Data-Type Detector Guide • formex

Detectors recognize a format, not just a type

formex ships a library of detectors: small functions that answer one yes/no question about a column — “is each value written in format X?” Each is a pure predicate:

input: a character vector x
output: a logical vector the same length (TRUE where the value matches, NA where x is NA); detectors never error.

The word format is doing the work here. “Is this a date?” is too loose to be useful — almost any short string of digits and slashes could be a date. The detector that earns its place answers something tighter: is this a M/D/YYYY date, with the month first, slashes as separators, and real month and day numbers? That is a format style: a specific written shape, with its delimiters in specific places, drawn from a specific alphabet, often of a specific length, sometimes carrying a self-checking digit. A detector is good exactly when it pins that structure and rejects everything that merely resembles it.

Stress-testing the format with AutoType

So the hard part is not accepting real values — it is rejecting plausible-but-wrong ones that share the data type but break the format. To measure that, formex borrows the negative-example machinery from the AutoType framework (Yan & He, Auto-Type: Synthesizing Type-Detection Logic …, SIGMOD 2018), which synthesizes type detectors and the hard negatives to test them. Starting from a handful of real positives, it mutates them into near-misses along four tiers — each one aimed at a different part of the format’s structure:

Tier	Name	What it changes	The structure it stresses
S1	preserve-structure	replaces only alphanumeric characters, leaving the delimiters in place	the alphabet / values (for loose types it often regenerates a valid value, so it stresses little)
S2	preserve-alphabet	replaces any character, but only with characters seen in real examples	the arrangement, while staying in-vocabulary
S3	random	replaces characters with arbitrary printable ASCII	the alphabet itself (the easiest negatives)
SP	scramble-structure	keeps the alphanumerics and their order, but permutes where the punctuation/delimiters sit	the format style directly — a decimal point, a grouping comma, a symbol in the wrong position

That last tier, SP, is the point of this guide’s framing: it holds the data type constant and attacks only the format style, catching a detector that checks “the right characters are present” but not “in the right places.” In that sense the fixtures test formats more than they test types.

The frozen mix of valid + labelled negatives is the bundled [data_type_tests] object. The report below applies each detector to it and reports pass rates per tier:

Recall — share of real positives accepted. Should be 1.00.
S1 / S2 / S3 / SP — share of generated near-misses wrongly accepted (a false-positive rate). Lower is better; blank means that tier produced no distinct negatives. A high S1 is expected for loose formats (mutating the digits of "$12.30" to "$47.99" is still a valid amount) — read S2/S3/SP for those.

What structure a detector leverages

Every detector in the catalogue names its strategy — the structure in the data it keys on. There are a handful of recurring kinds, from loosest to tightest:

exact lookup — the value must be a member of a closed reference set (ISO currency codes, USPS states, NAICS codes). The structure is the list.
small enum — a lookup against a short fixed vocabulary (census regions, NCES locales).
structured regex — a positional shape: delimiters and character classes in fixed places (EIN = two digits, a dash, seven digits; a GUID = the 8-4-4-4-12 hex pattern). This is where the SP tier bites hardest.
regex + range — a shape plus a numeric bound, so the parts have to be real (an IPv4 octet is 0-255; an ISO date’s month is 1-12).
anchor + length — a fixed width whose prefix must be valid (a county FIPS is five digits whose first two are a real state code).
checksum — the value validates its own check digit (ISBN, ORCID, NPI, DEA), so a single wrong character is caught arithmetically. The tightest structure available.
heuristic / composite — for loose real-world text (addresses), fires when several weaker components co-occur.

The catalogue below lists the detectors that ship with a frozen AutoType fixture, grouped by family, each with its one-line definition, the strategy (the structure it leverages), real examples, and the stress-test report. (A few newer detectors — is_currency_amount(), is_version(), is_boolean(), the phase and geometry detectors — are wired into the ontology but not yet in this frozen fixture, so they are not reported here.)

The catalogue

Currency

`is_currency_code()`

ISO 4217 three-letter alphabetic currency code (USD, EUR, JPY).

