Technology, AI and Digital Platforms
One Afghan name can be written twenty or more ways in the Latin alphabet, and almost none of those spellings are wrong. Screening systems, KYC onboarding checks and vetting files built on English-language name logic break on Pashto and Dari records — flagging thousands of ordinary Afghans and missing the records that matter. Ariana Nexus supplies the name-variant data, transliteration rules, tuning evidence and human review that make those systems work.
Exhibit · one name, twenty-two renderings
Twenty-two Latin renderings of one illustrative name, each built from a spelling pattern common in Afghan records. An exact-match system sees twenty-two people. A fuzzy matcher at 0.90 Jaro-Winkler still splits six of them off, mostly where a title, an abbreviation or a missing name changes the shape of the string — and it groups “Mohammad Ajmal”, a fragment many unrelated people share. A caseworker sees one person.
محمد اجمل زدران
Illustrative example · Mohammad Ajmal Zadran · given name, given name, qawm name
Solid marker: Jaro-Winkler similarity of 0.90 or higher against the first record (lower-cased, accents folded). The matcher groups it.Hollow marker: below 0.90. The matcher treats it as a different person.
Afghan name matching is the problem of deciding whether two records refer to the same Afghan person after a name written in Pashto or Dari script has been romanized into the Latin alphabet inconsistently, without a surname, and often without a reliable date of birth. Ariana Nexus builds the name-variant corpora, transliteration rule sets, screening test data and written analyst review that sanctions screening, KYC and vetting programs need to make that decision defensibly.
Every failure below is a data problem with a documented cause. None of them are solved by raising or lowering a match threshold.
محمد is attested in Latin script as Mohammad, Mohammed, Muhammad, Mohamad, Mohamed, Muhammed, Mohd and more. عبدالرحمن appears as Abdul Rahman, Abdulrahman, Abdurrahman, Abd al-Rahman and Abdul Rehman. The spelling on a customer record is usually the one a clerk, an airline, a resettlement agency or the person themselves chose years ago — not the output of any standard.
Afghanistan has no general tradition of family names, and the national identity document does not normally carry one. What looks like a surname on a Western form is usually a father's name, a grandfather's name, a tribal or qawm name, a place name, or a religious title. Mapping those into first, middle and last name fields silently destroys the link between records.
The United Nations 1988 Sanctions List records Afghan names across four numbered name fields, alongside the name in original script and separately graded good-quality and low-quality aliases. A system that reads field 1 as a forename and field 4 as a surname is reading a patronymic chain as a Western name and will mismatch on both ends.
Mullah, Maulavi, Haji, Qari, Hafiz, Sayed, Shah, Khan, Jan, Gul, Agha and Engineer attach to names in some records and not others. Some are religious titles, some are inherited, some are affectionate. Stripping all of them loses real identifiers; keeping all of them produces matches on the title instead of the person.
Afghan civil records run on the Solar Hijri calendar, whose year begins at Nowruz in late March. Conversions are frequently defaulted to 1 January or 1 Hamal, and sanctions entries themselves carry approximations and ranges rather than dates. Most screening configurations are tuned on the assumption that a matching date of birth confirms identity and a differing one clears it. On Afghan records, both assumptions are wrong.
Herat, Hirat and Herāt; Kandahar and Qandahar; Kunduz, Qunduz and Kondoz; Nangarhar and Nangrahar; Pul-e-Khumri and Pol-e Khomri. District names repeat across provinces, and administrative boundaries have been redrawn. Address matching adds noise instead of confidence unless the gazetteer is built for Afghanistan.
Soundex, Metaphone and edit-distance measures such as Levenshtein and Jaro–Winkler operate on Latin strings and encode English sound patterns. Pashto carries retroflex and affricate consonants — څ, ځ, ښ, ږ, ړ, ڼ, ټ, ډ — with no settled Latin equivalent, and Dari leaves short vowels unwritten. Matching on the romanization instead of the source script compounds the error rather than correcting it.
Perso-Arabic text arrives with Arabic yeh against Farsi yeh (ي / ی), Arabic kaf against keheh (ك / ک), heh against teh marbuta, zero-width non-joiners, two families of Arabic-Indic digits, bidirectional control marks and unnormalised presentation forms. Two byte-different strings render identically on screen and compare as different names.
Several of the most frequently held given names in Afghanistan appear inside entries on the UN 1988 List and the OFAC SDN List. A configuration tuned for European name distributions will treat an ordinary customer's name as a partial hit, generate an alert, and do it again at every rescreen.
Six lines of work. Most engagements begin with one and add the others as the institution's Afghan-population data is brought under control.
