Statement

Send a customer's bank statements, in any layout, and get the AML and fraud risk of the whole history: the band, the signals that fired with the rows they cite, the counterparties screened, and the typologies they point to.

Type: statement · Endpoint: POST /api/v1/statements/analyze · File (1M tokens per call)

Request

curl https://api.mandatelabs.ai/api/v1/decide \
  -H "X-API-Key: $WARD_API_KEY" -H "Content-Type: application/json" \
  -d '{"type": "statement", "subject_ref": "cust_1042", "subject_consent": true, "input": {
        "files": [{"file_base64": "...", "file_name": "2026-08.pdf"}, {"file_base64": "...", "file_name": "2026-09.pdf"}],
        "profile": {"monthly_income": 3200, "occupation": "nurse", "customer_type": "individual"},
        "focus": "aml"}}'

d = client.statement(["2026-08.pdf", "2026-09.pdf"], profile={"monthly_income": 3200, "occupation": "nurse"},
                     focus="aml", subject_ref="cust_1042", subject_consent=True)

const d = await client.statement(["2026-08.pdf", "2026-09.pdf"], { profile: { monthly_income: 3200, occupation: "nurse" },
  focus: "aml", subject_ref: "cust_1042", subject_consent: true });

The SDK helpers read the files and base64-encode them for you.

Input

Field	Description
`file_base64` + `file_name`, or `files[]`	One statement, or several statements of the same account (up to 200 files, 10 MB each, 60 MB total) merged into one history
`profile`	Declared `monthly_income`, `expected_monthly_inflow`, `occupation`, `customer_type`: movements are read against it
`focus`	`all` (default), `aml` or `fraud`: what the score is for
`screen_counterparties`	Screen counterparty names (default `true`)
`column_map`	A confirmed column reading for a layout Ward could not read alone (Files)
`thresholds`	Thresholds for this analysis only
`subject_ref`	Your customer id

Formats: CSV, TSV, XLSX, PDF with a text layer; English, Spanish and other headers; date order and decimal separator inferred per file. Card network exports (DE fields) and compliance alert exports are recognised too.

Main signals

Code	Family	Weight	Fires when
`STM_STRUCTURING`	AML	0.55	3+ cash or check deposits at 80–100% of the reporting threshold within 30 days
`STM_PASS_THROUGH`	AML	0.50	2+ credits of 1,000+ with 80%+ paid out within 3 days
`STM_FAN_IN`	AML	0.40	8+ distinct payers within 30 days
`STM_VOLUME_SPIKE`	AML	0.30	A month at 3x+ the median of the others
`STM_CASH_INTENSIVE`	AML	0.30	Cash is 50%+ of volume
`STM_GAMBLING`	AML	0.20	Gambling merchants or transfers
`STM_NEW_PAYEE_LARGE`	Fraud	0.35	First-time transfer of 3x+ the median debit and 1,000+
`STM_PEP_COUNTERPARTY`, `STM_QMAP_MERCHANT`, `STM_ADVERSE_MEDIA`	Lists	—	A counterparty matched a list: lifts `APPROVE` to `STEP_UP`

The full matrix (version in components.statement_matrix) is in the result's signals, each with its weight, explanation, the rows it cites and its typologies.

Result

Field	Meaning
`risk.band`, `risk.score`, `risk.score_kind`	`low` < 0.30 ≤ `medium` < 0.60 ≤ `high`
`signals[]`	Code, family, weight, detail, cited rows, typologies
`flagged[]`	The flagged transactions, highest score first
`summary`	Period, transactions, money in and out, top counterparties
`checks`	Sanctions, PEP, adverse media, WMID, enforcement on counterparties: status, names checked, matches
`coverage`	How much of the history Ward could read
`notes`	Every assumption (date order, currency, guessed columns)
`analysis_id`	Stored result: `GET /api/v1/statements/{id}`

Decision: low → APPROVE, medium → STEP_UP, high → DECLINE.

Unreadable files answer 422 with a stable code: unsupported_format, scanned_pdf, encrypted, no_header, not_transactions, needs_mapping, … See Files in any format.