cd_mkdir

See A Sample Attorney-Review LIBR Workpaper: https://exitprotocols.com/static/documents/Exit_Protocol_Syn...

Honestly never subscribed to Claude....GPT 5.5 and Grok only.

Anyways leave the CSS aside happy to answer about Workpaper it generates: https://exitprotocols.com/static/documents/Exit_Protocol_Syn...

Not a lawyer, so the Go-To-Market side in the legal space has been a steep learning curve. If anyone here has experience selling/deploying air-gapped, on-prem solutions to highly risk-averse, non-technical clients (like law firms), I would love to hear your battle stories.

Happy to answer any questions about the math, the OCR pipeline, or the architecture!

Sandbox link again: https://exitprotocols.com/simulation/uplink/

Problem was Implementing the 'Lowest Intermediate Balance Rule' (LIBR) legal algorithm in Python.

The Hard Part was OCR. I benchmarked Tesseract vs. Mistral OCR-3 on messy bank PDFs; Mistral won because it understands table layouts.

Built with Django/Celery to handle 10k+ transaction ledgers.

Happy to answer questions about the OCR pipeline or the graph theory implementation.

If you’ve ever had to prove which dollars in a drained bank account belong to you vs. a spouse, you’ve run into the Lowest Intermediate Balance Rule (LIBR). It’s a 50-year-old legal precedent (See v. See, 1966) that is a nightmare to calculate manually.

I’m a dev who got frustrated seeing forensic accountants charge $500/hr to do this in spreadsheets. So I built Exit Protocol to automate the forensic tracing and "impeachment" of financial lies.

OP here.

Just to share a bit of the engineering headache behind this: The hardest part wasn't the Django backend, it was getting the LIBR (Lowest Intermediate Balance Rule) logic to handle edge cases in the PDF parsing.

Bank statements often group transactions by date, not time. If a user deposits $5k and withdraws $5k on the same day, the order matters for the 'dip' calculation. I ended up having to write a heuristic that forces 'Withdrawals First' (worst-case scenario for the claimant) to ensure the report stands up to conservative judicial scrutiny.

If anyone here has worked with financial event sourcing for legal compliance, I'd love to know how you handle same-day timestamp ambiguity.

My friend got quoted $50k by a forensic accountant to trace "separate property" through a joint bank account during his divorce. I realized the math they use (Lowest Intermediate Balance Rule) is just transaction replay with some edge cases. So I automated it.

What it actually does:

Upload bank statements and text messages. The system cross-references them. When someone claims poverty under oath but bought a $195k yacht last month, it generates an impeachment report with suggested deposition questions.

Also does the boring stuff—LIBR tracing with SHA-256 verification so opposing counsel can't claim you doctored the evidence.

The interesting technical bits:

-Mistral OCR-3 for parsing messy bank PDFs (way better than Tesseract on real-world scanned documents)

-Force-directed graphs to visualize money flows through shell companies

-Django + Celery because some statements have 10k+ transactions

-Every report gets cryptographically hashed—anyone can verify authenticity via public tool

Things I'm unsure about:

-Will courts actually accept these? The math is correct and follows established legal precedent (See v. See, 1966), but "AI-generated report" makes some judges nervous.

-Is this B2C or B2B? Individuals use it once. Law firms use it repeatedly. Can't figure out which to focus on.

Pricing is a mess. So it's completely FREE for now.

Current status:

-Live at exitprotocols.com

-Processing real cases (5 so far)

-One family law attorney in Mumbai using it for a hearing next week

-Made front page on previous Show HN, got good feedback on the SHA-256 verification approach

The thing that surprised me:

The LIBR algorithm was the easy part. The hard part was the OCR—bank PDFs are cursed. Scanned at angles, different fonts, merged table cells, handwritten notes. Mistral's document OCR is genuinely impressive for this.

Also surprised by how many lawyers said "this would've saved my client $30k." Apparently forensic accounting is one of those things where everyone knows it's too expensive but nobody's bothered to automate it.

I'd love feedback on:

-Family law attorneys: would you actually submit this to a court? What would make you trust it more?

-Forensic accountants: am I missing edge cases in the LIBR implementation? (I have 12 test cases so far)

-Anyone: is the contradiction detection feature useful or just creepy? It cross-references financial affidavits against spending to catch lies—feels powerful but also invasive.

Sample report (anonymized data): https://exitprotocols.com/static/documents/Sterling_Forensic...

GitHub: https://github.com/Vinaygond/exit-protocol (will be public once I clean up some embarrassing code)

DEMO Credentials:

email: [email protected] ps: password123

Happy to answer questions about the implementation, the legal stuff, or why I thought this was a good use of time.

Thanks!

This is synthetic demo data kept it short so people could see the methodology clearly. Real reports have way more transactions (platform handles 10k+ via Celery).

It's meant to be public as a sample showing the output format. Real client data is obviously private/encrypted.

OP here – thanks for the feedback. I just pushed an update to address the Chain of Custody concerns.

The system now generates immutable forensic reports with SHA-256 integrity hashes for every document. Also added a regression suite to verify the tracing algorithm against known edge cases. The focus is definitely shifting from just "AI wrapper" to "Audit Compliance tool.

Case: https://exitprotocols.com/static/documents/Sterling_Forensic...