Signals reference
Every signal, its layer, severity, and weight.
Provenance
| Signal | Severity | Weight | Meaning |
|---|---|---|---|
| c2pa_ai_generated | critical | 95 | The file carries a C2PA manifest stating it was produced by a generative AI tool. |
| c2pa_invalid_signature | high | 60 | The file has a C2PA manifest, but its signature or content hashes do not validate, so the file changed after it was signed. |
| metadata_ai_generated | critical | 90 | IPTC DigitalSourceType or a software tag says the image came from a generative AI tool. |
| exif_editing_software | medium | 35 | Image metadata shows the file was saved by an editor such as Photoshop, GIMP, or Canva. |
| metadata_stripped | low | 10 | The file has no creation metadata at all, which is unusual for exported statements and invoices. Weak on its own. |
File structure
| Signal | Severity | Weight | Meaning |
|---|---|---|---|
| pdf_producer_mismatch | high | 55 | The PDF says it was created by one system but last written by a consumer PDF editor. |
| pdf_modified_after_creation | medium | 35 | The PDF modification date is well after its creation date. |
| pdf_incremental_updates | medium | 30 | The PDF contains later revisions that change content, not just signatures or validation data. |
| pdf_font_inconsistency | high | 55 | A value on the page uses a font or font embedding that appears nowhere else on the page. |
| pdf_text_overlay | high | 65 | A shape or annotation is drawn over existing text, or text is drawn on top of other text. |
| pdf_text_layer_mismatch | high | 60 | What the page shows differs from the text stored in the PDF. |
| pdf_anachronistic_producer | high | 60 | The software that wrote the PDF was released after the date the document claims. |
| pdf_signature_invalid | high | 60 | The PDF is signed, but the signature does not validate or content changed after signing. |
| pdf_signed_valid | info | -20 | The PDF is signed and nothing changed after signing. This lowers the score. It does not prove who signed it. |
| pdf_active_content | low | 15 | The PDF contains JavaScript, embedded files, or launch actions. |
Pixels
| Signal | Severity | Weight | Meaning |
|---|---|---|---|
| image_ela_anomaly | medium | 35 | Error level analysis shows a region that was compressed differently from the rest of the image. |
| image_copy_move | high | 55 | Part of the image appears to be copied from another part of the same image. |
| image_noise_inconsistency | medium | 30 | A region has a different noise pattern than the rest of the image. |
| image_double_compression | low | 15 | The JPEG shows signs of being decoded, edited, and saved again. |
| ai_image_classifier | medium | 40 | A third-party classifier scored the image as likely AI-generated. Never decides a verdict alone. |
Content logic
| Signal | Severity | Weight | Meaning |
|---|---|---|---|
| math_line_items_mismatch | high | 60 | The line items do not sum to the printed subtotal. |
| math_total_mismatch | high | 65 | Subtotal plus tax and tip does not equal the printed total. |
| math_running_balance_mismatch | high | 70 | A bank statement balance does not follow from the previous balance and the transaction. |
| paystub_math_mismatch | high | 65 | Gross pay minus deductions does not equal net pay, or year-to-date values are inconsistent. |
| paystub_tax_rate_anomaly | medium | 40 | Social Security or Medicare withholding does not match the statutory rate. |
| tax_rate_implausible | medium | 35 | The effective sales tax or VAT rate is outside the normal range for the country. |
| date_invalid | high | 50 | A date is in the future, does not exist, disagrees with the printed weekday, or the period makes no sense. |
| format_anomaly | medium | 35 | A misspelled card brand, an impossible card number fragment, or a routing number that fails its checksum. |
| internal_inconsistency | high | 55 | The same value is printed more than once with different amounts. |
| content_plausibility_note | low | 10 | A specific inconsistency noticed during review of the extracted fields. Low weight. |
History
| Signal | Severity | Weight | Meaning |
|---|---|---|---|
| duplicate_submission | high | 50 | The exact same file was checked before in this organization. |