Script-anchored name records with their attested Latin renderings, patronymic chains, honorific handling, qawm and tribal flags, frequency weighting and provenance for every variant. Built for your Afghan-population data and delivered as a dated, versioned file your matching engine or data platform ingests directly.
A single normalisation layer that reconciles the romanization standards your records were built on with the spellings your customers actually use — so Pashto and Dari names resolve to one canonical form before any matching logic runs.
We take your existing screening configuration and test it against an Afghan-specific variant corpus, by variant class, and report what it caught and what it missed. The output is written in the format your model validation function and your examiner will ask for.
Native Pashto and Dari analysts review level-two and level-three alerts on Afghan names and write the reasoning into the case record — in English, citing the script evidence, so the decision survives an audit years later.
Paper tazkira and e-Tazkira, Afghan passports including the machine-readable zone, driving licences, nikah nama, Se-Parcha and education documents examined for internal consistency, name-chain agreement and authenticity indicators — with a written opinion where a matter is contested.
Examination of the underlying document — Tazkira, e-Tazkira or passport — is a separate engagement.
Your existing Afghan-population records cleaned, de-duplicated and resolved to canonical entities — the remediation that makes every later screening run cheaper and every later audit shorter.
Six weeks is the baseline shape for a first engagement on a single system. A reference-data delivery alone moves faster; multi-system remediation takes longer.
Under a mutual non-disclosure agreement, we take a representative sample of your Afghan-population records in their current state — names, dates, places, document references — and report what is actually in the fields rather than what the schema says should be.
Names are anchored to source script where it exists and reconstructed where it does not. Patronymic chains, honorifics, qawm names and place elements are parsed into separate fields and expanded into their attested variants.
Your live screening configuration is run against the expanded corpus and against a held-back test set. Results are reported by variant class — what was caught, what was missed and at which threshold the behaviour changes.
Normalisation rules, the name-variant corpus, threshold recommendations and the supporting documentation are delivered together, in the format your model validation function and your examiner expect.
List changes, corpus updates and configuration drift are reviewed on a named schedule with a named owner on both sides. Nothing about this data set stays correct on its own.
of sanctions-screening alerts are false positives, and each one takes human review to clear.
Kim and Yang, Frontiers in Artificial Intelligence, 2024
carry an Afghan entry on the UN 1988 Sanctions List, alongside the original script and separately graded good-quality and low-quality aliases.
UN Security Council 1988 Committee, list structure
is recorded on the Afghan national identity document in the ordinary case. There is no general tradition of family names to match on.
Landinfo country of origin report on tazkera and Afghan ID documents
Every row below is a published source. None of them is a claim about Ariana Nexus — they are the documents an examiner will hold your configuration against.
| Instrument | Reference | What it governs |
|---|---|---|
| InstrumentBGN/PCGN National Romanization System for Afghanistan | Reference2007 system | What it governsThe joint United States and United Kingdom standard covering both Dari and Pashto names, designed to apply to Afghan names regardless of language origin. |
| InstrumentBGN/PCGN Romanization of Pashto | Reference1968 system, 2017 revision | What it governsPashto-specific romanization, including the additional consonants absent from Arabic and the handling of the definite article in names of Arabic origin. |
| InstrumentICAO Doc 9303, Part 3 | ReferenceMachine-readable travel documents | What it governsThe transliteration rules that govern the machine-readable zone of an Afghan passport, and the reason the name in the MRZ often differs from the name on the same page. |
| InstrumentUN Security Council 1988 Sanctions List | ReferenceResolution 1988 (2011) regime | What it governsThe Afghanistan and Taliban list. Four numbered name fields, the name in original script, and aliases graded as good quality or low quality — a structure most matching engines discard on import. |
| InstrumentOFAC SDN List and Afghanistan general licences | ReferenceE.O. 13224 · GL 14–20 | What it governsThe Taliban and the Haqqani Network are designated as Specially Designated Global Terrorists. General licences 14 through 20 authorise defined humanitarian, remittance, agricultural and NGO activity. |
| InstrumentALA-LC and ISO 233 | ReferenceLibrary and archival romanization | What it governsThe standards behind name spellings in academic, court and archival records, which frequently differ from both the passport spelling and the bank record. |
| InstrumentAfghan civil registration | ReferenceTazkira and e-Tazkira | What it governsIssued by the National Statistics and Information Authority. The paper tazkira carries no security elements and normally no surname; the e-Tazkira records a Latin-script name the applicant supplies themselves. |
| Failure mode | Generic multilingual screening | With Afghan name data in place |
|---|---|---|
