Warddocs
Guides

FraudClassifier and ScamClassifier categories

Many banks classify fraud and scams with two voluntary models published by the Federal Reserve: the FraudClassifier model and the ScamClassifier model. Ward uses the categories of both models, so the cases and outcomes you exchange with other institutions, and report to regulators, read the same way on both sides.

The models are classification structures, not scoring models. Ward's categories are advisory: they sit next to the decision and never change it. This feature is switched on per environment; ask your Mandate contact.

The two models

FraudClassifier asks, in order:

  1. Who initiated the payment? Authorized Party or Unauthorized Party.
  2. How was the fraud executed? (the method)
  3. The tactic.
Code Category
FC.AP Authorized Party
FC.AP.MAN Authorized Party Was Manipulated
FC.AP.MAN.PS Authorized Party Was Manipulated - Products & Services Fraud
FC.AP.MAN.RT Authorized Party Was Manipulated - Relationship & Trust Fraud
FC.AP.ACT Authorized Party Acted Fraudulently
FC.AP.ACT.EMB Authorized Party Acted Fraudulently - Embezzlement
FC.AP.ACT.FCL Authorized Party Acted Fraudulently - False Claim
FC.AP.ACT.SID Authorized Party Acted Fraudulently - Synthetic ID
FC.AP.MOD Unauthorized Party Modified Payment Information (the authorized party initiated; someone else changed it after initiation)
FC.AP.MOD.CC Unauthorized Party Modified Payment Information - Compromised Credentials
FC.AP.MOD.IMP Unauthorized Party Modified Payment Information - Impersonated Authorized Party
FC.AP.MOD.PHY Unauthorized Party Modified Payment Information - Physical Alteration
FC.UP Unauthorized Party
FC.UP.ATO Unauthorized Party Took Over Account
FC.UP.ATO.CC Unauthorized Party Took Over Account - Compromised Credentials
FC.UP.ATO.IMP Unauthorized Party Took Over Account - Impersonated Authorized Party
FC.UP.MIS Unauthorized Party Misused Account Information/Payment Instrument
FC.UP.MIS.DIG Unauthorized Party Misused Account Information/Payment Instrument - Digital Payment
FC.UP.MIS.PFC Unauthorized Party Misused Account Information/Payment Instrument - Physical Forgery/Counterfeit

ScamClassifier has four steps: (1) confirm the scam, the use of deception or manipulation intended to achieve financial gain; (2) confirm an authorized or unauthorized payment; (3) identify the scam category; (4) select the scam type.

Code Category
SC.RES.PAYMENT Authorized Payment (the authorized party made the payment)
SC.RES.ACCESS Unauthorized Payment (the authorized party enabled unauthorized account access)
SC.PS Products or Services
SC.PS.MERCH, SC.PS.INVEST, SC.PS.PROPERTY Merchandise Scam, Investment Scam, Property Sale or Rental Scam
SC.RT Relationship and Trust
SC.RT.ROMANCE, SC.RT.GOV, SC.RT.BANK, SC.RT.BUSINESS, SC.RT.FAMILY, SC.RT.OTHER Romance Impostor, Government Impostor, Bank Impostor, Business Impostor, Relative/Family/Friend and Other Trusted Party Scam

The codes are Ward's identifiers for the categories. Labels and definitions follow FraudClassifier Model and Definitions and ScamClassifier Model and Definitions (FedPaymentsImprovement.org). Answers carry the label. Every answer carries taxonomy_version and mapping_version.

In results

With the feature on, these carry an optional fed_classification when Ward's typologies point to a category:

  • the /decide envelope;
  • statement results, and each flagged transaction in them;
  • the account takeover (dispute) review;
  • the evidence pack (GET /api/v1/decide/{id}/evidence).
"fed_classification": {
  "fraud_classifier": {"code": "FC.UP.ATO", "path": ["Unauthorized Party", "Unauthorized Party Took Over Account"],
                       "label": "Unauthorized Party Took Over Account", "confidence": "exact"},
  "scam_classifier": null,
  "basis": ["K6"],
  "candidates": [{"source": "K6", "fraud_classifier": "FC.UP.ATO", "confidence": "exact", "rationale": "..."}],
  "advisory": true, "taxonomy_version": "fedclass-1.1.0", "mapping_version": "fedmap-1.2.0"
}

When a result names several typologies, Ward claims only what all of them share: the deepest category every candidate sits under, with the weakest confidence among them. The candidates stay listed for your analyst. When nothing maps (money laundering, sanctions, merchant-side schemes), there is no field: Ward never forces a category.

