own, sends nothing anywhere, and does not write the input strings to the log.
• Built by an automated system. This Actor, its rule tables, its tests and this page were built and are maintained by Augeo, an automated system of AI agents. It was not written by a person, and no person has read the code or the tables.

Accuracy on the bundled 300-case set: see next release notes

What it is used for

• Parse German number formats such as 1.234,56 and parse French number formats such as 1 234,56: both give the JSON number 1234.56.
• Convert a decimal comma to float in a column that follows a European number format, before loading it into a database or a spreadsheet.
• Normalize currency strings from invoices, price lists and exports: turn a currency string to number plus currency code, so 1.234,56 € gives 1234.56 and EUR.
• Parse European dates: convert a written date to ISO 8601, so 3. März 2024 gives 2024-03-03.
• Flag an ambiguous date such as 03/04/2024 instead of guessing in silence.

Input examples

What the Actor returns for each string when it is run with no locale and no dateOrder hint. The rows are outputs of the Actor's own code; tests/check_store_assets.js runs every input of this table through the code again and fails if a cell differs. Each item also carries alternatives, confidence, notes and error (see "Output").

Input	type	value	currency	iso	ambiguous
1.234,56	number	1234.56	null	null	false
1,234.56	number	1234.56	null	null	false
1 234,56	number	1234.56	null	null	false
1.234	number	1234	null	null	true
(1.234,56)	number	-1234.56	null	null	false
12,5 %	number	12.5	null	null	false
1 234,56 €	currency	1234.56	EUR	null	false
CHF 1'234.50	currency	1234.5	CHF	null	false
£12.50	currency	12.5	GBP	null	false
1.234,50 zł	currency	1234.5	PLN	null	false
1 234,50 kr	currency	1234.5	null	null	true
$1,234.56	currency	1234.56	null	null	true
03/04/2024	date	null	null	2024-04-03	true
31/12/2024	date	null	null	2024-12-31	false
31/12/24	date	null	null	2024-12-31	true
3. März 2024	date	null	null	2024-03-03	false
15 de marzo de 2024	date	null	null	2024-03-15	false
2024-12-31T23:59:59Z	date	null	null	2024-12-31	false
31.02.2024	unknown	null	null	null	false
1,23,456	unknown	null	null	null	false

The rows with ambiguous true are the ones to check or to settle with a hint: with locale en-US, 03/04/2024 gives 2024-03-04 and $1,234.56 gives USD (see the next sections).

Input

Field	Type	Required	What it does
items	array of strings	yes	1 to 100,000 strings, one number, amount or date per string, each up to 200 characters.
type	auto, number, currency or date	no (default auto)	auto detects the type of each string. The other values force one reading for every string.
locale	string	no	A tag such as de, fr, en-GB, en-US, de-CH or sv. Settles the decimal separator, the day and month order of all-numeric dates, and shared currency symbols.
dateOrder	auto, DMY, MDY or YMD	no (default auto)	Order of the parts in all-numeric dates. Overrides the order implied by locale. YMD applies only to dates with a two-digit year.

The options apply to every string of the run. For lists in mixed conventions, group the strings by convention and start one run per group.

An empty or missing items list, more than 100,000 strings or an unknown option value ends the run with a failure message; nothing is normalized and nothing is charged.

Output

One dataset item per input string, in the same order. Every item has exactly these ten fields:

Field	Type	Meaning
input	string	The input string, unchanged.
type	string	number, currency, date or unknown.
value	number or null	The numeric value for number and currency, otherwise null.
currency	string or null	ISO 4217 code when one reading is certain, otherwise null.
iso	string or null	The date as YYYY-MM-DD for date, otherwise null.
ambiguous	boolean	true when the string has more than one plausible reading.
alternatives	array	The other readings. Each has a reason and a value, an iso date or a currency.
confidence	string	high, medium or low. It is the outcome of the rules, not a measured probability.
notes	string	Semicolon-separated codes: the rule applied and any flags.
error	string or null	Error code when type is unknown, otherwise null.

