Reading the Desk screen

Desk is where you meet the team and read its day. It answers three questions for someone who is not an engineer: what did my desk do today, what did it decide, and why. Nothing on this screen can place an order.

The page is titled Your desk. One sentence under the title says who is working now, or who runs next and when. Three tabs follow.

Today on your desk

Today opens first. The top panel names the day and sums it up in one sentence: 13 of 24 sessions done, 6 trades proposed, 4 stopped early, $5.48 spent. Under it a strip of tiles covers the last two weeks, one tile per trading day. Pick a tile to read that day. Weekends and holidays show as a thin divider with their name. Days are the exchange's, in Central time.

Below it, Sessions lists the day, oldest first, one row per session. Each row starts with its time and kind, then says what the session did in one sentence:

• 8:00 AM, Morning plan: read the research brief, set 5 price levels to watch, no trades before the open.
• 10:21 AM, Woke up: woke because MSFT crossed $488.00, checked MSFT, no trade.
• 10:30 AM, Check-in: checked NVDA, MSFT and AMD, proposed 3 trades (trim 2 NVDA, exit AVGO, buy 2 AMD; 1 filled, 2 waiting for you). The counts are read from the orders as they stand now: once the two approval windows close with no answer, the same row reads 1 filled, 2 expired, on a screen you left open too.
• 3:15 PM, Daily review: wrote up the day and 3 things to check tomorrow.

Every row is built from what the session recorded: its orders and what became of them, the ideas it passed on, the levels it set, what woke it. It is never a summary of the agent's own prose. The small mark says who ran it (the Portfolio Manager's pie with a slice lifted out, the Researcher's broadsheet, the Critic's eye). The quiet words at the end name the model and the cost. A session that is running says working now. One that failed carries a Stopped early pill and the reason in words. The next slot says how long until it starts, and the rest of today's schedule folds into one Later today line.

On a wide screen the whole day sits on one grid: every row starts its mark, time, kind and sentence at the same place down the page, with the model and cost at the right edge, so the list reads like a ledger rather than a stack of cards. On a phone each row stacks the same parts. Above it, one tile per trading day carries that day's totals ("20 sessions · $11.40"). Today's tile says "nothing yet" until the first session runs, an earlier trading day with none says "no sessions", and a bad day adds "2 stopped early" underneath. A run of closed days collapses into a divider, and the date field with the calendar icon at the end of the strip opens any earlier day. The day's own header line sums what you are looking at with the same figures: Today · 20 sessions, all done · 3 trades proposed · $11.40 spent, or 12 of 15 sessions done while the day is still running.

Every role wears the same mark wherever it appears, here and on Activity, in the Inbox, in Consult's answers and in Learning: the Researcher a broadsheet, the Portfolio Manager a pie with one slice lifted out, the Trader two arrows passing, the Critic an eye, the Quant bars on a baseline, Consult a speech bubble, and the Executor a shield because it is code rather than a model. A small still dot at the corner of a mark means that session is running right now. A session that stopped early says so with the Stopped early pill in its row, not with a dot.

The kinds, in plain words:

Above the list, All sessions filters it by kind: Research, Morning plan, Check-ins, Wake-ups, Second opinion and Daily review, each with its count for the day. On one agent's page, which has fewer kinds, the same choices are a short row of chips. When a session was tried more than once (a retry after a failure), only the latest try shows. Its row says 3 tries, the latest at 8:00 AM. 2 earlier tries stopped early. with a Show every try link, and ticking Show 2 earlier tries above the list brings them all back. Three or more quiet check-ins in a row that all made no trades for the same reason fold into one row, 9 check-ins, 10:00 AM to 2:30 PM, which opens when you press it.

On a wide screen the list stays in view on the left while the open session scrolls beside it.

The session you open

Click a row. The latest Portfolio Manager session is open when you arrive, and the open row is marked with a thin orange rule on its left. It opens as one card. The title says what kind of session it was and when it ran (Daily review · 3:15–⁠3:19 PM CT). A quiet line under it names who ran it, the model, the cost and who asked for it. Then come four parts, outcome first and the evidence last:

