Understanding signals
The Detect signals in Ocrolus provide insights into potential fraud by highlighting indicators that may suggest tampering or manipulation within documents.
Detect runs three layers. Ocrolus checks that the math adds up (balances, totals, and dates) on supported document types. Resistant AI forensic analysis checks file metadata and the pixels of the page, including signs of AI generation, on any document type. Supported documents get all three layers. Serial fraud and template reuse detection is coming in early 2027.
These signals are automatically flagged during document processing and allow you to quickly identify suspicious activity. For example, the system may detect edits to account numbers, amounts, document origins, metadata anomalies, AI-generated artifacts, or screenshot-based submissions. Each signal is visualized to help you review highlighted areas that may require further investigation.
In addition to Dashboard, you can also use Detect via our API. To know more, refer to the following pages:
These pages explain the same functionality as this guide but in a format tailored for integration into your workflows. If you need further assistance, feel free to contact our support team at [email protected].
Fraud signals categories
We broadly classify fraud signals in one of two ways:
Resistant AI metadata and forensic signals
Resistant AI metadata and forensic signals are powered by Resistant Documents, their flagship document fraud detection software that evaluates the structural integrity and origin of a document through deep forensic inspection.
These signals analyze properties that are typically not visible during standard document review, such as how the document was created, whether it has been digitally modified, and whether its metadata aligns with what is expected from a legitimate issuer.
By inspecting documents at a forensic level, these signals help determine whether a document can be trusted or requires further verification. Most forensic signals are displayed as colored overlays on the document pages, making it easy to visually identify areas of concern. For more information, see Reviewing Detect signals and Interpreting Visualization.
Ocrolus algorithmic signals
Ocrolus algorithmic signals are generated by proprietary fraud detection algorithms and focus on validating the accuracy and consistency of data captured from a document. Instead of analyzing how a document was created, these signals evaluate whether numbers, dates, and calculations are internally consistent and align with expected rules.
These signals help identify cases where values may have been altered in ways that break internal consistency, even when the document appears visually untampered.
Reviewing Detect signals
This section explains how the information is displayed in the Ocrolus Dashboard after a document has finished processing, and guides you on how to access and review Detect signals.
Book overview
Detect signals will be available to review in the Dashboard on the Book Overview screen. Books with Detect signals for review will display a red flag in the Status column.
The Authenticity Score column displays the lowest Authenticity Score for any document within a Book. This will help you identify potential fraud more efficiently. This column is sortable and allows you to prioritize your workflow by focusing on documents with low scores, which may indicate severe fraud and require immediate attention, or high scores, which are likely authentic and can be approved with minimal review.
To access the Book Overview page, select the Book you want to review from the Book List page. All the Books are displayed on the Ocrolus Dashboard, sorted by the Authenticity Score column, with lower scores appearing at the top. Books processed before the Authenticity Score release on November 15 will not have an associated score and will be listed at the bottom, regardless of the sorting direction. This allows you to easily focus on more critical cases by directing attention to documents with lower Authenticity Scores that may indicate fraud.

Understanding signal icons
The following five icons indicate the status of a document:
- Green check circle: The document was successfully uploaded and processed. No fraud signals were detected, or the document received an Authenticity Score above 80, indicating high authenticity.
- Red flag: The document has been processed and Detect has identified fraud signals. The tooltip displays Detect signals found.
- Green partial check circle: The capture process has been completed successfully, but the Detect process is still running.
- Gray check circle with hourglass: The document is still being processed. Please check back later.
- Red tirangle with exclamation mark: The document was rejected. The tooltip displays the rejection reason such as Rejected: instant not supported.

The Book Overview page displays Detect signals at the Book and Document level. See below for details about the information contained in each panel on the Book Overview page.
The panels on the Dashboard are as follows:
Book-level signals
It displays the total number of Detect signals at the Book level, representing all signals detected across all documents within the Book combined. This makes it clear and consistent for users to understand the significance of Detect signals at the Book level.
Uploads
It displays the Documents within the Book and their respective statuses. To learn more about signal icons shown on UI, see the Understanding signal icons section.

