Technology, AI and Digital Platforms
When an Afghan document reaches your manual review queue, the question is not whether your model is good. It is whether anyone on the file can read a provincial clerk's Pashto handwriting, convert a Hijri Shamsi date, and tell a 2019 seal from a 2024 one. We do that work, and we write it down so your examiner can read it.
Written plainly, because answer engines quote the plain version.
Afghan identity document verification support is a specialist human review service for identity verification (IDV) and KYC vendors whose automated document checks cannot resolve Afghan documents. Ariana Nexus examines the paper Tazkira, the e-Tazkira national ID card and Afghan passports; reads the Pashto and Dari entries; converts Hijri Shamsi dates; reconciles name and transliteration variants; assesses the document against the issuance practice of its stated place and period; and returns a written determination with a field-level evidence file, a machine-readable payload for your case system, and a named reviewer's attestation.
The service is built to be a step inside your pipeline, not a portal your reviewers have to remember to open.
Your orchestration layer routes on issuing country, document class, or a failed template match.
Encrypted submission with the minimum fields required; images and metadata only, no downstream customer data.
Named reviewer, second review, and an audit record of both.
Structured payload with field values, confidence, reason codes and reviewer identifier, plus the human-readable evidence file.
Your rules engine consumes the reason codes. Retention and deletion run to your schedule, evidenced.
Example return
Illustrative structure. Field names, reason codes and confidence bands are agreed against your schema during onboarding.
Nine failure modes account for almost every Afghan file that lands in a manual review queue. None of them are fraud signals. All of them are read as fraud signals by a system that has never seen the document class before.
Template-matching libraries carry thousands of document types. The paper Tazkira is not one of them in any usable form: it is a handwritten booklet whose layout, paper stock, seal and field order vary by province, by issuing office and by decade. There is no machine-readable zone, no fixed field geometry and no substrate to authenticate. A confidence score against a template that does not exist is not a low score. It is a null result being reported as a low score.
Entries are written in Pashto and Dari in a cursive Arabic script with Nastaliq influence, by hand, by clerks with individual conventions. Optical character recognition trained on printed Arabic does not read it. Neither does a general vision model. A native reader does.
Afghan civil records use the Hijri Shamsi (Solar Hijri) calendar. Many records carry a year with no month or day, or an estimated age recorded at the counter rather than a date of birth at all. Systems that demand a full Gregorian date either reject the file or silently write a fabricated 1 January.
Afghan naming does not follow the given-name-plus-surname model most systems assume. A person carries a given name, a father's name and often a grandfather's name, a tribal or regional element, and sometimes an honorific. Romanised into English, a single common name yields a dozen defensible spellings. Exact-match logic misses the same person across two documents; fuzzy logic merges two different people.
The paper Tazkira identifies a person by family chain, volume and page, not by a unique citizen number. The 13-digit national identification number exists only on the e-Tazkira, which entered distribution in 2018. For anyone whose primary lifetime record predates that, there is no number to key on.
The United Nations Security Council 1988 list carries original-script names, approximate or range dates of birth, and national identification fields recorded as 'na'. Screened against common Afghan given names, fuzzy matching produces a wall of hits that a reviewer with no Pashto or Dari cannot clear, so files sit.
Seals, signatory titles, office naming, paper stock and print practice changed after August 2021. A genuine 2019 document and a genuine 2024 document do not look alike. Reviewers comparing one against the other, or against a single reference image of unknown date, reach the wrong conclusion in both directions.
Afghanistan is not a party to the Hague Apostille Convention, so no Afghan public document can carry an apostille. Documents follow a legalisation chain instead. A file presenting an apostilled Afghan document is not a verified file. It is a question.
Afghan customers present UNHCR Proof of Registration cards from Pakistan, Amayesh cards from Iran, consulate-issued extracts, third-country residence permits and school records in place of a national ID. Each carries different evidentiary weight. Treating them as interchangeable is how a file is either wrongly cleared or wrongly declined.
Stated up front, because a vendor who overclaims here costs you a licence, not a renewal.
We do not query Afghan government databases. No accessible, reliable registry exists for third-party verification of Afghan civil records, and any vendor who tells you otherwise is describing something else.
