A file someone sends you is hardly ever a single paper. An eight-hundred-page PDF can contain a statement of claim, forty invoices, three reports and a judgment. Read and split makes JUDLEX read the whole file and divide it into each of those papers, each in its own PDF.
What it does
It reads every page. If the PDF carries its own text, it uses it as it is. If a page is scanned, the artificial intelligence reads it by looking at it, as a person would. If the PDF's text comes with the letters scrambled, that page is read by looking at it too.
It separates the papers. It decides where one document ends and the next begins. Each split has to rest on a sentence that appears on the page; if there is none, the split is discarded. It works through the pages in batches, and at the start of each batch it looks at the last page of the previous one, so that a long document is not cut in two just because it straddles both. The blank sheets that separated the papers when they were scanned do not become papers of their own: they are joined to the neighbouring document.
It gives each paper its details. The date, the type of document and who signed it, when that is on record. From these it builds the paper's name: date, type, issuer and subject, in that order, so that outside JUDLEX they also sort by date.
What is not on record is not filled in. A date that is inferred rather than read is marked as inferred. Who signed a paper is never inferred: either it appears on the page or it is left blank. A page that cannot be read is declared illegible, not made up.
Where to start it
Read and split all (N), in the row of buttons in the client file, splits in one go every document in the matter that has not been split yet. It only appears when there is one.
Read it and split out its documents, in a file's menu under Originals, splits just that one.
The chat's Go ahead button, when papers have just come in, does the same with whatever arrived: Add papers to a matter.
Before it starts: what will be sent
Click the button
JUDLEX first reads, at no cost, the text already inside the PDFs, and counts what it would have to send to your provider. Nothing is sent yet.
Check what will be sent
A dialog tells you which model will be used and how much is sent: how many scanned pages need looking at, one call each, and how many pages need splitting, in how many calls. There is no price: your provider charges for it, and what was spent is shown at the end.
Click I confirm it
Not a single call goes out before that.
While it works: in the chat
When you press I confirm, the chat opens in the matter's conversation. Your request appears as a message from you ("From the client file · Read and split") and the chat tells you each step as it goes: which document it is reading and which papers came out of it. You can keep asking it questions while it works, and afterwards.
The line under the client file's buttons also shows where it has got to. The Stop button, in the chat or on that line, is the same one: it halts once the current page is finished; anything already requested is still paid for.
A troublesome document does not hold up the rest. If a document still has pages that could not be read, it is not split: it is skipped, you are told why, and the next one follows. The pages that were read are not paid for again.
Neither does a dropped connection. If the connection to your provider goes, JUDLEX waits for it to return and carries on with the same document, without you having to do anything. If it takes too long to come back, it stops and tells you.
What does stop everything is whatever waiting will not fix: your provider rejecting the key, your account with it running out of credit, or two documents in a row failing for the same reason. The chat tells you which.
If you close the client file or reload the window, when you come back the work carries on in the same chat conversation, where it left off.
What you see at the end
The chat closes with what happened:
How many documents were split, and into how many papers.
How many were left undone or skipped.
How many splits were discarded because the sentence that justified them was not on the page.
What was spent, in dollars.
If there are warnings — pages declared illegible, skipped documents and why — you have them in the bell, grouped by document.
The new papers appear in the tree under Unfiled until you click Classify: Classify and identify.
What it costs
This does cost money, because the artificial intelligence reads the whole file. Your key pays for it, at your provider's rates, and the call goes straight from your computer to the provider. Scanned pages weigh the most, because each one is looked at as an image; a PDF with clean text needs far fewer calls. A page that has already been read is never sent again.
The figure next to the matter's name, at the top of the client file, adds up what has been spent on sorting out that paperwork. Every call is also listed in the spending History in Settings.
Pending pages
Sometimes pages without trustworthy text remain in documents that have already been split: a scan that was never read, or a PDF whose text comes with the letters scrambled. Then the Look at N pending page(s) (with AI) button appears.
The artificial intelligence reads those pages by looking at them, once, and that is all, telling you about it in the matter's chat: no document is split again. The papers that contain them go back into the queue for Identify the rest, because now there is text to identify them by. If there are many pages, it asks you first and tells you how many it will send; if there are only a few, the button itself has already told you it uses AI, and it gets going.
Split again, or go back to the originals
If a split has gone wrong, there is no need to delete the matter and upload everything again.
Split again. Under Originals, a file that has already been split offers Split this original again; in the matter's menu, Split the originals again does it for every file still in the vault. Before it starts, it tells you what it will send, just like the first time. Papers that come out the same stay as they were, with their folder, name and numbering. The ones that change are replaced, and whatever JUDLEX had drawn from them — facts, people — moves to the new paper that contains that page. Each new paper is left amber in the folder it came from; then click Classify. What has already been read is not paid for again.
Back to the originals…, in the matter's menu, returns the matter to how it was when the files were uploaded. It asks first and tells you what will be lost:
The split papers and their exhibit numbering go, along with the folders, the facts, the people and parties that cited those papers and whatever was noted while reading them, including your own changes to all of that.
The originals stay, as does what was already read from their pages, the case files, the conversations, the spending history and the diary deadlines.
To get it all back you have to split and classify again with artificial intelligence, and that is paid for again. When it finishes, the notice offers Split now.
When it cannot be started
The option is greyed out, with the reason, if a split is already running or the original file has been withdrawn. It cannot be started without an active licence either.