What the Quant may change on its own, and what it never can
On Auto the Quant desk manages the machine settings of your book inside a mandate the code derives from your posture and one word of freedom (Tight, Balanced or Free hand). Every automatic change is a card with the exact before and after, a notice, a grade and a one-click Revert; it reverts itself unless it holds; the desk pauses itself on a drawdown, a run of misses, bad data or a halt. Gates, enforced rules, your approval line, credentials, halts and anything per symbol are never its to touch.
- Who it is for
- Owners & operators
- Reading time
- 7 min read
- Updated
- Where you set it
- The Tuning row of the card on Strategy › Profile: Off · Recommend · Auto, and on Auto one word of freedom — Tight · Balanced · Free hand
- What Auto does
- Applies the Quant's cards that fall inside your mandate at the next session boundary, in either direction; everything else waits for you and says why
- Where the mandate comes from
- The code derives it from your and the freedom you picked — nobody types thresholds; the card prints it as sentences, with the numbers behind Details
- What you always get
- A card with the exact before and after, a notice (email or push per your notification settings), an entry in the audit log naming the card, a grade at 5 / 20 / 60 days, a one-click Revert for 7 days
- What undoes itself
- An automatic change reverts after 7 days unless you keep it (Keep on its card); the desk pauses itself on a deep , two misses in a row, stale data or a
- What it never touches
- Gates, enforced rules, your , credentials, halts, parameters, and anything per symbol — names and sizes are the portfolio manager's
You asked for a desk that manages the book for you when you say so ("in auto mode i DO want the quant to automatically make shifts for me, otherwise it's not very automated") without a form full of thresholds to fill in. So Auto is one switch and one word. The switch says whether the Quant's cards apply themselves; the word says how much room they get; the code turns that word and your posture into every number.
The three freedoms
All three are read against the posture you last applied on Strategy › Profile. The figures below are for a Moderate book; a Conservative or Aggressive book shifts the ranges with it, and the Details disclosure under the sentences always shows yours.
| Tight | Balanced (default) | Free hand | |
|---|---|---|---|
| weights | up to 10 points a week between algorithms | up to 20 points a week; may take a losing member off or give a one a small starting weight; may change the (which changes no score) | up to 35 points a week; same member changes |
| Posture | stays where you put it | may step one rung down (Conservative) and back; one step a week | may also step one rung up (Aggressive); one step a week |
| Risk limits | only ever come in from your posture's values | move inside the Conservative–Moderate range, never looser than Moderate's values; two moves a week; any limit may always come in | move inside Conservative–Aggressive, never looser than Aggressive's; three moves a week |
| Halt thresholds | never loosened | never loosened by a knob; follow a posture step you allowed | same |
| Sleeve split | follows the posture | up to 5 points from the posture's split, never past the exposure cap; one move a week | up to 10 points; two moves a week |
| Guidance | yours | may add short book-wide soft notes in fixed wording (two a week); a note about one name always waits for you | may also retire soft notes that cost more than they save |
| Its own schedule | yours | yours | may move between nightly and weekly |
Tight fits inside Balanced, Balanced inside Free hand, and all three inside the bounds the code enforces for everyone: a property the test suite checks for every posture, and that the auto-safety evaluation re-checks with thousands of random sequences: no series of admitted changes ever leaves the envelope.
Custom. Behind Advanced you can replace a derived value with your own (a wider blend budget, a wider sleeve band). The mandate is then tagged overridden with the derived value beside it and a Reset to derived. Widening anything asks for your code and the words MORE FREEDOM, exactly like picking a freer word does; narrowing is one click.
How a card meets the mandate
The Quant writes the same cards on Auto as on Recommend; it does not decide what is inside. When a card is due, the console's own code re-reads the values the book holds now and asks the mandate: is this kind allowed at this freedom, is the new value inside the range, is this week's budget still open, has the confidence guard passed? Inside → it applies at the session boundary through the same editor you would use by hand, with quant:auto/<card id> as the actor. Outside → it stays a card that waits for you, with the reason in one line ("any one position 12% is looser than your mandate allows (10%)", "gates are always yours"). A value that moved since the card was drafted parks the card for you instead of overwriting your change.
A step toward more risk inside the range you armed is allowed on Auto (that is what arming the range with your code meant) and the notice says so in words: "loosens risk, inside your mandate." On Recommend the same card would ask for your code and MORE RISK.
Never silent, never stuck
- Notice. Every automatic change raises a notification (Automatic change by the Quant in your notification settings) and appears on Strategy › Recommendations under Applied by itself, newest first, with its grade as it arrives. Overview carries a chip (Quant: auto · balanced · last change 2h ago) that links there.
- Revert. One click for 7 days, through the same editor; undoing a tightening is a loosening and asks for your code like any other.
- Self-revert. Seven days after an automatic change, it reverts itself unless you kept it: Keep on the card (or Accept) is the , one click, logged, and Revert stays available for its window. The self-revert never reads the grades (a few sessions of P&L cannot tell a good change from noise); the grades stay on the card for your review. Outside market hours only; waits while the book is halted; never overwrites a value you have changed since; noted once on the card either way. A change that reverted unacknowledged is not applied automatically again for 30 days: if the evidence still supports it, it comes back as a card that waits for you.
- Self-pause. The desk stops applying anything (cards keep coming) when the book has drawn down 70% of the way to its drawdown halt, when two automatic changes in a row missed their interval, when its last run failed or read stale data, or when trading is halted. You get a notice. Pause auto is always one click for you, on the Tuning row or at the top of Recommendations; resuming after the desk paused itself asks for your code and the words RESUME AUTO.
- Budget. The weekly allowances above are counted over a trailing seven days across every automatic change, so many small runs cannot add up to a large move.
Who owns which numbers
Auto never blurs the desk's layers ("a desk of agents that work on Wall Street, working for us"), each with its own job:
| Layer | Owns | Never |
|---|---|---|
| every risk check on every order; halts | — nothing overrides it | |
| You | enforced rules, the approval line, the posture you declare, the freedom word, credentials | — |
| Quant | machine settings: which algorithms are on and their weights, their display order, the split between sleeves, posture and appetite limits inside your mandate, soft guidance drafts, its own schedule | a symbol, a position size, a , an order — it has no tool for any of them |
| Portfolio manager | which names, the , the , how big — inside the sleeve and the limits it is given, which it sees but cannot edit | a weight, a limit, the split |
| Trader | how and when the manager's intent reaches the Executor | choosing names or sizes |
When two layers disagree the higher one wins and the lower one is told why: Executor limits, then enforced guidance, then the Quant's settings, then the manager's targets, then the trader's tactics.
What the Quant can never do, in any mode
- Trade, or pick a symbol or a size.
- Make an algorithm a gate, or lower one.
- Write, change or retire an enforced guidance rule.
- Move your approval line (the first row of the Autonomy card).
- Loosen either halt threshold with a knob card.
- Touch credentials, the kill switch, or a halt.
- Change an algorithm's parameters.
- Move any value outside the bounds the code defines, or past the most permissive posture preset.
- Widen its own mandate: the freedom word and every custom range are yours, behind your code.