Skip to content
14 - Finance

Document Trust

Recipient
Fintech lender
Channel
Back-office
Status
Open demo
Type
Agentic
Languages
FR
Sector
Finance
Actions and integrations
Preserves and fingerprints original files · Analyses PDF structure, signatures, images and provenance · Localises anomalies and prepares human review

A document analysis platform combining deterministic checks, cryptographic verification and specialised detectors, with traceable findings for analyst review

Document Trust runs a reproducible control pipeline on every submitted file. It preserves and fingerprints the original, validates the actual format, then analyses PDF structure, revisions, signatures, graphic objects and available provenance. A versioned policy aggregates the findings, records checks that could not be completed and produces a traceable analyst-review file.

Control matrix

Four control levels, from file to reference source

Each level produces a separate result: file integrity, structural analysis, cryptographic evidence, then comparison with configured business rules or approved references.

01

File integrity

Checks the declared format against the file bytes, records a SHA-256 fingerprint, preserves the original and detects exact duplicates.

02

Structure and edits

Examines pages, revisions, fonts, images, metadata, text collisions, graphic overlays and elements added after a signature.

03

Signature and provenance

Verifies cryptographic integrity and byte coverage when a signature exists, and checks C2PA Content Credentials when available.

04

Context and source

Compares document rules, issuer references or approved official sources only when those connections are configured and authorised. Otherwise the result remains inconclusive.

Input scope

File formats and document profiles

The intake layer accepts native and scanned PDFs plus common raster formats. Document profiles, required fields and validation rules are configured per workflow.

  • 01Native and scanned PDF documents
  • 02PNG, JPEG, TIFF and WebP images
  • 03Payslips and bank statements
  • 04Identity documents and proof of address
  • 05Tax documents and company records
  • 06Other project-specific supporting documents

Accepting a file format never means declaring the document authentic. Issuer references, business rules and source connections are validated separately for every deployment.

Detection catalogue

Technical signals and covered tampering scenarios

Each signal is localised in the file, assigned an explicit control result and presented to the analyst with its technical context. The platform does not make a legal finding of fraud.

  • 01Text, dates or amounts added as overlays in a native PDF
  • 02Local font, spacing or layout inconsistencies
  • 03Graphic objects added to an otherwise flattened document
  • 04Revisions or modifications made after signing
  • 05A visible signature without matching cryptographic evidence
  • 06Broken digital-signature integrity or incomplete byte coverage
  • 07An exact duplicate already submitted
  • 08A mismatch between the declared format and the actual file
  • 09A scan, screenshot or image whose quality prevents a reliable check
  • 10Invalid, unapproved or AI-editing provenance indicated by C2PA or weak declarative metadata

AI-generation and editing traces: Document Trust can validate C2PA Content Credentials and report declared AI creation or editing actions. The absence of a credential does not prove that AI was not used, and declarative metadata alone remains a weak signal.

Analysis architecture

Modular forensic pipeline

Specialised engines execute their controls independently and return structured findings. The policy engine aggregates those results without masking disagreement, missing evidence or unavailable checks.

A

Admission and fingerprinting

Validates the file envelope, preserves the original and records its fingerprint.

B

PDF structural analysis

Inspects objects, revisions, fonts, overlays, signatures and rendering order.

C

Image analysis

Assesses raster quality and technical image signals without inferring missing evidence.

D

Signature validation

Separates a visual signature from verifiable integrity and measures what bytes are covered.

E

C2PA provenance

Reads and validates available credentials and distinguishes verified provenance from declared metadata.

F

Policy engine and analyst console

Combines explained findings, records unavailable checks and routes the file to accountable human review.

Decision policy

Evidence limits and analyst escalation

A result without an alert is not an authenticity certificate. Manipulations that leave no conclusive trace in the submitted file require an original, an issuer reference or another authorised source.

  • 01A retouch already flattened into an image
  • 02A cleanly regenerated business value or fully synthetic document
  • 03The official origin of a document without an external reference
  • 04Fraud intent or a credit decision

When the evidence is missing, conflicting or outside the configured scope, the system returns an inconclusive result and keeps the decision with the analyst.

Capabilities
  • Native and scanned PDF, PNG, JPEG, TIFF and WebP
  • File integrity, PDF structure, signature and provenance checks
  • Localised and explained findings
Guardrails
  • No automated fraud or credit decision
  • Inconclusive results routed to human review
  • External source checks only when authorised and configured
Next solution
VBN →
Call our Voicebots