One example per type

Number, input 1 234,56:

{
  "input": "1 234,56",
  "type": "number",
  "value": 1234.56,
  "currency": null,
  "iso": null,
  "ambiguous": false,
  "alternatives": [],
  "confidence": "high",
  "notes": "space_thousands_comma_decimal",
  "error": null
}

Currency, input 1.234,56 €:

{
  "input": "1.234,56 €",
  "type": "currency",
  "value": 1234.56,
  "currency": "EUR",
  "iso": null,
  "ambiguous": false,
  "alternatives": [],
  "confidence": "high",
  "notes": "dot_thousands_comma_decimal",
  "error": null
}

Date, input 3. März 2024:

{
  "input": "3. März 2024",
  "type": "date",
  "value": null,
  "currency": null,
  "iso": "2024-03-03",
  "ambiguous": false,
  "alternatives": [],
  "confidence": "high",
  "notes": "month_name",
  "error": null
}

Unknown, input 31.02.2024 (there is no 31 February):

{
  "input": "31.02.2024",
  "type": "unknown",
  "value": null,
  "currency": null,
  "iso": null,
  "ambiguous": false,
  "alternatives": [],
  "confidence": "low",
  "notes": "invalid_date",
  "error": "invalid_date"
}

An ambiguous item, input 03/04/2024 with no hint:

{
  "input": "03/04/2024",
  "type": "date",
  "value": null,
  "currency": null,
  "iso": "2024-04-03",
  "ambiguous": true,
  "alternatives": [{ "iso": "2024-03-04", "reason": "mdy_order" }],
  "confidence": "low",
  "notes": "day_month_order_ambiguous",
  "error": null
}

Error codes

empty_input, not_a_string, input_too_long (over 200 characters), unrecognized_input, unrecognized_number, unrecognized_date, invalid_grouping, invalid_date, date_invalid_for_hint, date_missing_year, date_missing_day, multiple_currencies, number_out_of_range, internal_error.

Ambiguity rules

Case	Example	Reading chosen	Alternatives	Flagged	Confidence
One separator followed by exactly three digits, no locale	1.234, 1,234	Thousands: 1234	Decimal: 1.234	yes	low
The same, with a known locale	1.234 with de gives 1234; with en-GB gives 1.234	The locale's decimal separator decides	none	no	high
Both separators, or a space or apostrophe as thousands separator	1.234,56, 1,234.56, 1 234,56, 1'234.50	The last of the two separators is the decimal mark	none	no	high
All-numeric date where both orders are valid, no hint	03/04/2024	Day first: 2024-04-03	Month first: 2024-03-04	yes	low
All-numeric date where only one order is valid	31/12/2024, 12/31/2024	The valid order	none	no	high
All-numeric date with a dateOrder or locale hint	03/04/2024 with en-US gives 2024-03-04	The hint; dateOrder wins over locale	none	no	high
Two-digit year	31/12/24	00 to 68 read as 2000 to 2068, 69 to 99 as 1969 to 1999	The same day in the other century	yes	medium at most
Two-digit year with dashes	24-12-31	Day first if valid; year first is also considered, and is chosen with locale sv	The other valid orders and the other century	yes	low or medium
Currency symbol shared by several currencies, no settling locale	1 234,50 kr, $1,234.56, 12 lei, ¥500, Fr. 12.50	currency is null; the value is still read	One per candidate currency	yes	medium at most
The same, with a settling locale	1 234,50 kr with sv gives SEK	The locale's currency; note currency_from_locale	none	no	high
Eight digits that also form a date, type auto	20241231	Number: 20241231	The date 2024-12-31	yes	medium
Bare en, en-CA or fr-CA as locale	03/04/2024	No date order is implied; treated as no hint	Month first	yes	low

Other rules that are choices of this Actor, not facts:

• A percent sign is noted (percent) and the value is left as written: 12,5 % gives 12.5, not 0.125.
• A time after a date is dropped and noted (time_ignored).
• £ is read as GBP, although other pound currencies use the same sign.
• Negative amounts are read from a leading or trailing minus and from parentheses: (1.234,56) gives -1234.56.
• 1.234,- and CHF 12.- are read as whole amounts.
• Values are JSON numbers (double precision). Numbers with more than 15 significant digits are flagged precision_loss.
• An unknown locale is ignored and noted (unknown_locale_ignored).
• Indian digit grouping (1,23,456) is not supported and returns invalid_grouping.
• Month names are read in English, German, French, Spanish, Italian, Dutch, Portuguese, Polish and Swedish. Dates in other languages, relative dates ("yesterday") and dates with no year or no day return an error.