A red flag appears when Detect identifies risk signals and the document’s Authenticity Score is below 80. A green check circle indicates that no fraud signals were detected, or the document received an Authenticity Score above 80, indicating high authenticity. Use the checkbox on the left to open a document and review its associated signals.
Documents
It displays the Documents that are contained within whichever file is selected in Uploads. If multiple documents are selected, they will be grouped by type. If multiple bank statements are selected, they will be grouped by bank account. The Authenticity column shows the Authenticity Status and Authenticity Score for each document. If there are any Detect signals for a document, click on it to review them.

Preview
It displays the selected Document and the data captured from it. Click on the View Details link to view details about the fraud signals, if any.

Detect
Clicking on the Preview link will reveal details and visualizations for each Detect signal.

Don't see anything?If you can't find this part of the Dashboard, you may need to get Detect enabled for your organization. Please contact our support team at [email protected] or your account manager.
The panels are as follows:
Book-level signals
This is the same panel that appears in the Book Overview.
Visualizations
The Visualizations panel displays specific regions of the selected Document that were tampered with. In some cases, multiple overlapping signals may be shown. In this case, you can view them individually by selecting the thumbnails in the bottom-right corner.
To know more about visualizations that may be presented, see Interpreting Visualizations.
Details
This panel displays Authenticity Score and Authenticity Status given to the document and highlights key observations related to the document. It also provides details for each Detect signal, grouped by type. You can expand each signal group to view additional information.
Each signal includes an indicator:
- A red dot indicates a risk signal. Hover over it to view the label as Risk.
- A green dot indicates a trust signal identified for the document.
- A blue dot indicates an informational signal.
You can use the Expand All option to view details for all signal types at once.

Different signals have different information associated with them. To know more about additional guidance on interpreting them, see Interpreting Detect signals.
You can also view the data captured from this document by selecting the Capture tab.

To return to the Book Overview page, click the X on this panel or the arrow above the Book-Level Signals panel.
Interpreting Authenticity Score
In addition to detailed fraud signals which indicate the context and method of tampering, we provide a single Authenticity Score indicating the likelihood that a document is authentic.
The Authenticity Score ranges from 0-100. The score weighs the context of what was tampered with and our confidence in the signal.
We categorize scores as follows:
- 0-55: LOW authenticity
- 56-80: MEDIUM authenticity
- 81-100: HIGH authenticity
To learn more about how our score is determined and how to choose a threshold for your workflow, see the Authenticity Score page.
In addition to a numerical score, high-level observations provide transparency into the assessment by summarizing key findings in the document.
Detect signals are generated from multiple sources, including metadata-based and algorithmic signals. These signals also serve as high-level indicators that can help determine the document’s review flow.

