Forensic authenticity analysis

Know whether a document is what it claims to be.

DFDS examines invoices, tenders and certificates for tampering, pasted-over edits, AI generation and financial inconsistency — then explains the evidence in a report an officer can act on and defend.

Access is issued by an administrator. Contact yours to be invited.

TENDER-4728743.pdfSuspicious · 62
Pasted patch

What DFDS looks for

Deterministic forensics first, AI second — so a finding can be measured, pointed at and defended.

Pasted & painted-over edits

Pixel-exact detection of rectangles pasted over an original — including the kind whose colour differs by 4 parts in 765 and is invisible to the eye.

Structural tampering

Fonts a document never otherwise uses, revisions made after it was signed, and metadata that disagrees with the content.

AI-generated documents

Synthetic and re-rendered pages, flagged from generator fingerprints and visual review rather than a hunch.

Financial inconsistency

Totals that don't follow from their line items, tax that doesn't compute, and summary figures that contradict the detail rows.

How it works

Three steps — and a person decides at the end of them.

Bring the document in

Upload it, or let a scheduled workflow collect it from Azure Blob, S3, FTP or SFTP the moment it lands.

Examine it

Metadata, per-page pixel forensics, structural checks and an AI review run together — every page, every embedded image.

Decide on the evidence

A risk score with findings, annotated evidence images and a downloadable report — routed for human validation before anything is accepted.

Built to be challenged

A fraud finding is worth exactly as much as the evidence behind it.

  • Every finding cites measured evidence — coordinates, colours and deltas, not adjectives.
  • Deterministic detectors run independently of the AI, so a verdict never rests on a model's opinion alone.
  • Annotated evidence images show exactly where, and a colour map makes an invisible edit visible.
  • Findings are decision support. A named reviewer validates before anything is accepted.
As captured2
Colour map2

Red = pixels exactly matching the pasted patch. Green = the surrounding element. They differ by 4/765.