PII coverage

The pii detector finds 83 kinds of personal identifier with nothing configured, 51 of them confirmed by a checksum rather than a pattern alone. Each carries its patterns, its validator where there is one, and its context words.

The patterns come from presidio-analyzer 2.2.364 (revision 779dbd286d5e, MIT). 9 of the recognizers are pistra’s own, where Presidio has none, marked ‡ and listed below with their evidence. The page is generated from the tables the detector runs, so it cannot list a type the detector does not produce.

The detector is named for the bulk of what it finds, not for all of it. It also finds 19 kinds of API key and private key material, and those have their own page. An API key is not personal data, and it annotates credential/… rather than pii/…. The two are described by different facts: an identifier by its checksum, its language and its fpe alphabet, a credential by the prefix that carries it and where that shape was read from.

84 recognizers · 83 entity types · 166 patterns · 51 checksum-validated · language tags: de 13, en 51, es 3, fi 1, it 5, ko 4, kr 1, pl 1, sv 2, th 1, tr 2.

Entity types

Score is what a match is worth on its own, over the recognizer’s patterns, and +context the same range once one of the Context words is found near it. That second number is the one most rules see, since every recognizer here carries context words. Scores below is how to read both, and why a rule should usually ask a.validated || a.context_supported rather than name a threshold. Checksum says the recognizer has a validator. A span it accepts scores 1.0 and one it rejects is dropped, though it may also abstain and leave the pattern’s score standing. fpe is the alphabet the fpe operator uses for the type out of the box; a type without one is declined by that operator until fpe_alphabets names it. Max span is the longest match the recognizer can produce, in bytes, and it is what a streamed response is held back by when this type is enabled (see the commit horizon); a † means a pattern has an open repeat charged at a cap rather than measured.

