A Compare tab on any review, showing both scores per clause against the previous round. Liability: Escalate → Negotiate. Fourteen clauses unchanged, unread, unbilled.
How it works →Everything that moved.
Vern 2.0 is a rebuild, not a refresh. Every change that touched a review is listed below, dated, with the surface it landed on. Including the ones we took out.
Their redline comes back, and you read only what moved.
The release that makes Vern 2.0. Round diffing is the one we'd been building towards: version three arrives, both scores are recalculated, and you read the four clauses that changed instead of all eighteen again. Everything else here is the housekeeping that a product handling other people's contracts owes you.
Forward the counterparty's revised version to the same thread and it attaches as the next round rather than a new contract, at the follow-up rate. Ask Vern can be asked what changed since last time.
An admin can erase the document, the findings and the message bodies in one action, immediately and without an undo. The record that a credit was spent survives; re-uploading the same file is a fresh review, never a free recall.
Month-to-month credits consumed, broken down by where the contract came in from, and one invoice line per delivered review.
The same round comparison the Compare tab uses is a documented endpoint, so you can post the deltas into your own deal room.
API reference →It was built, and then removed before release. An unauthenticated page holding contract analysis is at odds with everything else we publish about where your contracts live. A report link now works for anyone in your organisation who is signed in, and for nobody else.
Real tracked changes, not a comment about one.
Vern's suggested wording now goes into the document as a Word tracked change their counsel can accept or reject, under your name. Word on Windows, on Mac and in the browser.
One clause at a time, or every suggested wording in the contract under a single tracked edit.
See the add-in →Jump the cursor to the clause a finding came from, and leave the reasoning as a Word comment. The wording is the redline; the reasoning is the comment. Never both in the same place.
Where two clauses read alike, or the document has moved on since the review, Vern refuses to edit and leaves a comment instead. Measured against a corpus of deliberately awkward contracts, it has never edited the wrong clause.
If somebody typed "8.1" rather than using Word’s numbering, it keeps its formatting and never shows up as part of your change.
If the document has changed since Vern read it, the panel says so and offers a re-check rather than quietly working from the old text.
A paragraph the other side had moved appeared twice in the text Vern read. It doesn't now, and there's a regression test so it can't come back.
The dashboard, the API, and the panel in Word.
Everything you actually touch. Findings stop arriving as a finished report and start arriving as they're scored.
A fast pass puts a verdict against every clause within seconds, then the deep pass upgrades each one in place. You start reading before Vern has finished.
Generate the code in the dashboard, type it into the panel once. Single use, ten minutes, revocable from settings. No password in Word, ever.
Open a contract Vern has already read and the review comes straight back into the panel. No new run, no credit, no waiting.
The dashboard is just another client of it. Keys are shown once, stored hashed, and carry only the scopes you grant them.
API reference →Publishing snapshots a version. Every past review keeps the version it was scored against, so changing a position today never rewrites what you decided last quarter.
See playbooks →Provider errors used to reach you verbatim. Now you get plain English, a free retry, and no credit spent.
Two scores, and a citation under each one.
The review engine itself. This is the release the rest of the product hangs off.
Playbook Score for whether the clause matches the position you set. Vern Score for whether it would hold up under UK law, judged independently of what you would accept.
Both scores explained →Quotes are round-tripped against the document. An anchor that can't be verified is dropped rather than approximated, so checking a finding is a glance instead of a re-read.
Clauses that contradict each other, like an uncapped indemnity that walks straight through your liability cap, plus the positions your playbook requires that the contract never mentions at all.
"Fees" meaning something different in clause 3 than it does in clause 8 is the sort of thing that costs money quietly.
Reserved at intake, spent on delivery. Failures, timeouts and re-runs on the same version are free, and a later round of the same deal is charged at the follow-up rate.
How pricing works →Runs sit on a durable queue. Deploy in the middle of a sixty-page contract and it carries on from where it was, without charging you twice.
Reading the document properly.
Unglamorous, and the reason citations work at all. If "3.1" doesn't survive extraction, nothing downstream can point at it.
Decimal, letters, roman, multi-level, restarts. The number you see in Word is the number Vern cites.
The text layer is read directly where there is one; a thin-text scan routes to OCR instead of being read as three words a page.
Text hidden by formatting, or white on white, is either excluded and flagged or kept and flagged. Nothing invisible passes through silently.
Parsing runs in a separate process holding no keys at all. Contract text is treated as data, never as instructions, including the instructions somebody hid in white-on-white.
Trust centre →25 MB, around 300 pages, PDF and Word only, and the file type is sniffed from the bytes rather than trusted from the name.
Detected at the door and refused with a message, rather than dying halfway through and charging nobody's favourite kind of nothing.
Ask Vern.
Chat, but scoped to one contract and made to go and look. Most assistants answer a contract question from memory of contracts in general. This one fetches the clause first.
Anchored to one review. It pulls the clause, your playbook position and the surrounding text with tools before it answers, so it quotes rather than remembers, and it can only ever read the contract you have open.
See Ask Vern →Same conversation, same thread, without leaving the document. Switch tabs and the answer you were reading is still there.
No credits, no per-message charge, no quota. Asking a follow-up question about a review you have already paid for shouldn't be a decision.
An address you can forward a contract to.
The intake lane for the people who will never log in. A contract arrives in somebody's inbox, they forward it, and the review comes back as a reply.
Every organisation gets its own inbound address. Forward a PDF or a Word file and the review comes back on the same thread. Anything from a sender you haven't approved is dropped silently, so a spoofed address never gets an answer.
How email-in works →Invite the people who only ever forward an email. No seat, no login, revocable in one click, and named individually. We don't trust a whole domain because one person at it is a customer.
The full summary with both scores per clause, or a bare confirmation that it landed and the findings are in the dashboard. Set per organisation, because not everyone who forwards a contract should read the analysis of it.
If it ends up somewhere it shouldn't, an admin can replace it. The old one dies immediately, and a forwarded contract can't be duplicated into two reviews by a retry.
Three things next. Which one first?
We don't publish a roadmap with quarters on it, and we're not going to pretend the order is already decided. Pick the one you'd actually use.
Ask the assistant you already have open to review a contract. It calls Vern, so the answer comes from your playbook rather than the model's instinct. What it will do →
The review as a document you can send internally, and the redline as a Word file for the people who will never open the add-in.
Vern joins the email negotiation as a supervised junior colleague: reads what arrives, drafts the reply and the redline, chases what's outstanding. Nothing substantive leaves without a named human approving it, and it can never agree terms or sign anything.
Want the next one in your inbox?
One short email per release, written by the person who built the thing. No product newsletter, no drip sequence, and it goes out after the change is live rather than before.