It watches the instruments you choose across every market your broker carries, argues both sides of each one, and opens the trade itself when it is confident enough — sized to your risk, stopped at the level that says it was wrong, and managed to the exit. In your own account. You set the envelope; it does the work.
Not a curated list of seven tickers. Everything your connected venues carry — and the venues carry more than most terminals bother to surface.
A venue declares what it can actually do — whether it shorts, whether it does fractions, whether it will hold a resting stop for that asset class — and the terminal never offers what the venue cannot honour. The desk will not open a position it cannot protect.
This is what happens every few minutes, on every instrument in your envelope, without anybody watching.
Live bars from your broker, turned into a ten-line structure card — trend, where price sits against VWAP and the moving averages, the real swing levels, volatility, volume — plus a picture of the chart. If the data is stale or the venue is down it stops there and says so. It never guesses a price.
A bull makes the strongest honest case. A bear red-teams that case, armed with the desk's own memory of every past loss on the same setup. A desk head reads both and rules — long, short, or hold, with a level at which the idea is simply wrong.
Three independent tests, and all three must pass: the desk head's conviction clears your floor, the bull's case was strong enough to be worth testing, and the red team failed to find enough against it. A confident ruling beside a strong bear is a disagreement, and a disagreement is a hold. Most looks end in a hold — that is the job working.
Arithmetic — never the model — turns "wrong below 63,940" into a position size: your equity, your risk percentage, the distance to that level, scaled down when volatility is high. The stop goes to the broker with the entry, and if the venue cannot hold one the desk places it the moment the fill lands and closes the position outright if it cannot.
At +1R it banks a third and moves the stop to breakeven — the trade can no longer lose. At +2R it banks another third. After that the stop trails the high-water mark and only ever ratchets in your favour, so the last third runs as far as the move does.
Every action and every pass, explained at the time, in its own words, with the full debate attached. You read the morning's reasoning instead of reconstructing it from a fill.
Autonomy without a boundary is not a product, it is a liability. The boundary is one screen and it is the only thing you have to get right.
Named instruments, from any market your venues carry. Nothing outside the list, ever — the desk researches everything and acts only inside it.
Risk per trade, a daily loss ceiling, how many positions at once, and the confidence below which it does not act. Ask for more than the box allows and it is refused outright, never quietly reduced.
Every mandate expires — days, not forever. Renewing is a decision you make again, awake, rather than one you made once and forgot.
One button. No approval, no confirmation, no failure mode — it never refuses. Troy can stop it for you by text, too. Turning it on is the only thing an AI here cannot do for you.
Four guarantees. Each one exists because a real system got it wrong, and three of them were found in the system this replaced.
Every bar carries where it came from and how old it is, all the way to the pixel. Stale or unavailable data stops the desk on that instrument and says so — it does not fill the gap with something plausible.
THE PREDECESSOR FABRICATED A MARKET AND TRADED ITWhere a venue cannot carry protection on the entry, the desk places it the moment the fill lands — and closes the position within fifteen seconds if it cannot. Not a retry loop. A rule.
ITS CRYPTO TRADES HAD NO STOP AT ALLEvery message about your money is written by the code that received the broker's confirmation. If it says your stop moved to breakeven, the broker said so first.
ITS "BREAKEVEN STOP" WAS A LOG LINEOne button, no approval, no confirmation, no failure mode. It does not close your positions — their stops stay resting at the broker, which is where they are safest.
A CONTROL THAT CAN BE BLOCKED IS NOT ONEYour chief of staff carries the desk in your pocket — six tools, and the dangerous ones left out on purpose.
What you are holding, what a market is doing, and why the desk did — or refused to do — something. Over text, over WhatsApp, over a call.
READ ONLYStop the desk opening anything new, or close a position outright. Closing is the safe direction, so it is the one write Troy keeps.
SAFE WRITESThe desk trades by itself all day. What it cannot do is decide to start — that is one button, on one screen, by you. A chat agent improvising during an incident is the one failure this design refuses to allow.
BY DESIGNWritten here rather than discovered later.
MARKETS_LIVE unset → every order is composed, checked, sized and
recorded, and nothing reaches a broker.
/health reports all four from outside, so a box in rehearsal is
visible rather than assumed. This department gives no financial advice: it reports
what the desk did and why.