Entity	Recognizer	Language	Patterns	Score	+context	Checksum	fpe	Max span	Context
ABA_ROUTING_NUMBER	AbaRoutingRecognizer	en	2	0.05 – 0.30	0.40 – 0.65	yes	digits	38	aba, routing, abarouting, association, bankrouting
AE_EMIRATES_ID	AeEmiratesIdRecognizer ‡	en	1	0.50	0.85			63	emirates id, eid, id number, الهوية الإماراتية, رقم الهوية, بطاقة الهوية, icp
AE_PASSPORT	AePassportRecognizer ‡	en	1	0.10	0.45			33	passport, passport number, جواز, جواز السفر, رقم الجواز
AE_TRN	AeTrnRecognizer ‡	en	1	0.20	0.55			63	trn, tax registration number, vat, الرقم الضريبي, رقم التسجيل الضريبي, fta
AU_ABN	AuAbnRecognizer	en	2	0.01 – 0.10	0.40 – 0.45	yes	digits	47	australian business number, abn
AU_ACN	AuAcnRecognizer	en	2	0.01 – 0.10	0.40 – 0.45	yes	digits	38	australian company number, acn
AU_MEDICARE	AuMedicareRecognizer	en	2	0.01 – 0.10	0.40 – 0.45	yes	digits	42	medicare
AU_TFN	AuTfnRecognizer	en	2	0.01 – 0.10	0.40 – 0.45	yes	digits	38	tax file number, tfn
CA_SIN	CaSinRecognizer	en	2	0.05 – 0.50	0.40 – 0.85	yes	digits	38	sin, sin number, social insurance, social insurance number, canada, nas, numéro nas, numéro d’assurance sociale, assurance sociale
CREDIT_CARD	CreditCardRecognizer	en	1	0.30	0.65	yes	digits	71	credit, card, visa, mastercard, cc , amex, discover, jcb, diners, maestro, instapayment
CRYPTO	CryptoRecognizer	en	1	0.50	0.85	yes		242	wallet, btc, bitcoin, crypto
DATE_TIME	DateRecognizer	en	13	0.10 – 0.80	0.45 – 1.00			144†	date, birthday
DE_BSNR	DeBsnrRecognizer	de	1	0.20	0.55	yes		36	betriebsstättennummer, betriebsstätten-nummer, bsnr, betriebsstätte, praxisnummer, arztpraxis, praxis, kassenärztliche vereinigung, kv-nummer, kv nummer, praxisadresse, praxisstandort, nebenbetriebsstätte, hauptbetriebsstätte, behandlungsort, vertragsarztpraxis
DE_FUEHRERSCHEIN	DeFuehrerscheinRecognizer	de	1	0.35	0.70			42	führerscheinnummer, führerschein, fahrerlaubnis, fahrerlaubnisnummer, fahrerlaubnisklasse, führerscheininhaber, fev, kba, kraftfahrt-bundesamt, driving licence, driving license, driver’s license, licence number, license number, dokument nr, dokument-nr, feld 5
DE_HANDELSREGISTER	DeHandelsregisterRecognizer	de	1	0.50	0.85			91†	handelsregister, handelsregisternummer, amtsgericht, registergericht, hra, hrb, hr-nummer, registerauszug, handelsregistereintrag, firma, gesellschaft, gmbh, ag, ug, kg, ohg, einzelkaufmann, einzelkauffrau, handelsregisterblattnummer
DE_HEALTH_INSURANCE	DeHealthInsuranceRecognizer	de	1	0.30	0.65	yes		39	krankenversicherungsnummer, krankenversichertennummer, versichertennummer, kvnr, krankenkasse, krankenversicherung, gesundheitskarte, egk, elektronische gesundheitskarte, gkv, gesetzliche krankenversicherung, krankenversicherungsausweis, versichertenausweis, versichertenkarte, aok, tkk, barmer, dak
DE_ID_CARD	DeIdCardRecognizer	de	2	0.40 – 0.50	0.75 – 0.85	yes		35	personalausweis, ausweis, personalausweisnummer, ausweisnummer, ausweisdokument, dokumentennummer, seriennummer, npa, neuer personalausweis, personalausweisgesetz, pauwsg, bundespersonalausweis, identity card, national id
DE_KFZ	DeKfzRecognizer	de	5	0.20 – 0.30	0.55 – 0.65			34	kennzeichen, kfz-kennzeichen, kraftfahrzeugkennzeichen, nummernschild, fahrzeugkennzeichen, zulassung, kfz, fahrzeug, auto, pkw, lkw, fahrzeugschein, fahrzeugbrief, zulassungsbescheinigung, amtliches kennzeichen
DE_LANR	DeLanrRecognizer	de	1	0.30	0.65	yes		36	arztnummer, lanr, lebenslange arztnummer, arzt-nr, arzt nr, arzt-nummer, vertragsarzt, kassenarzt, niedergelassener arzt, kbv, kassenärztliche vereinigung, kv-nummer, rezept, verschreibung, behandelnder arzt, hausarzt, facharzt
DE_PASSPORT	DePassportRecognizer	de	1	0.40	0.75	yes		35	reisepass, pass, passnummer, reisepassnummer, passport, passport number, pass-nr, dokumentennummer, bundesrepublik deutschland, ausweisdokument, mrz
DE_PLZ	DePlzRecognizer	de	1	0.05	0.40			20	plz, postleitzahl, postanschrift, adresse, wohnort, ort, wohnanschrift, lieferadresse, rechnungsadresse, straße, strasse, hausnummer, postfach, bundesland, gemeinde, stadt, dorf
DE_SOCIAL_SECURITY	DeSocialSecurityRecognizer	de	2	0.30 – 0.50	0.65 – 0.85	yes		47	rentenversicherungsnummer, sozialversicherungsnummer, versicherungsnummer, rvnr, svnr, sv-nummer, rente, rentenversicherung, deutsche rentenversicherung, drv, sozialversicherung, sozialversicherungsausweis, rentenausweis