Precision: exact lookup
Strategy: Upper-cases the value and tests membership in the ISO 4217 code list.
Examples: AUD, CAD, CNY, DKK, EUR

Recall	S1	S2	S3	SP
1	0.011	0.011	0.006	NA

`is_currency_symbol()`

A currency symbol immediately followed by an amount, e.g. $12.30.

Precision: structured regex
Strategy: Regex: a leading symbol ($ euro yen pound) + digits + optional 1-2 decimals.
Examples: $12.30, €299.10, ¥19, £1222, $1

Recall	S1	S2	S3	SP
1	1	0.257	0.013	0.049

`is_currency_code_symbol()`

An ISO code, symbol, and amount together, e.g. USD$12.30.

Precision: regex + lookup
Strategy: Regex for code+symbol+amount, then validates the 3-letter code against ISO 4217.
Examples: USD$12.30, EUR€299.10, USD$1, EUR€21, JPY¥22

Recall	S1	S2	S3	SP
1	0.029	0.008	0	0.019

Web / network

`is_email()`

An email address.

Precision: structured regex (loose)
Strategy: Regex: local part + @ + dotted domain + TLD; pragmatic, not full RFC 5322.
Examples: cjudgkins0@de.vu, csoutar1@bigcartel.com, afeavearyear2@skype.com, fmacgilrewy3@yellowpages.com, aeadmeades4@tamu.edu

Recall	S1	S2	S3	SP
1	0.68	0.15	0.01	0.434

`is_url()`

An http/https/ftp URL.

Precision: structured regex
Strategy: Regex: scheme :// + dotted host + optional path/query.
Examples: http://ezproxy.baylor.edu/, https://www.ezproxy.unibo.it/, http://ezproxy.anzca.edu.au/, http://libproxy.aalto.fi/, http://elibrary.einstein.yu.edu/

Recall	S1	S2	S3	SP
1	0.067	0.01	0	0.034

`is_ipv4()`

An IPv4 dotted-quad address.

Precision: regex + range
Strategy: Regex for four numeric octets, then validates each is 0-255.
Examples: 54.219.219.166, 49.200.88.156, 189.7.190.3, 206.120.6.24, 80.78.48.54

Recall	S1	S2	S3	SP
1	0.421	0.077	0.003	0.06

`is_ipv6()`

An IPv6 address (full or ::-compressed).

Precision: structured regex
Strategy: Regex over eight hex groups plus common :: compressions.
Examples: 89a4:e654:d143:d1d0:b986:73e4:c8e1:4710, f385:9ac0:aede:bb7e:a620:b2a5:dd6b:ebfb, 4f:45b6:3366:d336:e41b:8df4:696:e02, 8316:886e:bebc:738:e0e0:215f:89de:89fb, 70eb:2fd0:84c2:9abd:7682:ff5b:afff:c24d

Recall	S1	S2	S3	SP
1	1	0.005	0	0

`is_mac()`

A MAC hardware address.

Precision: structured regex
Strategy: Regex: six colon- or dash-separated hex octet pairs.
Examples: 40-1F-3C-B6-3C-BC, 1E-42-64-5E-54-69, 1F-9C-F0-33-2E-E5, 22-1F-8C-02-75-1D, 9B-D7-6F-02-9F-D5

Recall	S1	S2	S3	SP
1	1	0.025	0	0

Color

`is_hex_color()`

A hex color such as #4d9cf8 or #fff.

Precision: structured regex
Strategy: Regex: # + 3 or 6 hex digits.
Examples: #4d9cf8, #1f3a8e, #eac3aa, #dc937b, #2cb0c3

Recall	S1	S2	S3	SP
1	1	0.44	0.024	0

`is_rgb_color()`

An RGB triple such as 220,220,220.

Precision: regex + range
Strategy: Three comma-separated integers, each validated 0-255.
Examples: 220,220,220, 255,250,250, 253,245,245, 240,240,240, 250,235,235

Recall	S1	S2	S3	SP
1	0.315	0.07	0.004	0.035

`is_hsl_color()`

An HSL triple such as 195, 100%, 27%.

