Redline: contract review that shows its work

The problem

Freelancers sign contracts with clauses that cost them later: rights that pass to the client before an invoice is paid, or payment terms the client controls. An AI reviewer can help, but only if its output can be trusted. A flag that paraphrases, or points at something the contract doesn’t say, creates more checking work than it saves.

The call

I built Redline around one property a reviewer can check: every flag quotes the exact clause it came from. If the tool can’t point to the text, it doesn’t raise the flag.

What it does

• Source-quoted flags. Each risky clause is cited exactly, by section, so it can be verified in seconds.
• Severity ranking. Flags are rated by how much harm the clause could do, such as Severe or High.
• A drafted counter-offer. Each flag comes with a suggested change you can send back. For example: assign rights only on payment and carve out the contractor’s existing tools, or set a fixed payment date from delivery instead of from the client’s approval.
• A clear result when nothing is wrong, rather than silence, so a reviewer knows the check ran.

How I ran it

• 10 decision records behind the product, so every choice has a written reason.
• 12 tickets, each delivered as working, tested code. All 12 shipped.
• A live test: I planted six risky clauses in a sample contract and ran Redline on it.

Results

• 12 of 12 tickets shipped.
• 5 of 6 planted risky clauses caught, each with the correct severity and an exact citation.
• Live at redline-eight-lovat.vercel.app.

Honest limits

• The 5-of-6 figure comes from one test run, not a measured accuracy rate.
• It’s a review aid, not legal advice.

If a reviewer can’t check a flag against the document, the flag isn’t worth showing.