We do not route documents, data or inquiries through authorities inside Afghanistan. Your customer's file does not travel there. This is a firm compliance and security line, not a preference.
We do not obtain, replace, expedite or procure Afghan documents, and we do not act as an agent before any authority.
We analyse documents. We do not produce, alter or replicate them, we do not advise on how to, and we publish no forgery tradecraft.
The determination is evidence for your decision. The onboarding decision, the risk rating and the filing stay yours.
Coverage is by document class and issuance era, not by a single reference image. Where a class cannot support a determination on its own, the register says so.
| Document | Script and name | What it is | Verification posture |
|---|---|---|---|
| DocumentTazkira (paper) | Script and nameتذکره | What it isHandwritten booklet or sheet issued by district and provincial civil registration offices; records lineage, place and issuing office. | Verification postureForensic and contextual review; no unique citizen number to key on |
| DocumentTazkira extract / certified copy | Script and nameنقل تذکره | What it isOffice-issued copy or extract, often the only version a diaspora applicant holds. | Verification postureReviewed for issuance-era and office consistency |
| Documente-Tazkira | Script and nameتذکره الکترونیکی | What it isPolycarbonate national ID card in distribution since 2018; carries a 13-digit national identification number and biometric enrolment. | Verification postureArtefact and data-consistency review; card security features assessed |
| DocumentAfghan passport | Script and nameپاسپورت | What it isMachine-readable travel document; earlier non-machine-readable booklets and later electronic issues are both still in circulation. | Verification postureICAO Doc 9303 structure, machine-readable zone and visual-zone cross-check |
| DocumentBirth certificate | Script and nameتصدیق تولد | What it isHospital or civil registration record; frequently issued years after the birth it records. | Verification postureReviewed as supporting evidence, not primary |
| DocumentMarriage certificate | Script and nameنکاح خط | What it isCourt or registrar record of marriage; central to dependant and beneficiary files. | Verification postureReviewed for lineage and date consistency |
| DocumentEducation records | Script and nameشهادتنامه · سه پرچه | What it isSchool leaving certificate and the three-part ministry record used for university admission and credential review. | Verification postureReviewed for issuance practice; useful as a corroborating lineage source |
| DocumentProperty deed | Script and nameقباله | What it isCourt-registered deed used as address and source-of-wealth evidence. | Verification postureReviewed where source of funds is in question |
| DocumentDisplacement documents | Script and name— | What it isUNHCR Proof of Registration (Pakistan), Amayesh and headshomari records (Iran), third-country permits. | Verification postureClassified by evidentiary weight; never treated as national ID |
Six stages. Each one produces a written artefact, so the determination at the end is traceable to the evidence that produced it.
Encrypted transfer, minimum necessary data, defined retention, named custodian. Nothing is routed through Afghanistan. Where the engagement is at volume, we agree the sampling frame and the escalation thresholds before the first file.
Substrate, print method, seal impression, signature, photograph attachment, laminate, binding, overwriting and erasure — assessed against the practice of the office and period the document claims.
Handwritten and printed entries read in Pashto and Dari by native readers who can read provincial clerical hands, not transcribed by optical character recognition.
Lineage chain, issuing office, volume and page conventions, serial structure, and calendar conversion with the method shown rather than asserted.
Name graph built across scripts and romanisation standards, dates reconciled across documents, and every inconsistency classified as explained, unexplained, or indicative.
Written determination, confidence, limitations, named reviewer, second review by the Cultural Compliance Bureau, and a machine-readable return to your system.
Every engagement returns the same artefacts, in the same structure, whether the file is one document or ten thousand.
One of five outcomes — consistent, consistent with qualifications, indeterminate, inconsistent, or indicators of alteration — with the reasoning that produced it and the limits of what it can support.
Every field extracted in the original Pashto or Dari, romanised, and rendered in English, with the calendar conversion method shown and each anomaly logged against the field it was found in.
Every defensible romanisation of the subject's name — ICAO Doc 9303 transliteration, library and geographic-board standards, and the colloquial spellings people actually use — so your screening can be re-run against the full set rather than one guess.
Whether apparent sanctions or watchlist hits are attributable to this person, the discriminators used to clear or retain them, and what remains unresolved.
