Arguments can now hold the original text they were built from. Attach a source text to an argument and it is kept as a piece of the argument in its own right — you can say whether the argument sets out to represent that text faithfully or merely started from it, link individual premises and claims to the exact passages they came from, and attribute the text to a real, publicly reachable source. You can also mark a premise or a claim as one the original left unspoken, so a reader can tell what the speaker actually said from what someone supplied on their behalf.

Imports now fill this in for you. When you paste a piece of writing and have it analysed, the text itself is kept alongside the resulting argument, and the claims the analysis produces are linked back to the passages of your text they came from — no highlighting by hand. The thorough analysis links the inferences between those claims as well; the fast one links the claims only, since it reads the claim set rather than your text when it works out how they hang together. An argument imported from X or Reddit also has its source named automatically, with nothing to fill in — and that automatic name is only a starting point, so you can correct it to a proper citation, or remove it, whenever you like. An import is always treated as having started from its text and never as claiming to represent it faithfully; that is a statement you make deliberately, if you want to make it at all. A pasted text with no source to name simply has none.

Where the analysis quotes something it did not actually find in your text, no link is made rather than a link to the wrong sentence.

You can now read all of this on the argument itself. An argument built from a source text shows that text beside it, with every passage its premises and claims trace back to marked in the text. Point at a claim and its passage lights up; point at a passage and every piece of the argument that came from it lights up in return — including a claim used in more than one place, which is marked at each of them. Where the author named the source of the text, the full reference is there too. A premise or claim its author marked as one the original left unspoken says so plainly, whether or not the argument has a source text at all. An argument with no source text looks exactly as it did before.

Tapping a passage marker opens the source text and takes you to that passage, and it stays marked once you look away — so the answer is still there when you arrive. If you had closed the source text panel, it opens back up. Everything works by keyboard as well as by pointer, and a screen reader reads the passage itself rather than a label standing in for it.

Writing all of this — attaching a text, drawing the passage links, marking content unspoken — is still to come; for now these are things you can read. What is already in place underneath behaves the way you would want: publishing a new version or forking an argument carries the source text, its passage links, and the unspoken marks straight across; pasting a text you have already stored reuses it rather than keeping a second copy; and removing a source text frees the space it was using, and so does deleting the argument that held it. Naming the source of a text is something you do once — since two arguments can share the same text, a later edit can't quietly rewrite or erase the name someone else gave it. A name filled in automatically doesn't count against that: it's a suggestion until you confirm or replace it.

You can now do all of this yourself, on the argument. Open the Source Text panel and paste in the text an argument was built from, say what role it plays, and name its source if you want to — nothing asks you to, and an argument whose text has no source to name simply has none. From a claim's or a premise's menu, "Link to the source text" opens the text and lets you select the exact passage that piece of the argument came from; the same menu marks a claim or premise as one the original left unspoken, which you can do whether or not the argument has a source text at all, because it is a statement about what was said rather than about anything stored here.

Selecting a passage works with the keyboard alone — move to the text, hold Shift and use the arrow keys — not only by dragging with a mouse.

When an argument sets out to represent its source faithfully, the panel points out the parts that trace back to no passage and offers to mark them unspoken. They are suggestions and nothing more: nothing is recorded until you say so, and dismissing one changes nothing at all. A forked argument that inherited that same claim to faithfulness is asked to review whether it still holds.

The showcase arguments now come with the writing they were read out of. Federalist No. 10, the Crito, and Mill's chapter on the liberty of thought each sit beside their full canonical text, with the passage behind every claim marked in it — point at a claim and the sentence it rests on lights up. Each names its source properly. These are the arguments where the claim to represent a text faithfully is simply true, so that is what they say. "The Singer Solution to World Poverty" is unchanged: it is a piece still in copyright, and reproducing it in full is not a call to make lightly.

You can now build an inference out of a whole grouped premise in one step. If a premise says "A and B", its menu offers "Build an inference from this premise" — pick the claim it rests on, and you get "A and B, if C" with the grouping left as it was. Before this you had to work from the inside out, which quietly added your new claim to the group instead of building around it.

Wrapping a single claim into an inference now keeps that claim as the conclusion, which is what the formula on screen already showed. It used to be saved the other way round, so what you read and what was stored disagreed.

Paging through Explore no longer skips arguments. Where several arguments were published at the same moment, moving between pages could hide some of them entirely and show others twice. The same applied to lists sorted by popularity, and to paging through your own list of imports on your profile.

Browsing argument lists also now stays fast as the catalogue grows, instead of getting slower with every argument added — however you sort them, and including the public Explore listing. You won't see much difference today; it's the ceiling that moved.

An argument that sets out to represent its source text faithfully now says so to its readers. That was always a deliberate statement by the author, but until now only the author could see it — and it is the thing that tells you how to read the rest: when an argument claims to represent a text, a premise that traces back to no passage is worth explaining rather than unremarkable. Arguments that merely started from a text say nothing extra, which is exactly what they meant.

A source text containing a long web address no longer runs off the edge of the pane. The address used to stretch the line far past the panel with no way to scroll to it, hiding the rest of the text — and any passage inside the hidden part — with no way to reach it at all.

Content marked unspoken that also has a passage of the source text attached to it now says so on its own label. Those two statements contradict each other — the original left this unsaid, and here is where the original said it — and the marker used to look identical to an ordinary unspoken one, so you could only find out with a screen reader.

A source text whose reference was pulled out of the writing automatically is now flagged as needing tidying up here too, the way it already is when you export the argument or read it on your phone.

Attributing a source text you typed in yourself no longer tells you it was filled in automatically.

Clicking a passage of the source text now works on the first click, and answers in the argument. Every passage after the first used to take two clicks: pressing one scrolled the pane back to the passage you had chosen before, and the press was lost. And the passage you picked lit up the premise it produced without ever bringing it into view, so on a long argument it could light up something a screenful away and look like nothing had happened. Both directions of the link now behave the same way: pick either end, and the other end comes to you.

The source text is now reachable when the sidebar is collapsed to its strip of icons. It has an icon there like every other section, which matters most for an argument whose text is attached but not yet linked to any passage — there was previously nothing at all to open it with.

The list of parts that trace to no passage of the source text no longer counts the same claim more than once. Behind every claim sits some bookkeeping the app maintains for you, and the list was reading that too: each claim was offered twice, and each one added a spurious row reading "an untitled premise". An imported argument where every claim was already linked to a passage offered sixteen rows when the right answer was none. Marking one of the spurious rows did nothing visible and could not be undone. The list now shows one row per premise and one per claim, exactly as they appear in the argument.

Setting a suggestion aside is now labelled Hide for now, because that is what it does — it clears the row for this visit and does not record a decision.

Pasting too much text into the importer now tells you so. The import was already refused, but the dialog said nothing at all: you clicked and nothing happened, with no way to find out why. It now shows the same message you get when you attach an over-long text by hand, naming the size and the limit.

A source text you attributed yourself no longer shows its reference twice.

Clicking Source on a claim now shows you the line, right there. A small panel opens beside the claim with the passage it was built from and a few words of the writing on either side, so you can see where it sat without opening the whole source text and losing your place. An ellipsis marks each end where there is more text you are not being shown — and only there, so a passage that opens the document does not pretend something came before it.

The whole source text is still one click away, on View full text in that panel, and it opens with the passage already highlighted and scrolled to.