At 6:40 on a Tuesday in March, a portfolio manager I speak to most months is doing what she's done for eleven years. Coffee going cold, screen at arm's length, reading down the overnight tape and deciding which items she's allowed to care about. Concentrated equity book, a bit over $180m, thirty-two names. Her scarce thing was never information. It's that she has until roughly 8:15, and after that the day starts happening to her instead of the other way round.
She used to spend forty of those ninety minutes deciding what to read. Not reading. Deciding what to read.
That gap is the entire product. Everything below is about closing it, and about the places where it still doesn't close.
Ninety-one rows, and she opens four
Her feed had ninety-one items in it that morning. Six columns: ticker logos, headline, actionability, sentiment, provider, time. The two middle columns are the reason the other four are tolerable.
Actionability arrives as a badge with a number on it. Actionable 0.72. Watch 0.51. Stale 0.19. The cut sits at 60 out of 100 and it's a hard line, not a suggestion: above it the item is something that happened, below it the item is a company clearing its throat with a timestamp attached. Sentiment sits next to it, signed and conviction-weighted, so Bullish +0.58 tells you the direction and how much the engine is willing to stand behind it in the same glyph. A confident small move and a shaky large one don't collapse into the same number.
She reads badges. She opens four rows. When she does open one, the drawer slides in from the right with the five sub-scores drawn as bars, each with its weight printed beside it: novelty 0.30, materiality 0.25, surprise 0.20, specificity 0.15, directness 0.10. Underneath that, the engine's own reasoning in a paragraph.
Here's the behaviour I did not expect and now watch more closely than anything else on the page. She reads the reasoning field before she reads the article. Not instead of. Before. It's become a filter of its own: if the reasoning is thin, she closes the drawer and never opens the source.
The stream itself is deliberately undramatic. New rows slide the list down over about six-tenths of a second and carry a highlight that fades out across five seconds, and then nothing. No sound, no counter spinning, no red. On a quiet morning the screen just sits there, which is the correct behaviour and took an embarrassing number of iterations to be brave enough to ship.
Five views, in the order she looks at them
A watchlist here holds a saved filter, not a list of symbols. Tickers and providers, plus four tag dimensions we label Topic, Market, Country and Sector. So "my energy book" and "US macro prints" and "anything touching fertiliser" are three separate lenses rather than three separate spreadsheets.
They live in the sidebar with a coloured bookmark each, and you drag them into whatever order your morning runs in. Hers goes energy (amber) first, then a red one for three specific chemical names, then macro, then two she built during Q1 earnings and has never deleted. That last detail matters more than it sounds. People do not tidy up, so the ordering has to be theirs and the colours have to be theirs, or the sidebar becomes a place they stop looking.
The bit that gets used hardest is the temporary edit. Change a filter chip while you're sitting on a saved view and the header quietly marks it (modified), the change goes into the URL, and two buttons appear: Save writes it back to the view you're on, Save as new spawns a fresh one. Most of the time neither gets clicked and the view snaps back when she leaves it. Occasionally she pastes the URL into a chat with her analyst, who opens exactly what she was looking at.
"I stopped keeping the tab open all day. I keep it open for two hours and then I close it, and I know I haven't missed the thing."
Portfolio manager, single-manager equity fund, ~$180m
Three places this doesn't help you
Say the limits out loud or a user finds them at the worst moment.
If your edge is the first tick, look elsewhere. A row appears when it has been judged, not when it published, and judgement is measured in seconds rather than microseconds. That gap is the product working as intended, and it's the wrong trade for anyone whose whole edge is arriving first.
On the Free plan you get 50 realtime items a day and no scores at all. The server strips the actionability and sentiment values before they ever reach the browser and drops an Upgrade pill in the two columns where the badges would be, which is honest but is also, functionally, the six-column table minus its two useful columns. Scored feed starts on Starter at $49 a month, $39 if you pay annually.
And the live table holds 100 rows. It's a window on the present, deliberately, and it forgets. Anything older lives behind the REST history, where the lookback is 48 hours on Starter, 30 days on Pro, a year on Quant. If you want to sit and read backwards through a quarter, that's an API job, not a feed job.
Warren can't see your feed, and that's on purpose
Warren sits in the sidebar under Chats. He'll search the web, read a URL you paste at him, and run the same Verdict Engine over a document you hand him, so the score he gives back and the badge on the row can't drift apart.
He also cannot see your feed. Not a bug, not a roadmap item I'm dodging: he has no read access to your private stream, and if you ask him what came through this morning he'll tell you so rather than guess. Ask him instead about the press release you just pasted, or the 10-Q you dragged in, and the answer comes back as a card you can pull apart before you size anything. One caveat on plans, because people hit it and email us: chat is on Starter, but document scoring isn't. That's Pro at 100 a day, Quant at 1,000.
What people ask for next is always the same thing, and it's the one I can't give them. They want the badge to know they're already long the name in size, so a 0.64 on a position they hold reads louder than a 0.91 on something they'll never touch. We don't hold your positions and I'm not in a hurry to start. Which leaves a fairly narrow path: let the view carry enough of your intent that the ranking approximates it without us ever knowing your book. That's the design problem on my desk this quarter, and I don't have it yet.