Confidence: exact (every case of the Ward typology lands on this category; it may be an upper level when the lower one depends on the case), likely (most cases do), partial (only part of it fits).

Sending outcomes with the codes

The outcome endpoints take an optional fed_classification with fraud_classifier, scam_result and scam_type:

  • POST /api/v1/decide/{id}/feedback and POST /api/v1/learning/outcomes
  • POST /api/v1/kyc/outcomes
  • POST /api/v1/assess/{id}/outcome and POST /api/v1/assess/fraud-reports
{"correct": true, "fed_classification": {"fraud_classifier": "FC.AP.MAN.RT", "scam_result": "SC.RES.PAYMENT", "scam_type": "SC.RT.ROMANCE"}}

Codes are checked against the models: an unknown code, a code from the wrong level, or codes that contradict each other (for example a scam that ended in a payment with an Unauthorized Party code) are refused with 422. When you send codes without a label, Ward sets its own label from them (for example FC.UP.ATO.CC becomes account_takeover, FC.AP.ACT.FCL becomes first_party_misuse), so your outcomes train your models like any other label.

One pair the two models place differently is accepted: tech-support and lottery scams are Products & Services Fraud in the FraudClassifier and Business Impostor Scam in the ScamClassifier, so FC.AP.MAN.PS with SC.RT.BUSINESS is stored as sent with a fed_note (frameworks_differ: ...) and the two document references. Other contradictions are still refused.

Your own labels: the label dictionary

Send your own outcome names as client_label on the same endpoints (for example "client_label": "ATO-confirmed"). Ward reads them through your label dictionary: each of your labels maps to a Ward label and to codes, with a note.

  • Set it with POST /api/v1/label-map (a full list, as a draft or active) or POST /api/v1/label-map/entries (add or replace entries). Import it from your own document with POST /api/v1/label-map/import: Ward reads one label per line or table row and returns draft entries; nothing is saved until you save them. A bad entry is refused with 422 invalid_entry.
  • Every change is a new version; GET /api/v1/label-map lists them. Mandate may propose entries; a proposal changes nothing until you activate it (POST /api/v1/label-map/{id}/activate) or reject it.
{"entries": [{"client_label": "ATO-confirmed", "ward_label": "account_takeover", "fraud_classifier": "FC.UP.ATO",
              "note": "Fraud ops closing code"}], "activate": true}

When your label and your codes disagree, Ward keeps both and answers "held": true with label_status: "disagreement" and the specific conflict. A label not in your dictionary, sent without codes, is held as unmapped: Ward never guesses. Held outcomes stay out of training until they are resolved, in GET /api/v1/label-map/outcomes and POST /api/v1/label-map/outcomes/{id}/resolve (label_right, codes_right, missing_entry or neither). The resolver must not be the person who submitted the outcome, and a resolution can add the missing entry to your dictionary so the same case does not come back. GET /api/v1/label-map/stats shows your agreement rate, top disagreements and unmapped labels.

Export

GET /api/v1/fed-classification/export?format=csv|jsonl&since=&until= returns your cases and outcomes: Ward's suggested categories next to the outcome and the codes you reported, your own label and its status (agreed, disagreement, unmapped, resolved), with both versions. format=summary returns your agreement statistics. It holds identifiers only (case ids and your own external references), no names or customer references. Each export is written to the audit log.

Wording

Say that a product is "aligned with" or "uses the categories of" the FraudClassifier and ScamClassifier models. The models' terms do not allow any use that suggests the Federal Reserve stands behind a third-party product without written consent, so do not describe Ward, or your use of it, that way. FraudClassifier and ScamClassifier are service marks of the Federal Reserve Banks.

Sources

  • FraudClassifier Model and Definitions and ScamClassifier Model and Definitions (Federal Reserve Banks), available after registration at FedPaymentsImprovement.org: https://fedpaymentsimprovement.org/?p=39306
  • FedPayments Improvement, Fraud and scam classification models: https://fedpaymentsimprovement.org/?p=39306
  • FraudClassifier model (interactive): https://fedpaymentsimprovement.org/fraudclassifier/index.html
  • About the ScamClassifier model: https://fedpaymentsimprovement.org/wp-content/uploads/about-the-scamclassifier-model.pdf
  • Using the FraudClassifier and ScamClassifier models together: https://fedpaymentsimprovement.org/wp-content/uploads/using-the-fraudclassifier-and-scamclassifier-models-together.pdf