The document assessed against the practice of the office and period it claims, with the reference basis identified.
A structured payload to your case management system: field values, confidence, reason codes, reviewer identifier, and retention metadata.
Named reviewer, credentials, method applied, materials examined, limitations, and date. Written from the first file to survive examination.
Where the file becomes contested, the same team produces a written expert opinion and appears for deposition or testimony.
There is no certification for Afghan document examination, no accredited body, and no registry to call. What exists is a standard of work. We set ours in public so it can be held against us.
Every document is read by a native Pashto or Dari reader who can read a provincial clerk's hand. Optical character recognition is used to index, never to decide. Pashto first, then Dari, in that order, because that is how the documents come.
The bench is built from alumni and scholars of Cornell, the University of Chicago, Brown, the University of British Columbia and Otto-von-Guericke University, among others. Document examination is an analytic discipline. Speaking the language is the entry requirement, not the qualification.
We do not route documents, data or inquiries through authorities inside Afghanistan. For a client onboarding a refugee, a defector or a dissident, that boundary is the whole engagement.
We analyse documents. We do not produce, alter or replicate them, we do not teach anyone to, and we publish nothing that would help. The boundary is written into every engagement letter.
Every part of the work is produced by our own people. No subcontracted examiners, no brokered specialists, no referral fee hidden in the margin. The firm that writes the determination is the firm that defends it.
Method, materials, reasoning and limitations are stated in every determination. A finding that cannot be explained to an examiner is not a finding.
The reviewer who reads your file is inside the same firm as the person who will write the expert opinion and sit for deposition if the decision is challenged.
Afghan public documents that can carry an apostille. Afghanistan is not a party to the Hague Apostille Convention, so the legalisation chain applies instead.
Year e-Tazkira distribution began. Anyone whose primary lifetime record predates it has no national identification number to key on.
Digits in the Afghan national identification number — present on the e-Tazkira, absent from the paper Tazkira.
Afghan languages in the firm's coverage. Pashto and Dari carry most civil documents; the other twenty-two carry the interviews around them.
Afghan document examination is not a language task with a compliance wrapper. It is civil-registration history, forensic document examination, onomastics, sanctions practice and data engineering, carried by people who grew up inside the record system they are reading and then went and studied the disciplines that explain it.
Program oversight
Managing Partner
Oversees the firm's operations and this programme. Founded Ariana Nexus in 2025 and built its three-layer model — the Human Intelligence Collective, the AI Data Factory and the Cultural Compliance Bureau. From Herat.
Senior Partner
Builds the firm's institutional systems and AI platforms. Owns the case system, the structured return schema and the name-graph tooling that makes a determination consumable by a rules engine rather than only by a person. Lived in Kabul.
Principal
Machine learning and data systems. Builds the reference sets, the matching and scoring methods, and the evaluation that tells us when a rule is performing and when it is only agreeing with itself.
Principal
Institutional law. Carries the civil-registration and legalisation side of the work: what an office could and could not issue, in what period, under what authority, and what the legalisation chain for a non-Apostille jurisdiction actually requires.
Principal
Screening and casework. Works sanctions and watchlist alerts against original-script entries, and writes the discriminator record that lets a reviewer clear or retain a name with reasons attached.
Named reviewers appear on every determination. Where a reviewer's own security is a factor, the firm — not the individual — is named on the client-facing document and the identity is held in the engagement record.
The frameworks below govern how identity evidence is collected, weighed and recorded. They are the design basis for the method, and the vocabulary your examiner will use.
