The third redline lands on a Thursday afternoon. Same agreement, forty-eight clauses. Somewhere in there they have moved the liability cap, added an indemnity and renumbered everything after clause nine. You have already reviewed this contract twice. You are about to read all forty-eight clauses again to find the six that changed.
That is the problem Vern 2.0 was built to remove. It went live on Monday 10 August. Thirty-seven changes, none of them breaking, with one idea sitting underneath the lot: a counterparty redline is not a new contract, it is the next round of the one you are already negotiating.
Every contract review tool we have used, including the previous version of ours, treats each upload as a fresh job. Same document in, same full report out, same price. That is fine for the first pass. By round three it is making you pay, in time and in money, to re-read the forty-two clauses nobody touched.
Round diffing
Vern now holds the negotiation as the object and each draft as a version of it. Send in a redline and it arrives as the next round of the same matter rather than as a new document with no history. The review gains a comparison view, and it shows you the deltas: what moved, which way and whether it is now good enough to sign.
Read that top line first. The matter went from 62 to 71 and is still sitting in Negotiate, so the deal improved without becoming signable. Underneath, the liability cap climbed from 12 to 48 because they conceded a cap at three times fees rather than the two times you asked for, and the restrictive covenant went from 41 to 76, which is a full concession and better than your own fallback.
The statuses do the triage. Improved is a clause that moved your way. New issue is one that appeared this round, and clause 15.2 is exactly why that status earns its place: a free right to novate the agreement to any group company, which was not in round one at all and quietly transfers your fee entitlement without consent. That, not the liability cap, is the thing to escalate. Still open is the quiet one. Payment terms read 54 before and 54 after, word for word identical to round one, which means they have not acknowledged it rather than refused it.
Both scores travel across rounds and they stay separate the way they always have. The Playbook Score, out of 100, answers whether the clause matches the position you set. The Vern Score answers whether it would hold up under UK contract law, judged without your playbook in the room at all. A clause can climb against your playbook and still read as Caution on its own merits, which is worth seeing, and is why Vern never averages the two into a single reassuring number.
Why renumbering does not break it
The obvious way to diff two contracts is to compare clause 11.2 with clause 11.2. It works right up until the other side inserts a definition at the top and every number below it shifts by one. Then a naive diff reports that all forty-eight clauses changed, which is worse than useless.
Vern matches clauses by content rather than by position. Each clause is identified by a hash of its normalised text, so the liability cap stays the same clause whether it is numbered 11.2, 12.2 or moved into a schedule. Review runs are immutable, which means any two rounds in a negotiation remain comparable months later. When they renumber the whole agreement, the diff still holds.
What it costs to negotiate properly
Re-scoring only what changed is a technical decision with a commercial consequence, so we passed it through to pricing. The first review of a contract costs a full credit. Every later round of the same deal costs half. Anything that fails or times out costs nothing and the retry is free.
The point is not the arithmetic, it is what the arithmetic stops encouraging. When every round costs full price, people start deciding which redlines are worth putting through review. That is exactly the wrong incentive, because the risk usually enters on round three, in the version everyone has stopped reading carefully.
Where the round actually gets done
Knowing which clause moved is only half of it. At some point somebody has to write the counter, and they are going to write it in Word, because that is where contracts have lived for thirty years and no amount of good software is going to change that.
So Vern runs inside Word as a task pane on Windows, Mac and the web. It has four tabs. Home notices which contract you have open and whether it has been reviewed before. Review lists every clause with both scores. Ask Vern answers questions about the clause you have highlighted. Settings shows the paired device and the playbook version you are scoring against.
Clause 8.1 is the one to look at there. A Playbook Score of 12 because indirect losses are uncapped and your playbook requires two times annual fees or better, and a Vern Score of Weak because it would struggle on its own terms under UK law. Both readings point the same way, which is the easy case. The interesting ones are where they disagree, and Vern shows you that rather than averaging it away.
Two things in that panel matter more than they first look. The first is the tiered fallbacks. Your playbook does not just hold your ideal position, it holds the ladder you climb down: the standard cap, the mutual version and the super-cap you would sign at three in the afternoon on the day the deal has to close. Vern shows each one with the rate the other side has actually accepted it at, alongside its own untested draft, so the person negotiating picks the wording most likely to land rather than the wording that sounds firmest. Every applied position is recorded, and what gets accepted tunes the defaults.