| Failure modeRomanization variance | Generic multilingual screeningFuzzy distance applied to the Latin string | With Afghan name data in placeVariants resolved to a script-anchored canonical record |
| Failure modeMissing surname | Generic multilingual screeningBlank or duplicated last-name field | With Afghan name data in placeGiven, father and grandfather names held in separate fields |
| Failure modeHonorifics | Generic multilingual screeningStripped globally or kept globally | With Afghan name data in placeClassified by type, with rules per class |
| Failure modeDate of birth | Generic multilingual screeningTreated as a strong confirming or clearing signal | With Afghan name data in placeWeighted as weak, with calendar provenance recorded |
| Failure modePlace of birth | Generic multilingual screeningFree-text comparison | With Afghan name data in placeResolved against an Afghan administrative gazetteer |
| Failure modeTuning evidence | Generic multilingual screeningAggregate hit rate | With Afghan name data in placeCatch and miss reported by variant class |
| Failure modeAlert review | Generic multilingual screeningEnglish-language analyst, no script access | With Afghan name data in placeFirst-language review with written script-level reasoning |
Every enterprise screening platform advertises multilingual matching. The phrase usually means Arabic, Cyrillic and Chinese, with Persian handled as a variant of Arabic and Pashto handled not at all. Afghan names are the hardest sub-case in the set and the least represented in the training data, the gazetteers and the test suites those platforms ship.
Ariana Nexus does not compete with those platforms. We supply the layer they do not have. Our people read the source scripts, know how a Kandahari name is built differently from a Herati one, can tell a qawm name from a patronymic on sight, and can put the reasoning in writing for a regulator, an examiner or a court.
The firm is run by alumni and scholars of Cornell, the University of Chicago, the University of British Columbia and Otto-von-Guericke University Magdeburg, working alongside Afghan analysts with first-language command of the scripts. Every engagement is produced in-house: one team, one statement of work, one point of accountability. Nothing is brokered out.
Matching decisions are anchored to the Perso-Arabic original, not to whichever romanization happened to reach your database.
Every recommendation arrives with the test that produced it, written for an audience that will challenge it.
Graduate degrees from Cornell, the University of Chicago and Otto-von-Guericke University Magdeburg, applied to a problem that is linguistic, legal and statistical at once.
No subcontractors, no outsourced review pool, no third-party data brokers in the chain of custody.
Afghan name matching sits between linguistics, computation, document law and financial-crime compliance. It cannot be delivered by a bilingual speaker, and it is not delivered here by one. The people below read the source scripts, trained at the institutions named, and sign the written work that leaves the firm.
Oversees the firm's operations and this programme. Sets the boundary conditions every identity engagement runs under, and owns the client relationship from the first data read to standing review.
Builds and runs the firm's institutional systems, technology and AI platforms, and the data engineering behind each name-variant delivery.
Leads matching methodology — variant expansion, scoring, control test design and the model documentation that accompanies every tuning recommendation the firm makes.
Leads document and civil-registration analysis: tazkira and e-Tazkira, passports and the machine-readable zone, and the reconciliation of a name chain across every document in a file.
Leads analyst review and the written record — disposition language, escalation criteria, and the reasoning that has to still make sense to an auditor several years after the alert was cleared.
Sanctions and PEP screening, KYC onboarding, periodic rescreening and remittance corridors serving Afghan customers and the Afghan diaspora.
Afghan name data, test corpora and language expertise supplied to an existing matching engine, so the platform can claim Afghan coverage it can actually evidence.
Biographic screening, applicant and partner vetting, and document examination where the underlying records are Afghan and the standard of proof is high.
Partner and staff vetting under sanctions constraints, and beneficiary deduplication across agencies working with the same displaced population.
Identity and civil-registration evidence in immigration, asylum, criminal and family matters, with written opinion and testimony where required.
Entity resolution, knowledge-graph construction and training data where Afghan names, places and documents have to resolve correctly at scale.
This page describes work on identity data. The limits below are part of the engagement, written into the statement of work, and they do not move.
The purpose of this work is narrower than the tools it improves. Most of the people caught by a poorly tuned Afghan name rule are not the people the rule was written for. Getting the data right takes them out of the alert queue and leaves the queue to the records that belong in it.
The Afghan national identity document, issued by the National Statistics and Information Authority. The older paper form carries no security elements and normally no surname.
The electronic, biometric form of the tazkira. The Latin-script spelling of the holder's name is supplied by the applicant on the application form, which is why two siblings can hold different official spellings.