Precision: regex + range
Strategy: Hue validated 0-360 and two percentages 0-100.
Examples: 195, 100%, 27%, 195, 100%, 47%, 196, 60%, 49%, 197, 72%, 55%, 200, 29%, 55%

Recall	S1	S2	S3	SP
1	0.506	0	0	0.039

`is_cmyk_color()`

A CMYK quadruple such as 0.0,0.31,0.37,0.0.

Precision: regex + range
Strategy: Four comma-separated values, each validated 0-1.
Examples: 0.0,0.31,0.37,0.0, 0.0,0.52,1.0,0.64, 0.35,0.0,0.25,0.0, 0.52,0.63,0.48,0.0, 0.30,0.6,0.0,0.0

Recall	S1	S2	S3	SP
1	0.097	0	0	0.009

Hash / UUID

`is_md5()`

An MD5 message digest.

Precision: fixed-length regex
Strategy: Exactly 32 hexadecimal characters.
Examples: 6e8a6448a62364225da9991fd8e1ef8c, 3df4ffb6eb24fffab3763041f13dbdea, b205e37bafd229efeb6345565ce428a0, d604c1cd26544a8c30f0bb20a689eac7, 793fce16a0a172e58c193b2899e0c025

Recall	S1	S2	S3	SP
1	1	1	0	NA

`is_sha1()`

A SHA-1 message digest.

Precision: fixed-length regex
Strategy: Exactly 40 hexadecimal characters.
Examples: 16de70e8fcd093964a448a86bc89b1c607152ead, efb6759998e36322258c172aa7e8ba6416b3235f, f1d48e18f51de36d01bafab5bfa68d16a8c0192b, 0ceee2d04e239856ecf18f93b5ee75d72e031f92, ac8a1226405cab9d6049304163d658c5ae611c1a

Recall	S1	S2	S3	SP
1	1	1	0	NA

`is_sha256()`

A SHA-256 message digest.

Precision: fixed-length regex
Strategy: Exactly 64 hexadecimal characters.
Examples: 96a162ea0e727b109724d95fa66659aacc2f5a1478c12bb779aa9c1fbd510450, 59d3b1aafad9d120e6df4a1a6ae6b5edf006be7a044e646b63a9feb8e909c900, 18f778c3525b95d98292595ebd8a0c5ac31bfb6cddeafefff04089468e304b96, e93042fecce04326c178d34b88c70f3ef8e90aaaefbc14e53a17a772c149b113, 21995f09908d3bb4f2c7eaf99a50cbad991aeb652208bb71c55e3161fcfaa493

Recall	S1	S2	S3	SP
1	1	1	0	NA

`is_guid()`

A GUID / UUID.

Precision: structured regex
Strategy: The 8-4-4-4-12 hexadecimal UUID pattern.
Examples: f2a59360-1d1a-49e8-9a2f-f9cdf0d5a811, 5ff3cc50-709c-4ba2-88a7-a2e4cca9f32c, 241c3e6c-1aa5-41c6-bbac-8ad08bdfcde9, 12e8b5a7-a1be-4dec-933a-6f9b040f8da3, 2eddc761-2258-4305-9e7b-c1087df3165c

Recall	S1	S2	S3	SP
1	1	0.035	0	0

Scientific / bio IDs

`is_ensembl_id()`

An Ensembl gene/transcript/protein ID (ENSG…, ENSMUSG…).

Precision: structured regex
Strategy: Regex: ENS + optional species + feature letter (E/G/T/P) + 11 digits.
Examples: ENSMUSG00000102659, ENSMUSG00000093178, ENST00000407559, ENSG00000139618, ENSG00000158636

Recall	S1	S2	S3	SP
1	0	0	0	NA

`is_uniprot_id()`

A UniProtKB protein accession (P33667, Q9BWD1).

Precision: structured regex
Strategy: The official UniProt accession-number pattern.
Examples: P33667, P17888, P02338, Q9BWD1, Q46806

Recall	S1	S2	S3	SP
1	0.311	0.311	0.017	NA

`is_icd9_code()`

An ICD-9-CM diagnosis code.