DE_TAX_ID	DeTaxIdRecognizer	de	1	0.50	0.85	yes		44	steueridentifikationsnummer, steuer-id, steuerid, steuerliche identifikationsnummer, steuerliche identifikation, persönliche identifikationsnummer, steuer identifikation, idnr, steuer-idnr, steuernummer, bzst
DE_TAX_NUMBER	DeTaxNumberRecognizer	de	3	0.20 – 0.50	0.55 – 0.85			52	steuernummer, steuer-nr, steuer nr, st.-nr, st-nr, finanzamt, umsatzsteuer, einkommensteuer, körperschaftsteuer, gewerbesteuer, steuerveranlagung, steuerbescheid
DE_VAT_ID	DeVatIdRecognizer	de	2	0.40 – 0.50	0.75 – 0.85	yes		41	umsatzsteuer-identifikationsnummer, umsatzsteueridentifikationsnummer, ust-idnr, ust-id, ustidnr, umsatzsteuer-id, mehrwertsteuer, vat, vat-id, vat id, steueridentifikation, bzst, bundeszentralamt für steuern, finanzamt, invoice, rechnung
EMAIL_ADDRESS	EmailRecognizer	en	1	0.50	0.85	yes		458†	email
ES_NIE	EsNieRecognizer	es	1	0.50	0.85	yes		37	número de identificación de extranjero, NIE
ES_NIF	EsNifRecognizer	es	1	0.50	0.85	yes		36	documento nacional de identidad, DNI, NIF, identificación
ES_PASSPORT	EsPassportRecognizer	es	1	0.05	0.40			33	pasaporte, passport, número de pasaporte, passport number
FI_PERSONAL_IDENTITY_CODE	FiPersonalIdentityCodeRecognizer	fi	2	0.10 – 0.50	0.45 – 0.85	yes		41	hetu, henkilötunnus, personbeteckningen, personal identity code
IBAN_CODE	IbanRecognizer	en	3	0.50	0.85	yes		142	iban, bank, transaction
IBAN_CODE	OmIbanRecognizer ‡	en	1	0.50	0.85	yes		91	iban, account, bank, آيبان, رقم الحساب, حساب
IN_AADHAAR	InAadhaarRecognizer	en	2	0.01	0.40	yes	digits	50	aadhaar, uidai
IN_GSTIN	InGstinRecognizer	en	3	0.10 – 0.80	0.45 – 1.00	yes		57	gstin, gst, goods and services tax, tax identification, gst number, gst registration
IN_PAN	InPanRecognizer	en	3	0.01 – 0.50	0.40 – 0.85			40	permanent account number, pan
IN_PASSPORT	InPassportRecognizer	en	1	0.10	0.45			32	passport, indian passport, passport number
IN_VEHICLE_REGISTRATION	InVehicleRegistrationRecognizer	en	9	0.01 – 0.85	0.40 – 1.00	yes		38	RTO, vehicle, plate, registration
IN_VOTER	InVoterRecognizer	en	2	0.30 – 0.40	0.65 – 0.75			37	voter, epic, elector photo identity card
IP_ADDRESS	IpRecognizer	en	12	0.10 – 0.60	0.45 – 0.95	yes		218†	ip, ipv4, ipv6
IT_DRIVER_LICENSE	ItDriverLicenseRecognizer	it	1	0.20	0.55			37	patente, patente di guida, licenza, licenza di guida
IT_FISCAL_CODE	ItFiscalCodeRecognizer	it	1	0.30	0.65	yes		54	codice fiscale, cf
IT_IDENTITY_CARD	ItIdentityCardRecognizer	it	3	0.01	0.40			35	carta, identità, elettronica, cie, documento, riconoscimento, espatrio
IT_PASSPORT	ItPassportRecognizer	it	1	0.01	0.40			34	passaporto, elettronico, italiano, viaggio, viaggiare, estero, documento, dogana
IT_VAT_CODE	ItVatCodeRecognizer	it	1	0.10	0.45	yes		55	piva, partita iva, pi
KR_BRN	KrBrnRecognizer	ko	2	0.05 – 0.10	0.40 – 0.45	yes		42	사업자등록번호, 사업자번호, 사업자, BRN, Business Registration Number, Korean BRN, business number, tax registration number
KR_DRIVER_LICENSE	KrDriverLicenseRecognizer	ko	1	0.05	0.40	yes		51	운전면허, 운전면허번호, 면허번호, Korean driver license, Korean driver’s license
KR_FRN	KrFrnRecognizer	ko	1	0.50	0.85	yes		53	외국인등록번호, Korean FRN, FRN, Foreigner Registration Number, Korean Foreigner Registration Number, 외국인번호
KR_PASSPORT	KrPassportRecognizer	kr	2	0.05 – 0.10	0.40 – 0.45			34	Korean passport, Korean passport number, 대한민국 여권, 여권, passport, passport number
KR_RRN	KrRrnRecognizer	ko	1	0.50	0.85	yes	digits	53	Korean RRN, Korean Resident Registration Number, Resident Registration Number, RRN, rrn, rrn#
KW_CIVIL_ID	KwCivilIdRecognizer ‡	en	1	0.30	0.65	yes	digits	48	civil id, civil number, paci, البطاقة المدنية, الرقم المدني, رقم البطاقة المدنية
MAC_ADDRESS	MacAddressRecognizer	en	2	0.60	0.95	yes		53	mac, mac address, hardware address, physical address, ethernet
MEDICAL_LICENSE	MedicalLicenseRecognizer	en	1	0.40	0.75	yes		35	medical, certificate, DEA
NG_NIN	NgNinRecognizer	en	1	0.01	0.40	yes	digits	44	nin, national identification number, national identity number, nimc, national identity, nigeria id, nigerian identification
NG_VEHICLE_REGISTRATION	NgVehicleRegistrationRecognizer	en	1	0.50	0.85			28	plate number, vehicle registration, license plate, number plate, plate, vehicle, registration