A name built from the person's given name followed by the father's and often the grandfather's given name. It is a lineage, not a surname, and it changes shape depending on who is filling in the form.
A community, tribal or solidarity-group name that sometimes appears in the surname position on Western forms and sometimes does not appear at all.
The conversion of a name from Perso-Arabic script into the Latin alphabet. Several competing standards exist, and most real records follow none of them.
On United Nations sanctions lists, an alias assessed as a reliable alternative name for the listed person, as distinct from a low-quality alias.
The Afghan civil calendar, whose year begins at Nowruz in late March. Conversion errors are one of the most common sources of date mismatch in Afghan records.
The two encoded lines at the foot of a passport page. Its transliteration follows ICAO Doc 9303 and frequently differs from the printed name above it.
Because no single romanization standard governs how Pashto and Dari names are written in the Latin alphabet, and because the spelling on most records was chosen by a person rather than produced by a system. Applicants supply their own Latin spelling on Afghan identity documents; airlines, banks, resettlement agencies and immigration systems each add their own. Twenty or more spellings of one name is normal, and none of them is a mistake.
Usually not in the Western sense. Afghanistan has no general tradition of family names, and the national identity document does not normally record one. What appears in a surname field is typically a father's name, a grandfather's name, a tribal or qawm name, a place name or a religious title — and the same person may supply a different one on different forms.
For a United States or United Kingdom institution the reference point is the BGN/PCGN National Romanization System for Afghanistan (2007), with the BGN/PCGN Pashto system for Pashto-specific names and ICAO Doc 9303 for anything derived from a travel document. In practice no single standard is enough, because your customer records were never built on one. The workable answer is a normalisation layer that maps the standards and the real-world spellings to one canonical record.
Most matching engines compare Latin strings. When a name arrives in Perso-Arabic script it is either transliterated on import — introducing a new variant that may match nothing — or held as a string that is byte-different from a visually identical name because of yeh and kaf variants, zero-width joiners or unnormalised Unicode. Both paths produce silent false negatives.
Because Afghan civil records run on the Solar Hijri calendar, whose year begins at Nowruz in late March, and because many people hold no recorded birth date at all. Conversions and intake forms default to the first day of the year. Sanctions entries themselves often carry approximate years or ranges rather than dates. Any configuration that treats date of birth as a strong confirming or clearing signal is mis-tuned for this population.
The tazkira is the Afghan national identity document, issued by the National Statistics and Information Authority. The older paper form carries no security elements and normally no surname. The e-Tazkira is the electronic biometric card; its Latin-script name is supplied by the applicant on the application form, which is why official spellings differ between members of the same family.
We can tell you what the evidence supports and what it does not, in writing, with the reasoning shown. On a well-documented file that is often a clear answer. On a file with two paper tazkiras, no shared date and a name chain that changes across documents, the honest answer is a stated confidence with the specific evidence that would resolve it. We do not manufacture certainty that the records cannot carry.
By testing rather than loosening. We build a variant corpus for your Afghan population, run your live configuration against it, and report catch and miss rates by variant class. That shows which alerts are structurally unavoidable, which come from a fixable data defect, and where the configuration is already missing true matches. Recommendations move specific rules with evidence attached; nothing moves on a global threshold alone.
We do not sell list data and we do not replace your screening vendor. We make your screening work on Afghan names, and we test your configuration against the way Afghan entries are actually structured on those lists — four numbered name fields, original script, and aliases graded good quality or low quality.
Yes. Pashto and Dari carry most Afghan identity records and have the deepest coverage. Uzbeki, Turkmeni, Balochi and the Hazaragi variety of Dari are covered for name and place handling, and the remaining Afghan languages are available on request, scoped case by case rather than promised as standing coverage.
Named members of the firm, not an anonymous pool. Matching methodology sits with Hussain Ahmad (M.Eng., Cornell; Ph.D., University of Chicago). Document and civil-registration analysis sits with Wasil Peroz (M.Sc., Otto-von-Guericke University Magdeburg). Analyst review and the written record sit with Maryam Safi (B.A., Cornell). Hassan Ukasha, Managing Partner, oversees every engagement. The work is produced in-house, with no subcontractors.
With a scoped data read under a mutual non-disclosure agreement. You send a representative sample of your Afghan-population records; we report what is actually in the fields and what it will cost you in alerts and in missed matches. That report stands on its own whether or not the engagement continues.
Send a representative sample of your Afghan-population records under a mutual non-disclosure agreement. We report what is in the fields, what your current configuration does with it, and what that is costing you in both directions.
Initiate an engagement