Document Provenance & Authenticity Check

The verify-document-provenance pipeline assesses whether a PDF document is what it claims to be. It examines the embedded metadata for signs that the document was created or modified by different software than claimed, reviews the visual layout and typographic elements for signs of digital editing or forgery, and checks for the presence or absence of digital signatures and their validity. This pipeline is designed for use cases such as HR document screening, legal document intake, procurement fraud detection, and any workflow where you regularly receive PDFs that need a fast authenticity check before further processing.

Forged documents — altered invoices, fraudulent employment certificates, manipulated contracts — rarely change just the visible text. Editing software leaves traces in the PDF's internal structure, font embedding, object modification history, and creation metadata. This pipeline surfaces those traces systematically.

Pipeline Details

Property	Value
Pipeline ID	`verify-document-provenance`
Tier	1 — AI-Only
Estimated Duration	~1–3 min
Credit Cost	600 credits per run
Crowd Stages	No
Files Required	Yes — enriched PDF

Required Inputs

Files

One enriched PDF file. The enrichment step extracts metadata.pageCount, metadata.author, metadata.creationDate, and metadata.wordCount. The file must have enrichmentStatus: "completed" before the run is created. See Files & Enrichment for details.

Context Keys

All context keys are optional. Providing them allows the AI to cross-reference the document's claimed identity against its embedded metadata and visual content.

Key	Required	Description
`issuing_authority`	Optional	Organisation claimed to have issued the document (e.g. `"Federal Labour Office"`, `"City of Munich"`). Cross-referenced against embedded author, producer, and signature metadata.
`document_type`	Optional	Type of document, e.g. `"invoice"`, `"employment certificate"`, `"passport copy"`, `"contract"`. Used to apply document-type-specific layout and formatting expectations.
`expected_date`	Optional	Expected issuance or validity date in ISO 8601 format (e.g. `"2023-12-01"`). Compared against the PDF creation date and any dates visible in the document body.

Stages

Metadata Forensics (AI)

Inspects the full PDF metadata block: author, creator application, producer (the software that generated the final PDF output), creation date, and modification date. The AI checks for inconsistencies between these fields — for example, a document claiming to have been issued by a government office but showing a consumer PDF editor as the creator tool, or a modification timestamp that post-dates the claimed issuance date by months. It also reviews the document's internal object modification history, which in PDF format records each time a page, image, or form field was altered.

Content Analysis (AI)

Reviews the document's visual layout, typography, and embedded objects for signs of digital editing. The AI examines font embedding patterns (documents modified post-creation often contain fonts from multiple different source applications), checks whether text rendering is consistent across the page, assesses whether any stamps, signatures, logos, or seals are rasterised overlays (which may indicate they were placed on top of an existing document rather than generated with it), and looks for copy-paste artefacts in text layers that suggest content was imported from a different source.

Verdict Synthesis (AI)

Aggregates the provenance and content signals into a final verdict. The AI distinguishes between minor formatting inconsistencies that are common in legitimate documents and substantive forensic indicators of forgery, and calibrates the confidence score accordingly. The verdict is accompanied by a list of the specific signals that were found and a plain-language explanation of their significance.

Starting a Run

curl -X POST https://api.crowdee.ai/v2/projects/{projectId}/verification-runs \
  -H "X-API-Key: crw_YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "pipeline_id": "verify-document-provenance",
    "file_ids": ["{fileId}"],
    "context": {
      "issuing_authority": "Federal Labour Office",
      "document_type": "employment certificate",
      "expected_date": "2023-12-01"
    }
  }'

Example Response

{
  "id": "run_01j9x...",
  "status": "completed",
  "verdict": "manipulated",
  "confidence": 94,
  "pipeline_id": "verify-document-provenance",
  "scorecard": {
    "metadata_consistency": "fail",
    "creator_tool_plausibility": "fail",
    "content_layout_integrity": "warn"
  },
  "explanation": "PDF creator tool is 'Adobe Acrobat 2019' but the document claims issuance by a government authority whose standard document production system generates PDFs via a different toolchain. Modification date is 14 months after the claimed issuance date. A rasterised text overlay was detected on page 2, consistent with a salary figure having been replaced after original document generation."
}

For interpreting verdict values (authentic, manipulated, inconclusive, unverified) and understanding how confidence scores are calculated, see Verdicts & Confidence.