One Card for the Whole Run

When Miss Lucy takes on a real piece of work — research a point, then draft on it, then check the citations — that is one job to you and about four separate processes to us. For a long time the screen showed you the processes.

There was a chain-of-thought panel. There was a second chain-of-thought that appeared under individual messages. There was a reasoning pane. There was a card that appeared when she handed work to a specialist, and another card when the citation checker ran. Five surfaces, one job, each of them true, none of them telling you the thing you actually wanted to know, which was: where is she, and how much longer.

Since last week there is one card.

What it says

The card covers the whole run and shows it as phases rather than as processes. She answers, she delegates, she verifies. Each phase is a row. A specialist — case research, an opinion, the citation desk, an SLP paper book — collapses into a single row rather than announcing itself as a separate event, and you can open any row if you want to see what happened inside it. There is a progress ring, an elapsed clock, and a status badge that has four states, one of which is failed.

That last one matters more than it sounds. A long research-and-drafting run can take twelve or fifteen minutes. Under the old screen a run that died halfway through mostly looked like a run that was still thinking, and the difference between those two things is the difference between waiting and starting again.

And when it is over, it tells you what it cost

The other new thing is on the welcome screen, before you open a chat. If you have run something before, there is a card for the last run you finished — what you asked, whether it worked, how long it took, and what it cost you in tokens.

Two decisions in there were argued about more than the rest of the card put together.

A brand-new user sees nothing. No empty frame, no "your runs will appear here". An empty placeholder on a first login teaches somebody that the product is empty. An absent one teaches them nothing, which is the correct amount to teach them at that moment.

A failed last run renders as failed. It is not hidden and it is not softened into "incomplete". This is the only place in the product that tells a returning user that her last attempt died, and she needs to know that before she spends anything on a second attempt.

The two lists underneath

The same pass redrew the two screens where work accumulates.

Matters is now a list rather than a board. Each matter carries the number of documents and the number of drafts on it, and a state in one word — active, quiet, or needs attention. A board looked impressive and made you drag things between columns to describe a status the system could work out on its own.

The Vault is a table, and the status of an uploaded file is one of three words: Reading, Ready, or Couldn't read. It used to be five labels, and two of them meant the same thing to everybody except us. You can now file a document to a matter from the row it is sitting in, which sounds small and removes one of the more irritating dead ends in the product.

The part we got wrong on the way here

All four of these — the run card, the matters list, the Vault, the dashboard — shipped behind flags. That was sensible. The phase model was new, and if it turned out to be wrong for some workflow the lawyer would see nothing, which is worse than the problem we were fixing.

What was not sensible was how the flags resolved. They defaulted on for localhost and on the development site, and off in production. Read that again slowly, because we did not, for longer than we should have. It meant every developer, every screenshot, every review and every browser check saw the new screen, while every paying customer saw the old one, and nothing on either screen said which was which.

A flag whose default depends on where the browser is pointed does not stage a rollout. It makes your own review structurally incapable of seeing what a customer sees. We deleted all four flags, and we deleted the pattern with them — if something genuinely needs staging in future it will be gated on something the customer's own account carries, never on where the browser happens to be.

None of this is behind a flag now. If you have a Miss Lucy account, it is what you get.

Why bother redrawing progress at all

Because the argument for showing the work is only worth anything if the showing is legible.

We have said from the beginning that you should be able to watch Miss Lucy reason, so that you can catch her before a court does. That is still the position. But five true surfaces that between them never answer where is she and how much longer is not transparency — it is instrumentation, pointed at the wrong reader. The information was all there. It was arranged for the people who built it.

One card, three phases, a clock, and a number at the end telling you what the run cost. That is the same information, arranged for the person paying for it.