Interpreting Detect signals
Detect signals provide a short description of what has been uncovered and additional information to help contextualize the signal. File tampering signals are also visualized as highlighted regions on the document such as red boxes around affected fields. To know more about visualizations, see Interpreting visualizations.
Resistant AI metadata and forensic indicators
These indicators are powered by Resistant AI and apply to any document type. They detect hidden manipulation, document origin issues, and structural anomalies that are typically not visible during standard review.
| Signal | Meaning |
|---|---|
anomalous_pdf_structure | Documents from this issuer typically follow a consistent structure. This one differs in layout, metadata, or file structure, suggesting it may have been edited, recreated, or saved using different software. |
suspicious_producer | The document was produced or modified using tools that are unusual for the identified issuer, suggesting possible reprocessing or tampering. |
overlapping_content | Content is layered over existing elements, such as text or images placed on top of backgrounds or scanned pages, potentially obscuring original content. |
anomalous_font | Fonts used in the document are not typically associated with this issuer or document type. |
digital_print | The document appears to be exported from a digital source, such as a PDF converted into an image, which may conceal editing traces or remove metadata. |
unusual_pdf_structure | Structural elements such as metadata, layout, fonts, or generation patterns do not match known issuer formats or standard document structures. |
content_tampering_suspected | Visual inconsistencies suggest that specific characters, fields, or regions may have been digitally altered. |
image_of_known_document | The document is submitted as an image of a type typically available as a native PDF, limiting access to metadata and structural details needed for verification. |
document_from_online_generator | The document originates from an online generation service, offering limited insight into content creation and requiring additional scrutiny. |
editing_software_detected | The document was modified or saved using third-party editing or design software, which may indicate tampering, especially when combined with other anomalies. |
content_diff_between_versions | Differences detected between versions of the same document indicate changes to content, which may suggest fraud. |
has_only_images | The document contains only images, possibly due to scanning or conversion, limiting the ability to extract and verify text. |
has_piece_info | Variable data elements detected in the document structure suggest that certain applications were used to create or modify the document. |
expected_pdf_structure | The document structure aligns with known authentic patterns for the document type. |
unmodified_per_metadata | Metadata timestamps show no evidence of incremental updates or modifications after creation. |
screenshot | The document appears to be a screenshot or screen capture, which may obscure editing artifacts or metadata. |
NoteThe list above represents the most common Resistant AI indicators and is not exhaustive. Resistant AI indicators are continuously updated as part of live fraud detection, and new indicators may be added over time.
Ocrolus algorithmic signals
These signals are generated by Ocrolus' proprietary fraud detection algorithms. They validate the accuracy and consistency of the data captured from a document and are specific to each document type. Signals listed here come from Ocrolus math checks on supported documents. Resistant AI signals (metadata and pixels) apply to any document type.
Each check runs only on the fields it depends on. If a field isn’t captured from the document, the check doesn’t flag.
Bank statement signals
| Signal | Meaning |
|---|---|
| Unreconciled Bank Statement Balance | The bank statement's beginning and ending balances do not reconcile based on the transactions for the given period. |
| Invalid Transaction Date | The transaction date does not fall within the statement period. |
| Date in Future | The captured date is in the future. |
| Year in Future | The captured year is in the future. |
Pay stub signals
| Signal | Meaning |
|---|---|
| Online Generated Paystub | The pay stub’s layout matches a pay stub produced by a known online generator. |
| Unreconciled Gross Pay | Gross pay does not match the sum of the earnings lines. |
| Unreconciled Hourly Current Pay | Total hourly current pay does not match the hourly lines it should add up from. |
| Unreconciled FICA Social Security Tax Amount | Social Security tax withheld does not match the expected 6.2% of gross pay. |
| Unreconciled FICA Medicare Tax Amount | Medicare tax withheld is less than the expected 1.45% of gross pay. |
| Missing FICA Tax Amount | No Social Security or Medicare tax amount appears on the pay stub. |
| Missing Federal Income Tax | No federal income tax amount appears on the pay stub. |
| Missing State Income Tax | No state income tax amount appears on the pay stub. |
| No State Level Withholding | The pay stub has no state-level deduction for the employee’s home state. |