• What it decided. One small card per name, built from the order itself. A proposed trade says what and at what price, and where the order is now: filled at $128.57 (links to Activity), waiting for you in the Inbox (links to the Inbox), expired before anyone decided once its approval window has closed, or turned down by the risk rules. An entry the rules refused offers Place it myself. An idea it passed on or is still waiting on carries one line with the label Why not, with the reason in one plain sentence ("The price has not reached its level yet."). A session with no trades says why in one sentence, never "0 proposals".
• What it set up for next time. The price levels it armed, one row each: the name, when (what the price has to do, such as Rises above $131.20) and then (what follows: the level wakes the PM for a look, or the PM acts on it at its next session, with the stop beside it when it set one). Then the next scheduled check and the brief it filed, in plain words. One sentence counts the rest: It left 3 things for its next session to check first. It also left 1 rule for itself until Fri Sep 11 and 1 open question. A brief links to itself on Research.
• Second opinion. What the Critic concluded about this session in a sentence (The Critic raised one concern about this session (1 serious). It is still open.), each concern as a card with one main action (Open the trade or Discuss in Consult), Show the Critic's note and More actions (Tell the desk, a note for the record, Dismiss), and one plain line on what its objections mean. A serious concern raised before an order goes out can hold it for your approval; every other concern is advice, shown to you and answered in writing by the next session. A card says It's your call now only while the order it is about still waits for you. On the Critic's own sessions this part links the batch it reviewed.
• What it looked at. The reads it made, one line each: the research brief, the market calendar, cash and buying power, positions, prices, the algorithm signals, the news. Three show at rest and the rest open from and N more reads. The names it checked are chips; tap one for its price and the usual actions. A wake-up leads with what woke it.

A session that cannot trade (a research brief, a daily review, a Consult answer, a Quant run) swaps What it decided for a part named for what it made, such as what the brief filed.

While a session is still running, a Live part at the top narrates it in plain words, newest line first. It is the same stream the Overview's mind card shows.

The note, as written

The rows in front are written by the console from the order's recorded fields: the symbol, the side, the size, the price, where the order is now, and a reason picked from a fixed list. They read the same way every day.

The agent's own sentences sit one click behind them. Each part that has any ends with a closed line: The PM's note, as written (or the Researcher's, or the Critic's). Open it to read, word for word, its note to you, the reason it wrote beside each order, and what it left for its next session. A finding card keeps the Critic's wording behind Show the Critic's note in the same way.

Nothing is hidden. You read the facts first and the prose when you want it.

Technical details

One closed line at the foot of the open session holds everything an engineer would want, and it appears nowhere else. A sentence in front of it says what is inside: It took 14 steps and 22 lookups and cost $0.62. Its full record is inside, for when you want to check its work. Open it for:

• the model, the step and lookup counts, the cost and budget, and the ids;
• Replay on this snapshot, which runs the session again with no side effects and compares the decision;
• the Transcript: its thinking between lookups, oldest first;
• the Tool calls, in order, with their results;
• the Orders as the risk rules checked them on a Portfolio Manager session: every check with its numbers, plus confidence and horizon. Desk › Today opens the latest session with this part already open;
• the write-up or journal as written, word for word;
• The briefing it read: a readable summary first, the raw card behind another click;
• the Second-opinion arithmetic: the Critic's track record, what it means that the Critic is advising only, and every concern with its written answer.

It stays closed in every state, a stopped session included: the header already says why it stopped, in words. Each part scrolls inside itself, so opening one never turns the page into a long scroll.

Agents

Desk › Agents is the roster, one card per agent. A card shows:

• the agent's figure, its name and a state pill: Idle, Running, Paused, Off or Coming soon. The Critic's pill reads Advising only: it is on, and its objections are advice that cannot hold an order;
• a line in its own voice and one sentence for the job;
• Runs on, the model in plain words (Fable 5.1, check-ins on Sonnet 5), and Works, when it runs;
• today's sessions and its next run;
• a Settings link. Someone with read only access sees View settings.

The name opens the agent's page.

The last tile is the Executor, marked Rules engine, not an AI. It checks every order any agent proposes against your limits, and it is the only part of the desk that can send one. See your limits goes to Risk.

One agent's page

The page opens with the agent's name, its state, the model it runs on, one sentence for the job and one line of today's facts. A strip of chips switches to another agent and keeps you on the tab you were reading. The tabs:

• Sessions: that agent's day, read the same way as Today (Portfolio Manager, Researcher, Critic). One closed line at the top, About the Portfolio Manager, holds how it talks and what it never does.
• Instructions: what its next session is told, as one page with a jump list down the side: Mandate, Limits, What it reads, Rules and Skills. Read only. See how the PM operates.
• Settings: the model, the budget, the schedule switches and the instruction modules. It is the one editor for that agent. See model policy.
• Memory and Schedule, on the Portfolio Manager's page only: what it remembers between sessions, and today's plan with the next trading days.

The Quant's page has no tabs: its runs and tuning come first, then its instructions, on one page.

On a phone

The bottom tab reads Desk. The open session is one column in reading order (decided, set up, second opinion, looked at) with Technical details closed at the end. The page never scrolls sideways. On an agent's page you move between agents with one menu beside Desk, and between days with the arrows beside the date.