Precision: structured regex
Strategy: Regex: three digits (or an E/V code) + optional decimal detail.
Examples: 078.11, 008.8, 005.9, 117.9, 003.0

Recall	S1	S2	S3	SP
1	0.652	0.294	0.013	0

`is_icd10_code()`

An ICD-10 code (I61, D76.0).

Precision: structured regex
Strategy: Regex: a letter + two digits + optional dotted detail.
Examples: I61, I63, G45, D76.0, D76.3

Recall	S1	S2	S3	SP
1	0.399	0.242	0.036	0

`is_snp_id()`

A dbSNP reference SNP ID (rs3918290).

Precision: structured regex
Strategy: Regex: rs followed by digits.
Examples: rs3918290, rs1801159, rs1801265, rs17376848, rs1801160

Recall	S1	S2	S3	SP
1	0.139	0.139	0	NA

`is_asin()`

An Amazon Standard Identification Number.

Precision: structured regex
Strategy: Regex: B0 followed by eight alphanumerics.
Examples: B01L7DYXL0, B0065RHC2Q, B004M8T1XS, B004M8T30O, B0065RRJGK

Recall	S1	S2	S3	SP
1	0.262	0.262	0.013	NA

`is_bibcode()`

A 19-character NASA ADS bibliographic code.

Precision: structured regex
Strategy: Regex: 4-digit year + 14 code characters + a final letter.
Examples: 2004PASP..116...98K, 2003ApJ...597..957H, 2003A&A...405..175M, 2003ApJ...590..895W, 2003A&A...404..249B

Recall	S1	S2	S3	SP
1	0.175	0.179	0.017	0.521

Codes / lookups

`is_roman()`

A Roman numeral.

Precision: structured regex
Strategy: The canonical well-formed Roman-numeral regex (rejects e.g. IIII).
Examples: I, II, III, IV, V

Recall	S1	S2	S3	SP
1	0.248	0.248	0.01	NA

`is_http_status()`

An HTTP status code.

Precision: regex + range
Strategy: Three digits within 100-599.
Examples: 100, 101, 200, 201, 202

Recall	S1	S2	S3	SP
1	0.701	0.701	0.015	NA

`is_unix_time()`

A Unix epoch timestamp in seconds.

Precision: regex + range (loose)
Strategy: 9-10 digits within a plausible epoch range; inherently loose.
Examples: 1513036980, 1503296497, 1451638800, 1360228214, 1497917283

Recall	S1	S2	S3	SP
1	0.996	0.996	0.004	NA

`is_zip_code()`

A US ZIP or ZIP+4 code.

Precision: structured regex
Strategy: Five digits, optionally + four.
Examples: 98052, 15206, 32792, 50613, 27292

Recall	S1	S2	S3	SP
1	1	0.743	0.019	0

`is_uk_postcode()`

A UK postcode.

Precision: structured regex
Strategy: The standard UK postcode outward/inward pattern.
Examples: SA43 2BU, PO12 2DU, KA20 3LN, MK43 7TA, DT4 8US

Recall	S1	S2	S3	SP
1	0.258	0.109	0.046	0

`is_ca_postal_code()`

A Canadian postal code or 3-character FSA.

Precision: structured regex
Strategy: Letter-digit-letter, optionally followed by digit-letter-digit.
Examples: V5K, K0H, T4A, T4B, T9S

Recall	S1	S2	S3	SP
1	0.279	0.233	0.152	0

`is_us_state()`

A USPS two-letter state code.

Precision: exact lookup
Strategy: Membership in the 50 states + DC.
Examples: WA, OH, DC, NY, PA

Recall	S1	S2	S3	SP
1	0.059	0.059	0.057	NA

`is_country_code()`

An ISO 3166-1 alpha-2 country code.

Precision: exact lookup
Strategy: Membership in the ISO 3166-1 code list.
Examples: MA, AR, GR, CD, KP

Recall	S1	S2	S3	SP
1	0.467	0.467	0.216	NA

Date / time

`is_iso_date()`

An ISO 8601 calendar date (YYYY-MM-DD).

