Bank statements are messy (scanned, skewed), and proving "Chain of Custody" is critical for legal admissibility.
A Django app that ingests PDFs or shots and uses Mistral OCR-3 to parse transactions (way better reliability than Tesseract for tables), and runs the LIBR algorithm to identify exactly which dollars belong to whom.
Recent Updates (Based on HN Feedback -- https://news.ycombinator.com/item?id=46350044): I just refactored the backend to support a strict Chain of Custody:
Immutable Reports: Reports are now snapshot-locked upon generation.
Cryptographic Integrity: Every report generates a SHA-256 hash of the raw calculation vector + the final PDF. This means you can mathematically prove the report hasn't been tampered with since generation.
LIBR Verification: Added a regression suite to handle edge cases like "zero-balance dips," ensuring the tracing algorithm adheres to the See v. See (1966) legal standard.
The code isn't fully open-source yet, but it's free to use right now. I'd love feedback on the PDF generation logic or the OCR accuracy!