| Invalid State Income Tax Description | The state tax deduction line does not name a state. |
| Date in Future | A date on the pay stub is in the future. |
| Invalid Date | A date on the pay stub is not a valid date. |
W-2 signals
| Signal | Meaning |
|---|---|
| Invalid Year | The tax year on the W-2 is not a valid year. |
| Year in Future | The tax year on the W-2 is in the future. |
| Social Security Wage Base Exceeds Annual Limit | Social Security wage base exceeds the limit for the year. |
| Unreconciled Social Security Tax Withholding | Social Security tax withheld does not match the expected 6.2% of the wage base. |
| Unreconciled Medicare Tax Withholding | Medicare tax withheld does not reconcile with the expected amount. |
| Invalid Medicare Wages and Tips | Medicare wages and tips are below expected amounts based on related fields. |
| Invalid Medicare Wages | Medicare wages are below expected amounts based on related fields. |
| Invalid Statutory Employee Federal Tax | A statutory employee should not have reportable federal income tax. |
| Invalid State Taxes Paid | State income tax is present for a state that does not tax wages. |
| State Taxes Not Paid | No state income tax is withheld for a state that does tax wages. |
| Unreconciled State Tax Base | State income tax paid does not reconcile with the wage base. |
| Missing Medicare Wage Base | The Medicare wage base is missing. |
| Missing Social Security Wage Base | The Social Security wage base is missing. |
Utility bill signals
| Signal | Meaning |
|---|---|
| Billing Date in Future | The billing date on the bill is later than the current date. |
| Invalid Service Period | The service period is not a valid date range. The start date falls on or after the end date, or one of the dates cannot be read as a date. |
| Due Date Before Billing Date | The payment due date falls before the billing date. |
1040 and 1040-SR
| Signal | Meaning |
|---|---|
| AGI Reconciliation Failure | Adjusted Gross Income (line 11A) should equal line 9 minus line 10. |
| Taxable Income Reconciliation Failure | Taxable income does not reconcile with reported income and deductions. |
| Tax Liability Mismatch | Reported tax liability does not match expected tax for the reported income. |
| Refund Reconciliation Failure | Reported refund does not reconcile with payments and total tax. |
| Tax Year in Future | Reported tax year is in the future. |
| Invalid Tax Year | Reported tax year is outside the valid range. |
1040 Schedule C
| Signal | Meaning |
|---|---|
| Unreconciled Gross Income | Gross income does not reconcile across lines 1 to 7. |
| COGS Does Not Reconcile | Cost of goods sold does not reconcile across Part IV lines 35 to 42. |
| Unreconciled Schedule C Net Profit | Net profit does not equal tentative profit minus home office expenses. |
| Home Office Deduction Exceeds Net Profit | Home office deduction (line 30) exceeds tentative profit (line 29). |
1040 Schedule E
| Signal | Meaning |
|---|---|
| Rental Income Does Not Reconcile | Rental income (line 21) does not equal rents (line 3) plus royalties (line 4) minus total expenses (line 20). |
1040 Schedule F
| Signal | Meaning |
|---|---|
| Net Farm Profit Does Not Reconcile | Net farm profit or loss (line 34) does not equal gross income (line 9) minus total expenses (line 33). |
| Cooperative Distribution Inconsistency | Taxable amount of cooperative distributions (line 3b) exceeds total cooperative distributions received (line 3a). |
1065
| Signal | Meaning |
|---|---|
| Ordinary Business Income Does Not Reconcile | Ordinary business income (line 23) does not equal total income (line 8) minus total deductions (line 22). |
1065 Schedules L, M-1, and M-2
| Signal | Meaning |
|---|---|
| Balance Sheet Total Assets Do Not Match | Schedule L total assets (line 14) do not equal total liabilities and capital (line 22) for the end-of-year column. |
| Book Income Does Not Reconcile to Taxable Income | Schedule M-1 income (line 9) does not equal net income per books plus additions (lines 2 to 4) minus subtractions (lines 6 to 7). |
| Partners’ Capital Account Does Not Reconcile | Schedule M-2 end-of-year balance (line 9) does not equal beginning balance plus capital contributed, net income and other increases, minus distributions and other decreases. |
1120
| Signal | Meaning |
|---|---|
| Taxable Income Does Not Reconcile | Taxable income (line 30) does not equal total income (line 11) minus total deductions (line 27) minus the NOL and special deductions (line 29c). |
1120 Schedules L, M-1, and M-2
| Signal | Meaning |
|---|---|
| Balance Sheet Does Not Balance | Schedule L total assets (line 15) do not equal total liabilities and shareholders’ equity (line 28) for the end-of-year column. |
| Book-to-Tax Reconciliation Does Not Reconcile | Schedule M-1 income (line 10) does not equal net income per books plus additions (lines 1 to 5) minus subtractions (lines 7 to 8). |
| Retained Earnings Do Not Reconcile | Schedule M-2 end-of-year balance (line 8) does not equal beginning balance plus net income and other increases, minus distributions and other decreases. |
1120S
| Signal | Meaning |
|---|---|
| Ordinary Business Income Does Not Reconcile | Ordinary business income (line 22) does not equal total income (line 6) minus total deductions (line 21). |