PHONE_NUMBER	PhoneRecognizer	en	1	0.40	0.75		digits	95	phone, number, telephone, cell, cellphone, mobile, call
PH_TIN	PhTinRecognizer	en	2	0.01 – 0.05	0.40	yes		51	tin, taxpayer identification number, bir, taxpayer id, tax id, rdo, revenue district office
PH_UMID	PhUmidRecognizer	en	2	0.30 – 0.50	0.65 – 0.85			50	umid, unified multi-purpose id, crn, common reference number, sss, gsis, philhealth, pag-ibig, umid number, umid card, unified multipurpose id
PL_PESEL	PlPeselRecognizer	pl	1	0.40	0.75	yes	digits	44	PESEL
QA_ID	QaIdRecognizer ‡	en	1	0.10	0.45		digits	44	qid, qatar id, qatari id, id number, البطاقة الشخصية, رقم البطاقة الشخصية, رقم الهوية, moi, metrash
SA_NATIONAL_ID	SaNationalIdRecognizer ‡	en	1	0.30	0.65	yes	digits	40	national id, iqama, id number, هوية, الهوية, الهوية الوطنية, رقم الهوية, إقامة, الإقامة, رقم الإقامة, absher, أبشر
SA_UNIFIED_NUMBER	SaUnifiedNumberRecognizer ‡	en	1	0.10	0.45		digits	40	unified number, unified national number, الرقم الموحد, رقم المنشأة, 700 number
SA_VAT_NUMBER	SaVatRecognizer ‡	en	2	0.10 – 0.50	0.45 – 0.85			60	vat, vat number, tin, tax number, tax id, الرقم الضريبي, رقم ضريبة القيمة المضافة, zatca, هيئة الزكاة
SE_ORGANISATIONSNUMMER	SeOrganisationsnummerRecognizer	sv	2	0.20 – 0.60	0.55 – 0.95	yes	digits	41	organisationsnummer, orgnr, org nr, företagsnummer
SE_PERSONNUMMER	SePersonnummerRecognizer	sv	2	0.10 – 0.50	0.45 – 0.85	yes	digits	49	personnummer, svenskt personnummer, svensk id, ssn, personal identity number, samordningsnummer
SG_NRIC_FIN	SgFinRecognizer	en	2	0.30 – 0.50	0.65 – 0.85			34	fin, fin#, nric, nric#
SG_UEN	SgUenRecognizer	en	1	0.30	0.65	yes		39	uen, unique entity number, business registration, ACRA
TH_TNIN	ThTninRecognizer	th	1	0.50	0.85	yes	digits	52	Thai National ID, Thai ID Number, TNIN, เลขประจำตัวประชาชน, เลขบัตรประชาชน, รหัสปชช
TR_LICENSE_PLATE	TrLicensePlateRecognizer	tr	2	0.30	0.65	yes		35	plaka, araç plakası, plaka numarası, kayıt plakası, tr plaka, license plate, number plate, plate, taşıt plakası, kayıt
TR_NATIONAL_ID	TrNationalIdRecognizer	tr	1	0.30	0.65	yes	digits	44	tc kimlik, kimlik no, kimlik numarası, tckn, tc no, nüfus cüzdanı, national id, turkish id, türk kimlik
UK_DRIVING_LICENCE	UkDrivingLicenceRecognizer	en	1	0.50	0.85	yes		62	driving licence, driving license, driver’s licence, driver’s license, dvla, dl number, licence number, license number
UK_NHS	NhsRecognizer	en	1	0.50	0.85	yes	digits	42	national health service, nhs, health services authority, health authority
UK_NINO	UkNinoRecognizer	en	1	0.50	0.85			36	national insurance, ni number, nino
UK_PASSPORT	UkPassportRecognizer	en	1	0.10	0.45			34	passport, passport number, travel document, uk passport, british passport, her majesty, his majesty, hm passport, hmpo
UK_POSTCODE	UkPostcodeRecognizer	en	1	0.10	0.45			23	postcode, post code, postal code, zip, address, delivery, mailing, shipping, correspondence
UK_VEHICLE_REGISTRATION	UkVehicleRegistrationRecognizer	en	3	0.15 – 0.30	0.50 – 0.65	yes		26	vehicle, registration, number plate, licence plate, license plate, reg, vrn, dvla, v5c, logbook, mot, car, insured vehicle
URL	UrlRecognizer	en	4	0.50 – 0.60	0.85 – 0.95			1118†	url, website, link
US_BANK_NUMBER	UsBankRecognizer	en	1	0.05	0.40		digits	68	check, account, account#, acct, bank, save, debit
US_DRIVER_LICENSE	UsLicenseRecognizer	en	2	0.01 – 0.30	0.40 – 0.65			75	driver, license, permit, lic, identification, dls, cdls, lic#, driving
US_ITIN	UsItinRecognizer	en	3	0.05 – 0.50	0.40 – 0.85		digits	38	individual, taxpayer, itin, tax, payer, taxid, tin
US_MBI	UsMbiRecognizer	en	2	0.30 – 0.50	0.65 – 0.85			42	medicare, mbi, beneficiary, cms, medicaid, hic, hicn
US_NPI	UsNpiRecognizer	en	2	0.10 – 0.40	0.45 – 0.75	yes	digits	42	npi, national provider, provider, npi number, provider id, provider identifier, taxonomy
US_PASSPORT	UsPassportRecognizer	en	2	0.05 – 0.10	0.40 – 0.45			36	us, united, states, passport, passport#, travel, document
US_SSN	UsSsnRecognizer	en	5	0.05 – 0.50	0.40 – 0.85	yes	digits	38	social, security, ssn, ssns, ssid
ZA_ID_NUMBER	ZaIdNumberRecognizer	en	1	0.20	0.55	yes	digits	52	id, identity, identity number, id number, south african id, rsa id, smart id, national id

