What Redline actually does to your contract

Paste text, upload a PDF, or give it a link to a public terms page. Within about a minute you get a ranked list of traps. Each one carries the exact sentence from your document, where in the document it sits, what it means in plain language, who it hurts, and what a careful person would do about it.

Ranking is not cosmetic. Every trap gets a severity from 1 to 5 and a likelihood of actually biting you, and the list is ordered by the two together. An auto-renewal buried in clause 14.3 that you will hit next month outranks a theoretical liability cap you will never invoke, and the overall verdict at the top — low, medium or high — is computed from what survived, not from how many flags were raised.

The result lives at its own link. Send it to the other person in the deal, keep it for the day the charge appears on your card, or print it to PDF with the quotes intact. Nothing about the report requires an account or an app.

For a document that lives at a URL — a SaaS terms page, a marketplace policy — Redline can keep reading it. Once a day it re-fetches the page, compares it clause by clause with what you scanned, and emails you only when something changed: which clauses moved, which traps are new, which disappeared, and whether the terms as a whole got better or worse.

Length is a strategy. A subscription agreement runs to nine thousand words because the cost of writing it is paid once and the cost of reading it is paid by every customer. Nobody has an hour per checkout, so the document is drafted knowing it will not be read.

The dangerous clauses are also the dullest. "Unless otherwise terminated in accordance with clause 9.2" is where the automatic renewal lives, and it reads exactly like the twenty neighbouring sentences that mean nothing. Attention runs out long before the trap arrives.

Search doesn't rescue you either. There is no reliable word to search for: the same trap appears as "evergreen term", "successive periods", "shall continue thereafter", or a defined term two pages away that quietly redefines "Renewal". You have to understand the clause to find it, which is the work you were trying to skip.

And a general-purpose chatbot will cheerfully summarise a contract, then invent a clause number, soften a quote, or reassure you about a paragraph that isn't in your document. When the output is going to inform a decision about money, a fluent summary with a fabricated citation is worse than nothing.

First the document is segmented by its own structure. Redline follows the numbering the drafter used — 4.2, §7, Article 3, ¶12 — so every finding can point at a real place on the page instead of at "somewhere in the middle". PDFs are read as text; a scanned image with no text layer is refused rather than guessed at.

Then the analyzer reads every segment against a catalogue of 79 known traps: automatic renewal, unilateral price changes, arbitration and class-action waivers, perpetual content licences, deposit forfeiture, non-competes, data resale, silent scope changes and the rest. Each catalogue entry carries a prior severity and likelihood, so the model is judging this document against a known baseline rather than inventing a scale each time.

Every candidate then has to pass the quote gate. The quote it produced must appear in your document character for character — one contiguous span inside one segment. Paraphrases, merged sentences, tidied punctuation and helpfully shortened quotes are all thrown away before a human sees them. This is the single most important step: it makes a fabricated finding structurally impossible to display, because a finding with no verbatim anchor is deleted rather than shown.

What survives goes to a second reviewer — an independent pass that reads the same document and argues against each flag. It can confirm a trap, call it overstated, or refute it outright. Refuted flags are dropped; overstated ones are downgraded to the severity the evidence actually supports. You see the count of what was checked, blocked, refuted and downgraded at the top of every report, so the filtering is visible rather than a claim.

Deep review is a slower mode for documents worth arguing over — a lease, a loan, a vendor contract. You pick your jurisdiction (United States, European Union, United Kingdom, Russia, or "not sure") and each trap gains a note on how that kind of clause tends to fare there, a suggestion of what to ask the other side to change, and a short list of questions to ask before you sign.

Your document is not kept. Pasted text and uploaded PDFs are processed in memory and discarded; what is stored is the report — the quotes and explanations — so your link keeps working. Pages fetched by link are snapshotted, because comparing today's terms with the version you scanned is the whole point of monitoring.

Redline is deliberately narrow about what it will say. It reports what the clause says and what it typically means for a reader in your position. It does not tell you whether to sign, does not predict how a specific court would rule, and does not replace a lawyer on anything that matters enough to need one.