high velocity
​meta.fraud.submissions

This feature identifies unusually high numbers of receipt submissions from a specific device_id across multiple time periods.

The system monitors submission rates during these intervals:

Fraud Color Assignment

Default: Fraud Color set to Yellow if triggered.

The fraud decision color for this feature is fully configurable. Thresholds can be customized for each signal by either:

*Depends on Veryfi Lens data

critical velocity

meta.fraud.submissions

This feature identifies extremely high numbers of receipt submissions from a specific device_id across several time periods. Similar to high velocity but with more aggressive thresholds.

Threshold values for critical velocity are configured with higher limits than standard high velocity.

When a device_id exceeds these elevated thresholds, the system flags potential fraudulent activity with greater urgency.

Fraud Color Assignment

Default: Fraud Color set to Red if triggered.

The fraud decision color for this feature is fully configurable. Thresholds can be customized for each signal by either:

*Depends on Veryfi Lens data

multiple profiles or devices

The system tracks and flags unusual patterns where:

This helps identify potential fraud, account sharing, or bad actors attempting to manipulate the

*Depends on Veryfi Lens data

fraud history

When a user device_id with known fraudulent activity attempts new submission, the system escalates the risk assessment based on this historical pattern of bad behavior..

This feature increases the fraud score when receipts are submitted from a device_id that has previously submitted fraudulent documents.

The system examines:

*Depends on Veryfi Lens data

emulated device

This feature immediately returns the Red color if the document was submitted via app emulation software.

When the system detects that a receipt submission originates from an emulated environment rather than a genuine mobile device, it automatically flags this as high-risk activity. This identifies attempts to use virtual devices, app simulators, or other emulation technologies that may indicate fraudulent submission patterns.

*Depends on Veryfi Lens data

blocked device

This feature immediately returns the Red color if device_id is in a blocked list.

When a receipt submission comes from a device_id that has been explicitly added to the system's blocklist, it automatically triggers the highest risk level. This allows for immediate identification and rejection of submissions from devices previously confirmed to be associated with fraudulent activity.

handwritten characters
​

meta.handwritten_fields

This feature detects handwriting in specific fields of the submitted document. When handwritten modifications are identified, the system provides a list of the exact areas where handwriting occurred using dot notation.
For example, line_items.1.total indicates that the total in line item 1 was manually altered or written.

Fields examined for handwriting detection include:

This analysis helps identify receipts that may have been manually modified to alter key financial information or the extracted value is handwritten. Some customers use this field as a separate signal to build workflows for data handling.
​
​Fraud Color Assignment:

The fraud decision color for this feature is fully configurable. Thresholds can be customized for each signal by either:

This flexibility allows organizations to adjust sensitivity levels based on their specific risk tolerance and fraud patterns.

digital tampering

meta.digital_tampering_fields
​

This feature visually detects instances where documents have been digitally altered or manipulated.

The detection works specifically on photo submissions where visual alterations are present, analyzing pixel-level inconsistencies and pattern anomalies.

This feature can be configured to focus on specific fields, including:

By identifying digital manipulation attempts, this feature helps prevent acceptance of fraudulently modified receipts where visual editing has occurred.
​

Fraud Color Assignment:

The fraud decision color for this feature is fully configurable. Thresholds can be customized for each signal by either:

Not a document
​is_document

This feature identifies if the submitted image is not a valid receipt or document, but rather a random photo of cat. It functions as an enhanced extension of the document-level field is_document.

The system employs document classification algorithms to determine whether the submitted image contains legitimate receipt content or is an attempt to submit irrelevant material. This feature helps prevent fraud attempts where users submit random images or unrelated documents instead of actual receipts.

Fraud Color Assignment:
The fraud decision color for this feature is fully configurable. Thresholds can be customized for each signal by either:

LCD photo

meta.fraud.pages.is_lcd

This feature identifies when the submitted image is a photograph of a digital screen (monitor, tablet, or phone display) rather than an original document.

The system uses advanced image analysis to detect the characteristic patterns, pixel structures, and lighting anomalies that appear when someone takes a picture of an LCD/LED screen displaying a receipt. This helps prevent fraud attempts where users submit screen captures or photos of digitally displayed receipts instead of original receipt images.

