Sample outputDeterministicTier 1 / 3

Receipt OCR Batch (sample)

A batch of 3 receipts processed through structured-extraction OCR. Output is JSON with vendor, date, totals, line items, and a confidence flag per field.

Deterministic includes its failure mode

What the machine produces when the input is well-captured. High confidence, structured output, ready for ingestion.

Input brief the machine received
Three receipts attached. Two are clean phone-camera shots of recent restaurant + office-supply receipts. Extract structured JSON.
Output the machine produced
[
  {
    "receipt_id": "rcpt_001",
    "vendor": "Joe's Coffee Roasters",
    "vendor_address": "412 Elm St, Portland, OR 97205",
    "transaction_date": "2026-04-12",
    "transaction_time": "08:14",
    "subtotal_usd": 12.50,
    "tax_usd": 1.13,
    "tip_usd": 2.50,
    "total_usd": 16.13,
    "line_items": [
      { "description": "Drip coffee, large", "quantity": 2, "unit_price_usd": 4.25, "line_total_usd": 8.50 },
      { "description": "Almond croissant", "quantity": 1, "unit_price_usd": 4.00, "line_total_usd": 4.00 }
    ],
    "payment_method": "Visa ending 4421",
    "confidence": "high",
    "field_confidence": {
      "vendor": "high",
      "transaction_date": "high",
      "total_usd": "high",
      "line_items": "high"
    }
  },
  {
    "receipt_id": "rcpt_002",
    "vendor": "Standard Office Supply",
    "vendor_address": "88 Industrial Blvd, Beaverton, OR 97005",
    "transaction_date": "2026-04-13",
    "transaction_time": "14:32",
    "subtotal_usd": 47.81,
    "tax_usd": 4.30,
    "tip_usd": null,
    "total_usd": 52.11,
    "line_items": [
      { "description": "Letter-size paper, ream", "quantity": 5, "unit_price_usd": 6.99, "line_total_usd": 34.95 },
      { "description": "Black ballpoint pens, 12-pack", "quantity": 1, "unit_price_usd": 4.49, "line_total_usd": 4.49 },
      { "description": "Manila folders, 25-count", "quantity": 1, "unit_price_usd": 8.37, "line_total_usd": 8.37 }
    ],
    "payment_method": "Mastercard ending 7782",
    "confidence": "high",
    "field_confidence": {
      "vendor": "high",
      "transaction_date": "high",
      "total_usd": "high",
      "line_items": "high"
    }
  },
  {
    "receipt_id": "rcpt_003",
    "vendor": "Lou's Diner",
    "vendor_address": "1502 N Mississippi Ave, Portland, OR 97227",
    "transaction_date": "2026-04-14",
    "transaction_time": "12:48",
    "subtotal_usd": 18.95,
    "tax_usd": 1.71,
    "tip_usd": 4.00,
    "total_usd": 24.66,
    "line_items": [
      { "description": "Reuben sandwich", "quantity": 1, "unit_price_usd": 12.95, "line_total_usd": 12.95 },
      { "description": "Side salad", "quantity": 1, "unit_price_usd": 6.00, "line_total_usd": 6.00 }
    ],
    "payment_method": "Visa ending 4421",
    "confidence": "high",
    "field_confidence": {
      "vendor": "high",
      "transaction_date": "high",
      "total_usd": "high",
      "line_items": "high"
    }
  }
]

// 3 of 3 receipts processed at high confidence. Total spend across batch: $92.90 USD. Output is plain JSON, ready for ingestion into your accounting tool.

How to read this sample

This is the first draft the machine produces. It has been sanitized (no real names, no real client data), but the structure, length, and voice are representative of what you'll receive. Because this is a tier-1 job (Deterministic), expect to review for accuracy and adjust voice before shipping. The machine gets you the scaffolding in 90 seconds, so you can spend your time on the 20% that actually needs your judgment. The toggle above shows the safe-failure mode, not just the clean-input ceiling. Difficult inputs return explicit ERROR codes rather than guessed numbers.

RADIO