Precision: regex + range
Strategy: Regex + a month/day range check.
Examples: 2016-03-15, 2016-03-13, 2016-03-12, 2016-03-11, 2020-01-01

Recall	S1	S2	S3	SP
1	0.337	0.085	0	0

`is_iso_datetime()`

An ISO 8601 datetime with optional fraction/zone.

Precision: structured regex
Strategy: Regex over date + T + time + optional .fraction and Z/offset.
Examples: 1987-02-06T06:00:37, 1988-01-12T06:34:58, 1930-07-06T02:08:45, 2009-06-15T13:45:30, 2010-06-15T13:45:30

Recall	S1	S2	S3	SP
1	0.165	0	0	0

`is_us_date()`

A US numeric date (M/D/YY).

Precision: regex + range
Strategy: Regex + month/day validation.
Examples: 3/3/16, 1/27/16, 12/31/99, 7/4/2016, 2/29/2020

Recall	S1	S2	S3	SP
1	0.615	0.125	0	0.032

`is_us_datetime()`

A US numeric datetime with AM/PM.

Precision: structured regex
Strategy: Regex: M/D/YYYY (slash or dash) + time + optional AM/PM.
Examples: 2/6/1987 6:00:37 AM, 1/12/1988 6:34:58 AM, 7/6/1930 2:08:45 AM, 7/15/1971 6:43:21 AM, 05-13-2009 12:35 PM

Recall	S1	S2	S3	SP
1	0.085	0	0	0.01

`is_abbr_month_date()`

Day-month-year with an abbreviated month (1-Jul-14).

Precision: structured regex
Strategy: Regex: day + separator + 3-letter month + year.
Examples: 1-Jul-14, 2-Jul-14, 3-Jul-14, 4-Jul-14, 21/Nov/2012

Recall	S1	S2	S3	SP
1	0.02	0.005	0	0

`is_rfc2822_datetime()`

An RFC 2822 / HTTP datetime.

Precision: structured regex
Strategy: Regex: weekday, DD Mon YYYY hh:mm:ss + zone.
Examples: Fri, 06 Feb 1987 06:00:37 GMT, Tue, 12 Jan 1988 06:34:58 GMT, Sun, 06 Jul 1930 02:08:45 GMT, Thu, 15 Jul 1971 06:43:21 GMT, Mon, 01 Jan 2001 00:00:00 GMT

Recall	S1	S2	S3	SP
1	0	0	0	0

`is_long_date()`

A verbose English date (Friday, June 16, 1967).

Precision: structured regex
Strategy: Regex allowing optional weekday, full/abbrev month, day, year (with era) and time; requires a weekday or a year so it does not overlap Month-Day.
Examples: Friday, June 16, 1967, Thursday, June 9, 1955, Wednesday, May 17, 1967, Friday, June 16, 1967 8:46:44 AM, Friday, February 6, 1987 2:00:37 PM

Recall	S1	S2	S3	SP
1	0	0	0	0

`is_month_year()`

A Month-Year value (February 1987).

Precision: structured regex
Strategy: Regex: month name + optional comma + 4-digit year.
Examples: February 1987, January 1988, July 1930, July 1971, August, 2005

Recall	S1	S2	S3	SP
1	0	0	0	0

`is_month_day()`

A Month-Day value with no year (February 6).

Precision: structured regex
Strategy: Regex: month name + 1-2 digit day.
Examples: February 6, January 12, July 6, July 15, March 1

Recall	S1	S2	S3	SP
1	0.005	0	0.005	0

`is_clock_time()`

A clock time (6:00:37 AM).

Precision: regex + range
Strategy: Regex + hour/minute range validation.
Examples: 6:00:37 AM, 6:34:58 AM, 2:08:45 AM, 6:43:21 AM, 11:59 PM

Recall	S1	S2	S3	SP
1	0.115	0.02	0	0

`is_date_range()`

Two dates joined by ‘to’ or ‘-’.

Precision: structured regex
Strategy: Regex over two ISO/US dates and a joiner.
Examples: 07/21/16 to 08/19/16, 08/20/16 to 09/20/16, 09/21/16 to 10/20/16, 1972-01-01 - 1972-07-01, 1972-07-01 - 1973-01-01

Recall	S1	S2	S3	SP
1	0.144	0	0	0

`is_hijri_date()`

A Hijri (Islamic-calendar) date (11 Shawwal 1430).

