05Risk Management

Risk is designed in, not added on.

Discipline is an engineering problem. The risk architecture we are building defines limits before strategies, and supervision before automation.

Decision architecture

Five layers and an independent supervisor.

Independent risk supervisionConceptual architecture. In the execution environment, five layers stand between an idea and a market: mandate, limits, pre-trade controls, execution and post-trade review, whose lessons update the mandate. A separate supervision environment, with its own infrastructure and permissions, observes every layer and holds an independent stop that can halt execution.EXECUTION ENVIRONMENTSUPERVISION ENVIRONMENTSEPARATE INFRASTRUCTURE01MandateWritten first02LimitsExposure, loss, size03Pre-tradeChecks instructions04ExecutionSimulation only05Post-tradeReconcile and reviewEVERY LAYER IS OBSERVEDReal-time supervisionPositions, twin, system healthIndependent stopCannot be disabled by strategy codeHALTLESSONS UPDATE THE MANDATEOperationalIn developmentPlanned
Independent risk supervisionConceptual architecture. In the execution environment, five layers stand between an idea and a market: mandate, limits, pre-trade controls, execution and post-trade review, whose lessons update the mandate. A separate supervision environment, with its own infrastructure and permissions, observes every layer and holds an independent stop that can halt execution.EXECUTION ENVIRONMENT01MandateWritten first02LimitsExposure, loss, size03Pre-tradeChecks instructions04ExecutionSimulation only05Post-tradeReconcile and reviewLESSONS UPDATE THE MANDATESUPERVISION ENVIRONMENTReal-time supervisionPositions, twin, system healthIndependent stopCannot be disabled by strategyOperationalIn developmentPlanned
Supervision runs in a separate environment, observes every layer and holds a stop that strategy code cannot disable. Conceptual architecture · status as of October 2026

The layers

Between an idea and a market.

  1. 01

    Mandate

    What a strategy or agent is permitted to do, written before it exists.

  2. 02

    Limits

    Exposure, concentration and loss limits defined per strategy and in aggregate.

  3. 03

    Pre-trade controls

    Every instruction is to be checked against limits and data-quality rules before it can be executed.

  4. 04

    Real-time supervision

    Independent monitoring of positions, the digital twin and system health, with the authority to stop activity.

  5. 05

    Post-trade review

    Every outcome is reconciled, recorded and reviewed, and the lessons feed back into the mandate.

Independence

The supervisor never answers to the executor.

Supervision is designed to run on separate infrastructure, with separate permissions and its own authority to halt activity.

A failure in a strategy, a model or a data feed must never be able to disable the controls designed to contain it.

Controls

Designed into the system, not into a policy document.

Control

Independent stop mechanisms

Kill switches designed to operate outside strategy code, so they cannot be disabled by the systems they control.

Control

Data-quality gates

Stale, missing or anomalous data is designed to halt decisions instead of being silently used.

Control

Change control

No model, parameter or limit changes without review, versioning and a recorded rationale.

Control

Complete audit trail

An ordered, tamper-evident record of every decision, so any event can be reconstructed after the fact.