By detecting screen photography, this feature helps ensure that only legitimate receipts captured directly from physical documents are processed.
​

Fraud Color Assignment:
The fraud decision color for this feature is fully configurable. Thresholds can be customized for each signal by either:

screenshot

Veryfi provides two complementary signals to identify non-paper document submissions:
​
#1 fraud_type: screenshot with fraud_attribute: Digital Background
A binary classifier that detects whether a document originated digitally rather than as a physical paper document.
​
#2. meta.pages.screenshot_type
Provides granular classification of the digital document type:
- mobile_screenshot — Capture from a mobile device screen
- other_screenshot — Desktop screenshots, PDF exports, and other digital documents
- null — Physical paper document capture
- ai_generated — (Deprecated) AI-generated documents. Use meta.pages.ai_generated instead for this detection.
​
These signals help you flag and filter submissions of digital receipts, screen captures, and non-paper documents in expense management, reimbursement workflows, and compliance scenarios.
​
​Note: The screenshot classification model is not applied to text-based PDFs, so meta.pages.screenshot_type will be absent from the JSON response for those documents.
​

Fraud Color Assignment:

The fraud decision color for this feature is fully configurable. Thresholds can be customized for each signal by either:

generated document

meta.pages.ai_generated

This feature is designed to detect images created by AI tools (such as ChatGPT, Gemini or other generative AI systems). It employs a dual detection approach:

This feature helps prevent fraud attempts involving artificially created receipts that never existed in reality but appear legitimate.
​

Fraud Color Assignment:

The fraud decision color for this feature is fully configurable. Thresholds can be customized for each signal by either:

fraudulent pdf

This feature performs comprehensive analysis on PDF documents to detect signs of manipulation or fraudulent activity. It includes three specialized detection mechanisms:

Fraud Color Assignment

The fraud decision color for this feature is fully configurable. Thresholds can be customized for each signal by either:

duplicate

is_duplicate

This feature identifies when a submitted document is likely a duplicate of a previously submitted one. The detection algorithm employs straightforward matching of extracted values to identify repeated submissions of the same receipt.

The system utilizes the document-level field is_duplicate to flag instances where key receipt data points (such as vendor, date, total amount, etc) match those of previously processed documents.

This helps prevent fraud schemes involving multiple submissions of the same receipt for additional reimbursements or rewards.
​

Fraud Color Assignment:

The fraud decision color for this feature is fully configurable. Thresholds can be customized for each signal by either:

Learn more about Duplicate vs Similar documents

similar documents

meta.duplicates

This feature identifies when a submitted document's text has high similarity with previously submitted documents, even if they aren't exact duplicates.

The system performs comprehensive text comparison analysis and returns a list of similar documents in the meta.duplicates field, complete with similarity scores. This allows for detection of slightly modified versions of the same receipt or documents that contain highly similar content but with minor alterations/differencies.

Similarity thresholds are fully configurable, allowing businesses to adjust sensitivity based on their specific needs and fraud patterns.

Fraud Color Assignment:

The fraud decision color for this feature is fully configurable. Thresholds can be customized for each signal by either:

warnings

Veryfi Warnings system works alongside the Fraud Detection Suite to identify calculation inconsistencies and logical discrepancies that may indicate potential issues. Unlike fraud signals that detect deliberate manipulation, warnings highlight natural anomalies such as mismatched line item totals, duplicate entries, or tax calculation errors that might occur in legitimate receipts but still warrant verification.

This additional layer of scrutiny helps organizations maintain data integrity even when submissions pass fraud detection, ensuring complete confidence in extracted financial information.

line_item_amount_missmatch

Line item totals x quantity do not add up to the subtotal value.

line_item_repeats

When a line item has the same description or SKU with the same quantity but a different total.

tax_rate_missmatch

When Tax Rate x Subtotal != Tax Amount

item_counts_missmatch

When number of products on line items doesn't match Number of Items Sold value

totals_missmatch

When subtotal from line items doesn't match subtotal on the document

barcode_decoding_issue

When barcode is present on the document but is not decodable

logo_vendor_missmatch

When vendor name from logo doesn't match vendor name output from model

barcode_code_missing_in_ocr

Decoded Barcode numbers were not found on the document

malware

Supplied pdf file contains executable javascript code or eicar malware