Precision: structured regex
Strategy: Regex: day + Islamic month name + year.
Examples: 11 Shawwal 1430, 5 Muharram 1300, 19 Rajab 1460, 6 Muharram 1300, 1 Ramadan 1445

Recall	S1	S2	S3	SP
1	0.006	0	0	0

Address

`is_street_address()`

A street-address line (2325 140th St).

Precision: heuristic regex
Strategy: Regex: house number + name + a street-type suffix (St/Ave/Blvd/…).
Examples: 2325 140th St, 303 Aspen Ct, 1225 S Walnut St, 30 Grandview Ct, 100 Main St

Recall	S1	S2	S3	SP
1	0.013	0.013	0.003	0.047

`is_po_box()`

A PO box (PO Box 12, P.O.B. 3).

Precision: structured regex
Strategy: Regex covering PO / POB / Post Office Box variants + a number.
Examples: PO Box 1234, P.O. Box 56, POB 789, Post Office Box 42, PO BOX 100

Recall	S1	S2	S3	SP
1	0.025	0.01	0	0.073

`is_city_name()`

A US city / place name.

Precision: exact lookup (partial coverage)
Strategy: Normalized membership test against ~1,000 larger US cities; smaller towns miss by design (precision over recall).
Examples: Chicago, New Orleans, Springfield, Pittsburgh, Bloomington

Recall	S1	S2	S3	SP
1	0	0	0	0

`is_city_state()`

A ‘City, State’ value.

Precision: regex + lookup
Strategy: Splits on the comma and validates the trailing token is a real state (abbrev or name).
Examples: Bloomington, IN, Cheshire, CT, Bloomington, Indiana, Los Angeles, CA, New York, NY

Recall	S1	S2	S3	SP
1	0.31	0.062	0.05	0.307

`is_state_zip()`

A ‘State ZIP’ value.

Precision: regex + lookup
Strategy: Regex whose leading token must be a real state, followed by a ZIP.
Examples: IN 47401, CT, 06410, Indiana 47401, CA 90210, TX 78701

Recall	S1	S2	S3	SP
1	0.03	0.015	0	0

`is_address()`

A full postal address.

Precision: heuristic composite
Strategy: Scans one string for street / PO box / city / state / ZIP components and fires when at least three co-occur.
Examples: 1225 S Walnut St, Bloomington, IN 47401, 30 Grandview Ct, Cheshire, CT 06410, 1600 Pennsylvania Ave, Washington, DC 20500, 350 Fifth Avenue, New York, NY 10118, 742 Evergreen Terrace, Springfield, IL 62704

Recall	S1	S2	S3	SP
1	0.005	0	0	0.07

Geography

`is_state_fips()`

A 2-digit state FIPS code.

Precision: exact lookup
Strategy: Membership in the valid state/territory FIPS set.
Examples: 06, 36, 48, 17, 12

Recall	S1	S2	S3	SP
1	0.528	0.528	0.023	NA

`is_county_fips()`

A 5-digit county FIPS (GEOID).

Precision: anchor + length
Strategy: Five digits whose first two are a valid state FIPS.
Examples: 06037, 36061, 48201, 17031, 12086

Recall	S1	S2	S3	SP
1	0.742	0.742	0.016	NA

`is_tract_fips()`

An 11-digit census-tract GEOID.

Precision: anchor + length
Strategy: Eleven digits with a valid state prefix.
Examples: 06037206200, 36061014500, 17031839100, 48201100000, 12086001000

Recall	S1	S2	S3	SP
1	0.785	0.785	0	NA

`is_block_group_fips()`

A 12-digit census block-group GEOID.

Precision: anchor + length
Strategy: Twelve digits with a valid state prefix.
Examples: 060372062001, 360610145001, 170318391001, 482011000001, 120860010001

Recall	S1	S2	S3	SP
1	0.75	0.75	0	NA

`is_block_fips()`

A 15-digit census-block GEOID.