| Instrument | What it governs | How this service meets it |
|---|---|---|
| Instrument31 CFR 1020.220 — Customer Identification Program | What it governsVerification of identity within a reasonable time using documentary or non-documentary methods, on a risk basis. | How this service meets itSupplies the documentary-method analysis and the written record of what was examined and why it was sufficient. |
| InstrumentFFIEC BSA/AML Examination Manual | What it governsExaminer expectations for customer due diligence, enhanced due diligence and documented rationale. | How this service meets itDeterminations are written in the structure an examiner reads: evidence, method, conclusion, limitation. |
| InstrumentFATF Recommendation 10 | What it governsIdentification and verification using reliable, independent source documents, data or information. | How this service meets itAddresses the case the recommendation anticipates but most stacks do not handle: reliable documents from a jurisdiction with no queryable source. |
| InstrumentNIST SP 800-63A-4 (July 2025) | What it governsIdentity evidence strength and validation of evidence against the issuing or an authoritative source. | How this service meets itProvides the reasoned assessment where validation against the issuing source is unavailable, and states plainly what that limits. |
| InstrumentICAO Doc 9303 | What it governsMachine-readable travel document specification, including machine-readable zone structure and transliteration of Arabic-script names. | How this service meets itPassport review follows the specification; the name graph uses its transliteration as one of its standards. |
| InstrumentUNSC Resolution 1988 list · OFAC SDN | What it governsName screening obligations against lists whose Afghan entries carry original-script names, approximate dates of birth and absent identification numbers. | How this service meets itScreening notes work the original script, not a single romanisation, and record the discriminators used. |
| InstrumentRegulation (EU) 2024/1624 (AMLR) and AMLA | What it governsHarmonised customer due diligence across the European Union, with reliance and record-keeping rules. | How this service meets itEvidence files are produced in a form a relying institution can hold and an authority can inspect. |
| InstrumenteIDAS 2.0 and the EU Digital Identity Wallet | What it governsPerson identification data and electronic attestation of attributes as Member State wallets come into service. | How this service meets itAttribute-level output is structured for attestation flows, not only for a human file. |
| InstrumentGDPR Articles 9 and 35 · UK GDPR | What it governsSpecial category data and data protection impact assessment for high-risk processing. | How this service meets itMinimum-necessary intake, defined retention, named custodian, and impact assessment support for the client's own record. |
| InstrumentEU AI Act | What it governsObligations attaching to automated systems used in identity and access decisions, including human oversight and accuracy. | How this service meets itProvides the qualified human review and the documented reasoning that an automated decision cannot produce on its own. |
| Engagement | Shape | What happens | Who buys it |
|---|---|---|---|
| EngagementEscalation queue | ShapeContinuous | What happensYour tier-three queue routes Afghan files to us. Determination and machine-readable return against an agreed service level. | Who buys itIDV and KYC platforms, banks, money services businesses |
| EngagementCorridor readiness review | ShapeOne engagement | What happensBefore you open Afghan coverage: document class inventory, decision rules, reviewer guidance, acceptance thresholds and a reference set your team keeps. | Who buys itPlatforms opening a new market or corridor |
| EngagementReviewer enablement | ShapeProgramme | What happensWe train your adjudicators to clear the files that do not need us, and to recognise the ones that do. Raises first-pass rate; reduces the volume you send out. | Who buys itIn-house review and financial crime teams |
| EngagementScreening false-positive pass | ShapeRecurring | What happensYour alert backlog on Afghan names, worked against original-script list entries, with the discriminators documented for your examiner. | Who buys itSanctions, AML and financial crime operations |
| EngagementExpert opinion and testimony | ShapeMatter-based | What happensWritten opinion prepared to withstand examination, and testimony where a decision is contested. | Who buys itCounsel, regulators, adjudicators |
Afghan document classes that fall out of automated review, and the decision rules that keep them from falling out next time.
Diaspora account opening, customer identification programme evidence, and remediation of legacy files.
Corridor onboarding at volume, where declines cost customers and false clears cost licences.
Enhanced due diligence on customers from a jurisdiction with no queryable identity source.
Worker verification where the identity document is the only record that exists.
Beneficiary and dependant verification from lineage records rather than numbers.
Document analysis and expert opinion where a decision turns on whether a record is what it claims.
Caseload document review under timelines that do not forgive rework.
Shahadatnama and Se-Parcha review for admission and credential decisions.
Document analysis under a defined scope of work, with a cleared, credentialed and auditable delivery record.
A single common Afghan given name, romanised by a passport office, a resettlement caseworker, a bank branch, a sanctions list and the person themselves. Exact matching misses every pairing. Fuzzy matching merges people who are not the same.
Screening configuration, entity resolution and the full variant corpus are a separate engagement — see Afghan name matching, transliteration and identity data for screening and KYC.