Scores

A score is the recognizer author’s confidence, not a calibrated probability. For the rows extracted from Presidio it is Presidio’s own number, kept verbatim; a pistra row picks its own on the same scale. Three things move it between the pattern matching and a rule reading it:

	score becomes
a context word within 5 words before the match	min(1.0, max(score + 0.35, 0.40))
a checksum validator says valid	1.0
a checksum validator says invalid	dropped, no annotation at all
a second detector found the same span and type	+ agreement_boost, off unless configured

The validator is tri-state, as it is upstream. A recognizer whose Checksum column says yes can still abstain on a span it does not recognise the shape of, and then the pattern’s own score stands. So the column is a claim about the recognizer, not a guarantee about every match.

Nothing is dropped by default

guardrails.score_threshold defaults to 0, matching Presidio’s default_score_threshold, so every annotation above reaches policy however weak. That is deliberate. A threshold is a silent subtraction, and which findings a deployment can afford to lose is not a question a detector can answer. It does mean that a rule that asks only what was found will fire on the weakest patterns here:

IT_IDENTITY_CARD           ItIdentityCardRecognizer           0.01
IT_PASSPORT                ItPassportRecognizer               0.01
DE_PLZ                     DePlzRecognizer                    0.05
ES_PASSPORT                EsPassportRecognizer               0.05
US_BANK_NUMBER             UsBankRecognizer                   0.05
AE_PASSPORT                AePassportRecognizer               0.10

Those exist to be lifted by the words around them, and they are scored where they are for that reason. On their own they are close to “a run of digits of about the right length”.

Ask for corroboration, not a number

The rule that wants findings worth acting on is this one:

annotations.exists(a, a.category == "pii" &&
                      (a.validated || a.context_supported))

validated is the checksum; context_supported is the enhancer. Together they are what a threshold of 0.40 approximates, and they say it directly, so the rule keeps meaning what it meant if a row’s score is retuned upstream. Reach for a.score when you want to rank findings or tune against your own traffic; reach for these two when you want to act on them. See the policy reference for the annotation vocabulary.

Added by pistra

Presidio has no recognizer for these, so they are pistra’s own, written in guardrails/pii/native/*.yaml in the shape of a custom recognizer, built into the tables by the same generator, covered by the same prefilters, and on by default like every other row.

Evidence is the part worth reading. A national identifier is a run of digits that only a checksum tells from a phone number, and no issuer publishes its algorithm, so the check each row declares is the one the published validators agree on. Where a row declares none it says why, a wrong validator being worse than no validator. Each row is held at test time to the examples beside its declaration, synthetic values that satisfy the scheme and belong to nobody. That is proof of consistency, not of truth; Find Gulf identifiers says how to hold one to an identifier that is real.

