The Quant desk
A desk that proposes blend, sleeve and parameter changes as Inbox cards you apply or, on Auto, applies bounded blend changes itself at the next session boundary, revertible for 7 days.
- Who it is for
- Owners & operators
- Reading time
- 3 min read
- Updated
The Quant is a desk role that measures rather than trades. After the close it runs a deterministic engine over realized attribution, the backtests and how the live has done, and turns that into : small, specific changes to how the are combined. It never places or sizes an order, never reads news, briefs or chat, and never touches the limits the enforces.
Where it lives
Its one home is the Quant desk card on Desk › Quant: the on/off switch· Desk › Quant, the Manual / Auto dial· Strategy › Profile › Autonomy › Tuning (Off / Recommend / Auto), the scope· Desk › Quant › Scope (blend · sleeve advice · parameters · advice; parameters are opt-in), the model· Desk › Quant › Model and cadence· Desk › Quant › Cadence (nightly after the by default), the code bounds it may never cross, when the last run happened and the desk's own note, and Run now. Everywhere else the desk is read-only: Algorithms shows a chip for the dial, a pointer ("2 desk suggestions affect this blend") and provenance on any value the desk changed; the Algorithm blend· Algorithms itself is still edited there and only there.
Manual: cards you apply
Manual is the default. Every proposal is a Desk suggestion card in the Inbox (filter Desk suggestions) with the same shape as a trade card: what it wants to change, the engine's rationale and the Quant's note, a preview of the effect, "Why these numbers?" (the walk-forward table with the chosen row), Ask the Quant why, and Apply or Dismiss. Applying writes the change through the same audited route you use by hand; dismissing is remembered so the Quant does not repeat itself. An undecided card expires after 24 hours and says so. Advice cards (posture, sleeve targets) have no Apply: they deep-link to Strategy › Profile, where those settings are edited.
Auto: bounded, revertible, never Layer 1
Raising the dial to Auto needs your and the word AUTO typed out; turning the desk off drops it back to Manual. On Auto, bounded blend changes apply themselves at the next PM session boundary. They are applied by the console's own service, never by the model's process, which has no permission to write configuration at all. Before applying, the live values are re-read; if anything moved since the proposal was made it comes back to you as a manual card instead. Nothing older than 24 hours is applied; a single algorithm never moves more than 0.10 and a run never moves more than 0.25 in total; advice, parameter changes and anything set to Required are never auto-applied. Every auto change is audited with the Quant as the actor, raises a notification, and can be reverted for 7 days from the card or from the provenance line on Algorithms ("changed by Quant (auto) · revert").
Warningif the applier is not running, the card says "overdue" and the next run raises an alert; scheduled cards then simply expire. Nothing is applied late.
What it never proposes
Layer-1 limits (per-order caps, exposure caps, daily-loss and halts, thresholds, the , , the ) stay on Risk › Limits and its neighbours, owner-only, and keep applying on the next Executor tick when you change them. The Quant never sets an algorithm to Required; that ceremony is yours alone. And it never proposes a trade. The PM learns of applied changes through a Quant desk section of its next (evidence it may cite or disagree with, not memory), never through chat.