We receive the document images and the fields required to review them. Not your customer record, not your risk score, not your transaction history.
Retention is set in the engagement, evidenced on deletion, and defaults to the shortest period that supports your audit obligation.
Every file has a named custodian and a named reviewer, recorded, for the life of the engagement.
Work on United States engagements is performed by team members located in the United States unless the engagement specifies otherwise in writing.
Client documents are never used as training data, reference material or sample content. The reference set is built from separately sourced material.
We supply the processing description, retention schedule and method record your data protection impact assessment requires.
The Tazkira is the Afghan national identity document. In its historic paper form it is a handwritten booklet or sheet issued by a district or provincial civil registration office, recording a person's name, father's and grandfather's names, place, and the office and volume that issued it. It establishes citizenship and lineage. It does not carry a unique citizen number, and it has no machine-readable zone.
The paper Tazkira is a handwritten civil register record. The e-Tazkira is a polycarbonate national identity card, in distribution since 2018, carrying a 13-digit national identification number, printed data fields and biometric enrolment. They coexist: a person may hold both, and the two records may not agree on spelling or date, because they were created years apart by different processes.
No. There is no accessible, reliable registry that a third party can query to confirm an Afghan civil record, and Ariana Nexus does not route documents or inquiries through authorities inside Afghanistan. Verification is therefore forensic and contextual: the artefact, the issuance practice of the office and period it claims, the internal logic of the record, and its consistency with every other document in the file.
Because the automated check is asking a question the document cannot answer. There is no template to match, no machine-readable zone on the Tazkira, no unique citizen number before 2018, handwritten Arabic-script entries that optical character recognition does not read, and dates in the Hijri Shamsi calendar that are frequently incomplete. The result is a null reported as a low confidence score, and the file lands in manual review.
No. Afghanistan is not a party to the Hague Apostille Convention, so no Afghan public document can carry an apostille. Documents follow a legalisation chain ending in consular legalisation instead. A file presenting an apostilled Afghan document is a question, not a verification.
Afghan civil records use the Hijri Shamsi calendar, and many carry a year with no month or day, or an age estimated at the counter rather than a recorded date of birth. We convert per date rather than by fixed offset, show the conversion method in the evidence file, and state the precision the record actually supports instead of writing a default date your system will later treat as fact.
For common given names, a dozen or more, all defensible. Afghan naming uses a given name, a father's name and often a grandfather's name, with tribal or regional elements and honorifics, and no universal family name. We return a name graph covering ICAO Doc 9303 transliteration, library and geographic-board standards and colloquial spellings, so screening can be run against the full set.
Validity is your decision under your own risk policy, and policies differ by regulator and by institution. What we supply is the factual basis for it: whether the document is internally consistent, whether it matches the issuance practice of the office and period it claims, and what the file can and cannot support. Seals, signatory titles and print practice changed after August 2021, so a genuine document from either side of that line does not resemble the other.
Yes. Afghan passports are reviewed against the ICAO Doc 9303 specification, with the machine-readable zone parsed and cross-checked against the printed visual zone, the booklet assessed as an artefact, and the biographic data reconciled against the Tazkira or e-Tazkira in the same file. Both machine-readable and electronic issues are in circulation, as are older non-machine-readable booklets.
Service levels are set in the engagement against your queue, not published as a single number, because a single-document escalation and a ten-thousand-file remediation are different commitments. Escalation queues run on an agreed turnaround with an expedited path; portfolio reviews run to an agreed schedule with interim reporting.
Yes, and it is written for that from the first file. Where a decision is contested, the same team produces a written expert opinion prepared to withstand examination, and appears for deposition or testimony. Ariana Nexus is retained directly and does not introduce outside experts to counsel.
Yes, and that is the normal shape of the engagement. We do not replace your platform. We take the document classes it cannot resolve, return a determination and a structured payload your rules engine can consume, and where you want it, hand your own reviewers the decision rules and reference material that reduce how often you need us.
The fastest way to assess this service is to put a real Afghan document in front of it. Send one file from your escalation queue under a mutual non-disclosure agreement and we will return the determination, the evidence file and the structured payload in the format your system would receive them.
