How the desk decides when to think

A small book cannot afford a model session every hour, and it should not have to go quiet either. The desk separates watching, which is free, from thinking, which costs money, and spends its thinking where it matters.

Watches

Positions, armed stops, watchlist names, open ideas, the PM's price levels, the market; code, no model

Reads

One cheap first read per event: how large, whether it is bad for a holder, one sentence; a fifth of a cent

Decides

Code, not the model: wake the PM, note it for the end-of-day check, or let it go, reason written down

The one dial

The model budget in basis points of equity a year; derived by default, lower is one click, raising asks for your code

Never waits

Stops, halts and the daily loss halt are the Executor's code and act whatever the budget says

See it at

Desk once switched on: what the desk watched, read and declined, and the budget left

Watching is free, thinking is priced

Code sentinels observe halts, gaps at the open, large moves, unusual volatility, price levels, results and filings, analyst actions and macro releases. When one of them touches a name you hold or watch, the desk writes down the observation and grades it. The grade is a single short call to a small model that sees only the fact, never your book, so one read serves every book on the platform that the same event touches.

What happens next is a rule in code, in this order:

1. 1A protective event on a position (price near an armed stop, a halt on a name you own, a step down in the book's drawdown) wakes the PM at once. If the budget is already spent the session still runs and the row says it went over; the platform covers that look so one rough morning does not silence the rest of your week.
2. 2A severe event on any position (a guidance cut, an FDA decision, an offering, a halt, a gap of 5 percent or more) wakes the PM during market hours, and before the open when it is bad news.
3. 3A material event on a large position (results, a lawsuit, a 3 to 7 percent move on a name that is 8 percent of the book or more) wakes the PM when today's budget covers the session and what is at stake outweighs its price. Otherwise it becomes a note.
4. 4Everything smaller, and material events on names you only watch, become notes. The end-of-day check reads the day's notes for free, in one block, and says what it would do about them.
5. 5Noise (commentary, options-flow chatter, a stranger's earnings) is let go without a read, and the row says why.

Every one of these is a row you can read: the event, the first read's sentence, the verdict in words, the estimated cost, the budget after.

The budget is one dial

The model budget is the most the desk's model sessions may cost your book in a year, as a share of its equity. On a 10,000 dollar paper book the default 100 basis points is about 40 cents a trading day: enough for continuous first reads and roughly one session; unspent days roll forward within the week and what is left on Friday goes to the weekly review (a day on which the platform itself was down earns nothing: the desk did no watching that day). Three things about the dial:

• Lower is one click. Higher, up to the platform's paper ceiling of 300 basis points, asks for your authenticator code because it spends more of your money.
• Above the ceiling the desk only accepts a number you mark sponsored and confirm by typing I PAY. Every page that shows the budget then says it is sponsored, and every session paid from it carries the mark.
• The scheduled sessions (the morning brief, the end-of-day check) answer to the same yearly number through the schedule, not through this purse: the page shows what they cost this week beside the attention budget, so a busy schedule never silently empties it and nothing about their cost is hidden.

What the desk will not do to save money

It never delays a stop, a halt or an exit for a model or a budget: those are code and run on their own clock. It never wakes the PM for an event on a name you neither hold nor watch. It never spends on a look whose price is larger than anything the look could change, and it writes that arithmetic down when it declines.

What a woken session is told, once

When an event convenes the PM, the first line of its wake says what the look is budgeted at, its priority, how hard it should think (a protective look gets the PM's full effort; lesser events get less, never more than you configured) and what is expected of it: review the protection on a position, reassess a position, pre-plan the open, act on a level you asked about, or revisit an open idea. It is said once. Nothing later in the session counts the money down, and the session's spend is capped by the platform whatever the model does.

When the desk pauses itself, and what needs you

If nothing woke the desk for a week and you did not use it in that time (no question to the desk, no trade decided, no alert acknowledged, no guidance written), the desk stops paying for first reads on names you only watch and puts one notice in your Inbox saying so. Anything on a position, and every protective event, still wakes it, and the next thing you do with the desk resumes the rest. There is no setting for this.

The Desk also keeps a short list of things only you can decide: material events on your positions that went to the end-of-day check because the week's budget was spent (raise the budget, sponsor it, or leave it), and protective looks the platform had to cover. The Inbox shows how many are waiting. It also shows how often a woken session did nothing, by the kind of action it was expected to take, so a lane that wakes and never acts is visible as a number rather than a feeling.