The second is what happens when you apply one. Vern writes a genuine Word revision, real w:ins and w:del markup, attributed to you and timestamped like any other tracked change. Their counsel opens the document and accepts or rejects it in the normal way. That sounds like a small engineering detail. It is the whole difference between a tool that suggests wording and a tool that participates in the negotiation. Everywhere else hands you plain text to copy, which means somebody retypes it, the other side cannot see what moved and the version you send back carries no history at all.
You can apply one position or apply all of them at once. If Vern cannot locate a clause precisely enough to edit it, it refuses rather than guessing, which is the correct behaviour for software allowed to write into a legal document. Pairing is per device and per person: an eight-character code good for ten minutes swaps for a token bound to one person in one organisation, with no shared credentials and no cookies living inside the Office WebView. An admin revokes any device in one click. And reopening a contract Vern has already analysed costs nothing, because the add-in hashes the document text and recognises it rather than charging you to think about it twice.
And where the work is drifting to
The other place people now negotiate contracts is a chat window. Not officially. Nobody has written it into a process document. But the habit is already there, and the version of it we see in client environments is somebody pasting a Terms of Business into ChatGPT to ask what looks unusual.
We spend a good part of our working lives telling clients why that is a problem, so building the sanctioned version of it was not really optional. Vern connects to ChatGPT and Claude as a remote connector. You add it from the assistant's directory and sign in with your own Vern account. That is the setup.
It exposes four tools and no more. analyze_contract submits a document for review and spends a credit. get_findings reads results that already exist. ask_vern answers questions about a contract. list_playbooks reads your positions and fallback tiers. So you can ask your assistant to review the attached Terms of Business and tell you what to push back on, or what your position on liability caps is, or whether Meridian has moved on clause 8.1 since round one. The answer comes from your playbook rather than from the general internet.
The part worth dwelling on is what does not happen. The assistant never receives the contract to read. It calls Vern's tools, and the document text stays where it was. Your role and scope carry across from your Vern account and are enforced on our side, not on the assistant's honour, so a user who cannot see a matter in the app cannot reach it through Claude either. Every call is attributed and audit-logged, tokens are revocable per person and there is no bulk export through the connector.
That is the difference between shadow AI and governed AI, and it is worth being precise about it. The behaviour is identical from where the user sits. One of them puts your client's contract into a third party's chat history and one of them does not.
The rest of 2.0
Three more changes that customers will feel day to day.
Email-in now understands threads. Forward the counterparty's revised version into the same email thread and it attaches as the next round rather than opening a new review. No portal, no re-upload.
Reviews can be deleted properly. An admin can erase the document, the findings and the message bodies in one action, immediately, with no undo. That is deliberate. A deletion that quietly leaves the text sitting in a message body is not a deletion, and it is the first thing a decent vendor security questionnaire asks about. Usage reconciles to the invoice, with credits broken down month to month by where the work came in, upload, email, Word or API, plus a line per review. And diffs are on the API, so round comparison data can feed a deal room or a matter management system through documented endpoints with signed webhooks.
One thing we took out
Public share pages were built for 2.0 and cut before release. The idea was a link you could send to someone outside the platform to show them a review. It worked. We could not make the sharing model line up with how we handle other people's contracts, and a link that is hard to withdraw is a data protection problem waiting to become somebody's incident.
So it is not in the release. It is worth saying out loud, because release notes that only ever list additions are marketing rather than release notes, and because it is the same call we would tell a client to make.
Seven releases since the end of June
Vern 2.0 is a rebuild, but it did not arrive as one. It arrived as a run of small releases that each did one thing and shipped when that thing worked.
We publish that log for the same reason we publish what got cut. If you are weighing up whether to trust a product, the useful signal is not the feature list on the homepage, it is whether the thing has moved in the last six weeks and whether anyone will tell you when it did not.
It is also the honest answer to a question we get asked most weeks, usually by a firm that has a workflow of its own worth productising. The engineers who built and run Vern are the same ones you get on a build with us. Vern is not a demo we spun up to look credible. It has paying customers, a support queue, a versioned API and an on-call rota, and this post is roughly what it looks like when a product like that has a good month.
Try it, or get one built
If you review contracts for a living, the product, the pricing and a demo you can book all live at askvern.ai. Bring a real negotiation with a few rounds already on it. Round diffing is much more convincing on a contract you have argued about than on a sample.
If you are on the other side of the table, running procurement or a vendor security review, or you want a product like Vern built and run for your business, that conversation goes through us. Call 0333 050 0729 or email hello@assurepath.co.uk.
Vern is a contract review tool, not a legal service. It flags what deviates from the positions you set. It does not give legal advice, and it is not a substitute for a lawyer on the clauses that warrant one.