Entity	Recognizer	Evidence
AE_EMIRATES_ID	AeEmiratesIdRecognizer	Pattern only, on purpose. A Luhn over all fifteen digits is widely used and real cards are documented failing it, and a wrong validator drops the id it should have found.
AE_PASSPORT	AePassportRecognizer	The shape Microsoft Purview publishes. Nine characters from the digits and C F G H J K L M N P R T V W X Y Z, not all digits; no checksum.
AE_TRN	AeTrnRecognizer	A check digit is said to exist and is not published; shape and context.
IBAN_CODE	OmIbanRecognizer	ISO 13616. Oman’s IBAN is mandatory since 1 July 2024 and Presidio’s IBAN map does not have it yet; goes away when it does.
KW_CIVIL_ID	KwCivilIdRecognizer	Weighted sum modulo 11 (weights 2 1 6 3 7 9 10 5 8 4 2), as several independent validators have it; the issuer publishes nothing.
QA_ID	QaIdRecognizer	Century, birth year, nationality code, serial. No check digit, so the score is low and the context words carry it.
SA_NATIONAL_ID	SaNationalIdRecognizer	Luhn over all ten digits, as every published validator has it (SAP KBA 2384001 among them); the issuer publishes nothing.
SA_UNIFIED_NUMBER	SaUnifiedNumberRecognizer	No checksum is published; the score is low and the context words carry it.
SA_VAT_NUMBER	SaVatRecognizer	Shape only. 3, nine digits, 00003 for the VAT account; the looser fifteen-digit shape scores 0.1.

Phone numbers

PhoneRecognizer is the one recognizer above that is not a pattern. Presidio hands the text to python-phonenumbers, Google’s libphonenumber, at leniency VALID and reports what it finds at 0.4. The detector does the same through the Go port of that library, with its numbering metadata from libphonenumber v9.0.37, so the answer is the library’s rather than a pattern’s approximation of it. phone_regions on the detector lists the regions whose national formats are read (default US GB DE FR IL IN CA BR SA AE QA KW BH OM, Presidio’s eight and the six Gulf states); a number written with its country code is found whatever the list says. Each region is a library call per run of digits long enough to be one of its numbers, so the list is also a cost on text dense with long numbers, and none on prose. A year or a time is shorter than any region’s shortest number and is never shown to the library. A number is reported only within the shape the table’s span is measured from, which is up to sixteen digits with the common separators, so an extension is left off the match rather than the match dropped, and a number in an exotic separator is not found.

Language

The language tag is Presidio’s. Its analyzer runs only the recognizers registered for the language of the request. The pii detector has no request language and runs every active recognizer over every segment, so a German registration plate is found in an English prompt. Narrow with entities: or recognizers: on the detector. Those are also the lever over the streaming horizon.

Digits

Every script’s decimal digits are digits. The engine folds each of them onto its ASCII value before any pattern runs (Arabic-Indic ٠–٩, Devanagari ०–९, fullwidth ０–９, every set Unicode files under category Nd), so an identifier typed from an Arabic keyboard is found by the same recognizer, validated by the same checksum, and reported at its own offsets in the text as sent. This is wider than Presidio, whose \d is unicode and whose [0-9] is not, by accident of authoring; here every pattern sees every digit, custom recognizers included.

What the pattern tier does not cover

Everything above is a structured identifier: a format, sometimes with a checksum. Names, places, organisations, nationalities and dates have no format, and come from the nlp detector (a token-classification model in this process) or the remote detector (a presidio-analyzer with its own NLP engine). Their labels are renamed into the entity vocabulary by entity_map, whose default is:

Model label	Entity
GPE	LOCATION
LOC	LOCATION
MISC	MISC
ORG	ORGANIZATION
PER	PERSON
PERS	PERSON

A label the map does not name keeps its own.

Your deployment’s vocabulary

This page is the floor. A running node’s own list (these types, narrowed by each detector’s selection, plus custom recognizers, a model’s labels and a remote analyzer’s filter) is served live as GET /admin/v1/guardrails/entity-types, with the detector that produces each type. fpe_alphabets and an MCP server’s restore: list are checked against that vocabulary at load.

Regenerate with make gen-docs; re-extract from upstream with make gen-recognizers.