Abstract

Edited PDF bank statements are a common vector for loan origination fraud. No single signal reliably detects manipulation. This note describes a layered approach that combines independent weak signals into a robust indicator.

Signal families

  • Metadata: producer, creation and modification history
  • Structure: object streams, incremental updates and font subsets
  • Typography: glyph metrics and spacing inconsistencies
  • Arithmetic: running balances that do not reconcile

Recommendation

Prefer source-verified data, such as consented Account Aggregator data, wherever available, and apply forensics to documents that must still be uploaded.

The full research is available on request. We will send it to your work email.

Request this research