Precision: anchor + length
Strategy: Fifteen digits with a valid state prefix.
Examples: 060372062001001, 360610145001000, 170318391001002, 482011000001003, 120860010001000

Recall	S1	S2	S3	SP
1	0.731	0.731	0	NA

`is_cbsa()`

A 5-digit Core Based Statistical Area code.

Precision: exact lookup
Strategy: Membership in the Census CBSA code list.
Examples: 10100, 10140, 10180, 10220, 10300

Recall	S1	S2	S3	SP
1	0.141	0.141	0.013	NA

`is_csa()`

A 3-digit Combined Statistical Area code.

Precision: exact lookup
Strategy: Membership in the Census CSA code list.
Examples: 101, 220, 364, 408, 348

Recall	S1	S2	S3	SP
1	0.275	0.275	0.01	NA

`is_metro_area_name()`

A CBSA / CSA area title (metro-area name).

Precision: exact lookup (partial)
Strategy: Normalized membership against the official metro-area titles (full and state-stripped forms); distinct from city_name, which will not match these hyphenated multi-city titles.
Examples: Aberdeen, SD, Abilene, TX, Akron, OH, Athens-Clarke County, GA, Corpus Christi, TX

Recall	S1	S2	S3	SP
1	0	0	0	0

`is_nces_locale()`

An NCES locale code (11-13, 21-23, 31-33, 41-43).

Precision: small enum
Strategy: Two-digit enum for City/Suburb/Town/Rural x Large/Midsize/Small.
Examples: 11, 12, 13, 21, 22

Recall	S1	S2	S3	SP
1	0	0	0	NA

`is_urban_rural()`

An urban / rural designation label.

Precision: small enum
Strategy: Membership in a designation vocabulary (urban, rural, suburban, exurban, …).
Examples: Urban, Rural, Suburban, Suburb, Exurban

Recall	S1	S2	S3	SP
1	0	0	0	NA

`is_census_region()`

A US census region or division name.

Precision: small enum
Strategy: Membership in the four regions + nine divisions.
Examples: Northeast, Midwest, South, West, New England

Recall	S1	S2	S3	SP
1	0	0	0	0

Government / entity IDs

`is_ssn()`

A US Social Security Number.

Precision: regex + rules
Strategy: ddd-dd-dddd with issuance validity rules (area not 000/666/900+, group not 00, serial not 0000).
Examples: 540-39-4268, 518-07-5084, 531-33-7396, 212-61-9091, 478-46-4157

Recall	S1	S2	S3	SP
1	0.945	0.21	0	0

`is_ein()`

A US Employer Identification Number.

Precision: structured regex
Strategy: Two digits, a dash, seven digits.
Examples: 15-4464349, 58-9042990, 17-5287634, 11-9778467, 79-3293856

Recall	S1	S2	S3	SP
1	1	0.325	0	0

`is_dea()`

A DEA registration number.

Precision: checksum
Strategy: Two letters + seven digits, with the DEA check digit validated.
Examples: AP5836727, BJ6125341, AD0865937, AS9432042, AJ3247194

Recall	S1	S2	S3	SP
1	0.018	0.018	0	NA

`is_chinese_resident_id()`

A Chinese resident identity-card number.

Precision: checksum
Strategy: Eighteen characters with an ISO 7064 mod-11-2 check character.
Examples: 141002197808194474, 431026197506218370, 210811198609188735, 371625197902109782, 140427197602187066

Recall	S1	S2	S3	SP
1	0.093	0.093	0	NA

`is_uei()`

A SAM.gov Unique Entity Identifier (the 12-character code that replaced the DUNS number).

Precision: structured regex
Strategy: Twelve alphanumerics excluding the letters I and O, first character not 0.
Examples: ZQGGHJH74DW7, R7KZ8N2QP4M5, H3JD9FGT2WX6, AB1CD2EF3GH4, M5N6P7Q8R9S2

Recall	S1	S2	S3	SP
1	1	1	0.005	NA

`is_naics()`

A NAICS industry code (2-6 digits).