Use from an AI agent or MCP

The Actor takes one small JSON input and returns flat JSON items, so an agent can call it as a tool through Apify's MCP server or the Apify API. The tool arguments are the four input fields above; only items is required. The result is the run's default dataset: one item per string, in the input order, always with the same ten fields. The output below is what the Actor's code returns for this input; tests/check_store_assets.js compares the two JSON blocks against the code on every run. Pass a hint whenever the source of the strings is known: it settles the separator, date order and shared-symbol cases listed in the table above.

Input:

{
  "items": ["1.234", "$1,234.56", "03/04/2024"],
  "locale": "en-US"
}

Output (three dataset items):

[
  {
    "input": "1.234",
    "type": "number",
    "value": 1.234,
    "currency": null,
    "iso": null,
    "ambiguous": false,
    "alternatives": [],
    "confidence": "high",
    "notes": "single_separator_three_digits_locale",
    "error": null
  },
  {
    "input": "$1,234.56",
    "type": "currency",
    "value": 1234.56,
    "currency": "USD",
    "iso": null,
    "ambiguous": false,
    "alternatives": [],
    "confidence": "high",
    "notes": "comma_thousands_dot_decimal;currency_from_locale",
    "error": null
  },
  {
    "input": "03/04/2024",
    "type": "date",
    "value": null,
    "currency": null,
    "iso": "2024-03-04",
    "ambiguous": false,
    "alternatives": [],
    "confidence": "high",
    "notes": "mdy_from_hint",
    "error": null
  }
]

A simple rule for an agent: use an item as it is when ambiguous is false and error is null; otherwise look at alternatives, or call again with a locale or dateOrder hint.

Price

Pay per event: USD 0.001 per result, which is USD 1.00 per 1,000 strings.

• One result event is charged per dataset item, and one dataset item is written per input string. result is the only event this Actor's code charges.
• Items of type unknown (strings that could not be read) are dataset items too and are charged like the others.
• If you set a maximum charge for the run, the Actor stops when it is reached: the remaining strings are not processed and not charged, and the status message says how many were left.
• A run that fails input validation charges nothing.

Where the rule tables come from

The full statement is in src/DATA_SOURCES.md, shipped with the code. In short:

• No file from Unicode CLDR, ICU or any other locale data project was copied, downloaded or transformed, so no third-party data licence applies to this Actor.
• The month names, currency symbols, decimal separator conventions and date order conventions were generated entry by entry by the AI model that built the Actor, from its general knowledge. They were not compared against an external reference dataset, so an entry can be missing or wrong.
• The currency codes are a subset of the ISO 4217 alphabetic codes.
• The two-digit year pivot is the rule of the POSIX %y conversion.
• The only runtime dependency is the Apify SDK (the apify package). The normalizer itself uses Node core only.

How it was checked

A fixture script asserts the core module against expected values with node:assert, including 20,000 fixed-seed garbage strings that must never throw and must always return all ten fields. A second fixture script runs the Actor's input validation and batch loop against the examples on this page. Both were written by the same automated system that wrote the code: they are consistency checks, not an independent verification.

Limits

• Deterministic, no LLM: fixed rules, so the same input and options always give the same output. It cannot read a string its rules do not cover.
• Ambiguous inputs are flagged and may be wrong: the Actor picks one reading, sets ambiguous to true and lists the others.
• No accuracy guarantee. Unflagged items can be wrong too.
• Your input is not retained by the author beyond Apify's platform storage: it stays in your run's storage on Apify and the code sends it nowhere else.
• Built by an automated system of AI agents (Augeo), not written by a person; the checks above were written by the same system.
• 100,000 strings per run, 200 characters per string.
• One number, amount or date per string. It does not extract values from sentences or find several values in one string.
• No currency conversion and no exchange rates.
• No time zones: the time part of a timestamp is dropped.