1120S Schedule M-2
| Signal | Meaning |
|---|---|
| Accumulated Adjustments Account Does Not Reconcile | Schedule M-2 end-of-year AAA balance (line 8, column a) does not equal the sum of lines 1 to 5 minus distributions (line 7). |
4835
| Signal | Meaning |
|---|---|
| Net Farm Rental Income Does Not Reconcile | Net farm rental income or loss (line 32) does not equal gross farm rental income (line 7) minus total expenses (line 31). |
940
| Signal | Meaning |
|---|---|
| Taxable FUTA Wages Do Not Reconcile | Total taxable FUTA wages (line 7) do not equal total payments to all employees (line 3) minus the subtotal of exempt and excess payments (line 6). |
| FUTA Tax Does Not Reconcile | FUTA tax before adjustments (line 8) does not equal 0.6% of total taxable FUTA wages (line 7), or total FUTA tax after adjustments (line 12) does not equal lines 8 to 11. |
941
| Signal | Meaning |
|---|---|
| Social Security Tax Does Not Reconcile | Social Security tax (line 5a, column 2) does not equal 12.4% of taxable Social Security wages (line 5a, column 1). |
| Medicare Tax Does Not Reconcile | Medicare tax (line 5c, column 2) does not equal 2.9% of taxable Medicare wages and tips (line 5c, column 1). |
| Total Tax Liability Does Not Reconcile | Total taxes do not reconcile: line 6 does not equal lines 3 plus 5e plus 5f, line 10 does not equal lines 6 to 9, or line 12 does not equal line 10 minus line 11g. |
| Deposits Do Not Reconcile with Tax Liability | Total taxes after adjustments and credits (line 12) do not equal total deposits (line 13) plus balance due (line 14) minus overpayment (line 15). |
2555
| Signal | Meaning |
|---|---|
| Foreign Earned Income Exclusion Exceeds IRS Annual Limit | Foreign earned income exclusion (line 42) exceeds the prorated annual limit, line 41 multiplied by line 38 divided by 365. |
| Foreign Housing Exclusion Does Not Reconcile | Foreign housing exclusion does not reconcile: line 33 must equal line 30 minus line 32, and line 36 must equal line 33 times line 35. |
4562
| Signal | Meaning |
|---|---|
| Section 179 Deduction Exceeds IRS Annual Limit | Section 179 expense deduction (line 12) exceeds the maximum amount (line 1). |
4797
| Signal | Meaning |
|---|---|
| Gain or Loss Does Not Reconcile | Total gain (line 24) or adjusted basis (line 23) does not reconcile with the reported sales price, cost basis, and depreciation lines for a property. |
| Section 1245 Recapture Amount Inconsistent | Section 1245 recapture (line 25b) exceeds depreciation allowed (line 25a) or total gain (line 24) for a property. |
6252
| Signal | Meaning |
|---|---|
| Gross Profit Percentage Does Not Reconcile | Gross profit percentage (line 19) does not equal gross profit (line 16) divided by contract price (line 18). |
| Installment Sale Income Does Not Reconcile | Installment sale income (line 24) does not equal payments received (line 22) multiplied by gross profit percentage (line 19). |
8825
| Signal | Meaning |
|---|---|
| Net Rental Income on 8825 Does Not Reconcile | Net income or loss from rental real estate (line 17) does not equal gross rents (line 2) minus total expenses (line 16) for one or more properties. |
1099-K
| Signal | Meaning |
|---|---|
| Transaction Count Is Implausible for Declared Volume | Gross amount (Box 1a) and number of transactions (Box 3) are inconsistent: one is zero while the other is positive, or the amount per transaction is implausibly low. |
1099-MISC
| Signal | Meaning |
|---|---|
| 1099-MISC State Tax Withheld Exceeds State Income | State tax withheld (Box 16) exceeds state income (Box 18) for one or more state rows. |
1099-R
| Signal | Meaning |
|---|---|
| Distribution Code Inconsistent with Taxable Amount | Box 7 distribution code is H or Q while Box 2a taxable amount is greater than 0. |
SSA-1099
| Signal | Meaning |
|---|---|
| SSA-1099 Net Benefit Does Not Reconcile | Net benefits (Box 5) do not equal benefits paid (Box 3) minus benefits repaid (Box 4). |
W9
| Signal | Meaning |
|---|---|
| W-9 TIN Format Does Not Match Entity Type | Federal tax classification is a corporation or partnership, but a Social Security Number is populated and the Employer Identification Number is empty. |
Interpreting visualizations
Detect highlights specific regions of the document where anomalies or potential tampering are detected. These areas are marked with red boxes, making it easy to identify fields or content that may require further review.
The name of the current visualization type is shown at the bottom of the visualization panel. You can hover over the grey question mark to view additional details about the visualization.
The visualization panel on the left displays the document pages with highlighted areas. Each red box corresponds to one or more signals listed in the Detect panel on the right. You can click on individual signals in the Detect panel to understand what was detected in each highlighted area.
Signals from both Resistant AI document fraud detection, such as content manipulation, suspicious edits, and unusual PDF structure, and Ocrolus algorithmic checks, such as invalid Social Security taxes or Medicare taxes, are displayed together in a single panel to provide a unified view of detected anomalies.

Updated 11 days ago