Precision: exact lookup
Strategy: Membership in the full 2022 NAICS code set, rolled up to every level.
Examples: 11, 236, 311, 541511, 5415

Recall	S1	S2	S3	SP
1	0.137	0.137	0.004	NA

Phone

`is_phone()`

A US / NANP phone number.

Precision: loose regex
Strategy: Regex over the common US formats and separators.
Examples: 510-577-3201, (309) 647-1820, (609)-993-3001, 650-384-9911, 1-650-384-9911

Recall	S1	S2	S3	SP
1	0.875	0.021	0	0

`is_intl_phone()`

A +-prefixed international phone number.

Precision: loose regex
Strategy: Regex: + country code + 8-15 digits with separators/parentheses.
Examples: +1-4250013981, +61 3 6216 0864, +64 3 477 4000, +44 20 7946 0958, +49 30 12345678

Recall	S1	S2	S3	SP
1	1	0.28	0	0.367

`is_msisdn()`

A bare MSISDN subscriber number.

Precision: loose regex
Strategy: 10-15 digits; very loose, best paired with a variable-name hint.
Examples: 919961345678, 13109976224, 12481234567, 27821234567, 447912345678

Recall	S1	S2	S3	SP
1	1	1	0	NA

Research / scholarly IDs

`is_doi()`

A Digital Object Identifier.

Precision: structured regex
Strategy: Regex: 10./.
Examples: 10.1109/ICDCSW.2017.43, 10.1109/IRI.2016.84, 10.1109/ACC.2016.7526681, 10.1109/PIMRC.2015.7343424, 10.1109/ICICA.2014.47

Recall	S1	S2	S3	SP
1	0.045	0.005	0	0.131

`is_orcid()`

An ORCID researcher iD.

Precision: checksum
Strategy: dddd-dddd-dddd-dddC with an ISO 7064 mod-11-2 check digit (C may be X).
Examples: 0000-0002-0488-8591, 0000-0002-6390-6569, 0000-0001-8666-7241, 0000-0001-6352-7991, 0000-0003-2738-1803

Recall	S1	S2	S3	SP
1	0.058	0.013	0	0

`is_isbn()`

An ISBN-10 or ISBN-13.

Precision: checksum
Strategy: Validates the ISBN-10 (mod 11) or ISBN-13 (mod 10) check digit.
Examples: 140955525-9, 038027303-9, 489911182-7, 099337131-0, 712334370-0

Recall	S1	S2	S3	SP
1	0.058	0.025	0	1

`is_issn()`

An ISSN serial number.

Precision: checksum
Strategy: dddd-dddC with a mod-11 check digit (C may be X).
Examples: 0001-2092, 0001-253X, 0002-0729, 0003-021X, 0004-4687

Recall	S1	S2	S3	SP
1	0.092	0.033	0	0

`is_npi()`

A US National Provider Identifier.

Precision: checksum
Strategy: Ten digits with a Luhn check over the 80840 prefix + first nine digits.
Examples: 1225155880, 1649397217, 1578680153, 1346368917, 1487772901

Recall	S1	S2	S3	SP
1	0.085	0.085	0	NA

Adding a new detector

A detector is just a pure is_<type>() predicate, so adding one is self-contained. Collect 10-25 example values of the new type and build the same positive + hard-negative fixture the report above uses with [generate_autotype_test_cases()] (it generates the S1/S2/S3/SP negatives via [generate_autotype_negatives()]):

cases <- generate_autotype_test_cases(
  P         = c("ENSG00000139618", "ENSMUSG00000102659", "ENST00000407559"),
  data_type = "ensembl_id")

Write the is_ensembl_id() predicate, then score it against that fixture with [build_autotype_results()] and [generate_autotype_report()] to read the same recall/precision table shown here before you adopt it. Once you are happy:

Drop the function into an R/dt-<family>.R file (a character-in, logical-out predicate that never errors).
Add its positive examples to data-raw/datatypes/<family>.R.
Register one line in R/00-data-type-registry.R so guess_data_type() and score_data_type() pick it up, and add a row to the ontology (inst/extdata/research_data_type_ontology_v7.csv) for its semantic type.

See ?generate_autotype_test_cases, ?build_autotype_results, and ?generate_autotype_report for details.