05Quant Agent Managers
Capital, managed by agents.
A preview of the agent desk we are building: quantitative agents designed to manage capital under a written mandate, explicit limits and a record of every decision. Shown as a conceptual model; execution runs in simulation only.
Loading the model…
The 3D model needs WebGL. The full roster is listed below.
The Vordeon desk · conceptual model
Drag to orbit · select an agent
Director
Director
Agent · orchestration
In development
- Mandate
- Sets the research agenda within the mandate written by people, and assigns work across the desk.
- May decide
- Which hypotheses are studied first
- How research time is allocated
- Cannot
- Approve a strategy
- Allocate capital or open a position
- Override a risk limit
Sample record format (illustrative) agenda.updated · priority list v12 reason: data coverage complete for 3 markets
Research
Researcher I
Agent · quantitative research
In development
- Mandate
- Turns observations into hypotheses that are written down before any test is run.
- May decide
- Propose a pre-registered hypothesis
- Request a dataset
- Cannot
- Change validation thresholds
- Test on data it has already seen
- Place orders
Sample record format (illustrative) hypothesis.registered · H-0xx out-of-sample cut fixed at registration
Research
Researcher II
Agent · quantitative research
In development
- Mandate
- Works a second line of research independently, so ideas are not filtered by a single view.
- May decide
- Propose a pre-registered hypothesis
- Challenge another researcher’s idea
- Cannot
- Change validation thresholds
- Place orders
Sample record format (illustrative) hypothesis.registered · H-0xx null model attached
Data
Data pipeline
Service · versioned data
In development
- Mandate
- Delivers versioned, point-in-time datasets with lineage, so every result traces back to its exact data.
- May decide
- Accept or quarantine a dataset by quality rules
- Cannot
- Rewrite a sealed dataset
Sample record format (illustrative) dataset.sealed · version v003 checksum recorded
Validation
Validation protocol
Code, not an agent
In development
- Mandate
- Tests every hypothesis out of sample, with realistic costs and explicit null models. Most ideas stop here.
- May decide
- Pass or reject a hypothesis by fixed rules
- Cannot
- Be overridden by any agent
Sample record format (illustrative) validation.completed · protocol v1.0.0 result: rejected · 4 checks failed
Committee
Committee chair
Agent · allocation committee
In development
- Mandate
- Chairs a committee of three. A proposal and its capital allocation move forward only with a majority.
- May decide
- Vote on validated proposals and their allocation
- Cannot
- Bypass validation
- Bypass risk limits
Sample record format (illustrative) committee.vote · approve 2 of 3 in favour
Committee
Committee member I
Agent · allocation committee
In development
- Mandate
- Votes independently on every validated proposal and its allocation, with a written reason.
- May decide
- Vote on validated proposals
- Cannot
- Vote on unvalidated ideas
Sample record format (illustrative) committee.vote · reject reason: overlap with an existing strategy
Committee
Committee member II
Agent · allocation committee
In development
- Mandate
- Votes independently on every validated proposal and its allocation, with a written reason.
- May decide
- Vote on validated proposals
- Cannot
- Vote on unvalidated ideas
Sample record format (illustrative) committee.vote · approve
Risk Control
Head of Risk
Agent · supervision
In development
- Mandate
- Watches exposure across every desk and can halt one desk, or all of them, at any time.
- May decide
- Halt one desk or every desk
- Request a review
- Cannot
- Open or enlarge a position
- Lift a limit set in code
Sample record format (illustrative) risk.halt.requested · desk FX reason: daily loss limit reached
Risk Control
Risk engine
Code · hard limits
In development
- Mandate
- Enforces limits per order, per desk and per day in code. Models cannot negotiate with it.
- May decide
- Reject an order before it is placed
- Close and lock a desk until the next session
- Cannot
- Be disabled by strategy code
Sample record format (illustrative) pre_trade.check · passed exposure within desk limit
Risk Control
Independent stop
Separate infrastructure
Planned
- Mandate
- A stop that runs apart from execution and cannot be switched off by any strategy.
- May decide
- Halt execution
- Cannot
- Be disabled by strategy code
Record format · not yet active
Desk Managers
Gold desk manager
Agent · capital manager
In development
- Mandate
- Manages the capital allocation assigned to the gold market within its limits, running approved strategies in simulation.
- May decide
- When to act within an approved strategy
- How to use its allocation within its limits
- Cannot
- Change its own strategy
- Change its own limits or allocation
Sample record format (illustrative) order.simulated · within limits
Desk Managers
FX desk manager
Agent · capital manager
In development
- Mandate
- Manages the capital allocation assigned to the currency market within its limits, running approved strategies in simulation.
- May decide
- When to act within an approved strategy
- How to use its allocation within its limits
- Cannot
- Change its own strategy
- Change its own limits or allocation
Sample record format (illustrative) order.simulated · within limits
Desk Managers
Index desk manager
Agent · capital manager
In development
- Mandate
- Manages the capital allocation assigned to the equity index market within its limits, running approved strategies in simulation.
- May decide
- When to act within an approved strategy
- How to use its allocation within its limits
- Cannot
- Change its own strategy
- Change its own limits or allocation
Sample record format (illustrative) order.simulated · within limits
Desk Managers
Digital assets desk manager
Agent · capital manager
In development
- Mandate
- Manages the capital allocation assigned to the digital assets market within its limits, running approved strategies in simulation.
- May decide
- When to act within an approved strategy
- How to use its allocation within its limits
- Cannot
- Change its own strategy
- Change its own limits or allocation
Sample record format (illustrative) order.simulated · within limits
Digital Twin
Digital twin
Independent replica
In development
- Mandate
- Mirrors every simulated position on separate infrastructure, so the state of each desk can always be checked independently.
- May decide
- Flag a mismatch between a desk and its replica
- Cannot
- Place orders
Sample record format (illustrative) twin.reconciled · 0 mismatches
Record
Record keeper
Ordered audit trail
In development
- Mandate
- Keeps an ordered, tamper-evident record of every event: who decided what, when and why.
- May decide
- Append events
- Cannot
- Edit or delete past events
Sample record format (illustrative) ledger.append · hash chained
Roster · one role per agent
No agent decides alone, and none can change its own limits.
Each manager administers the capital allocation assigned to it, within its limits. Research proposes, validation filters, the committee allocates, risk can stop everything, and the record keeps every step.
The roles mirror how an institution separates duties. Select an agent to see its mandate, what it may decide and what it cannot.
-
Sets the research agenda within the mandate written by people, and assigns work across the desk.
-
Turns observations into hypotheses that are written down before any test is run.
-
Works a second line of research independently, so ideas are not filtered by a single view.
-
Delivers versioned, point-in-time datasets with lineage, so every result traces back to its exact data.
-
Tests every hypothesis out of sample, with realistic costs and explicit null models. Most ideas stop here.
-
Chairs a committee of three. A proposal and its capital allocation move forward only with a majority.
-
Votes independently on every validated proposal and its allocation, with a written reason.
-
Votes independently on every validated proposal and its allocation, with a written reason.
-
Watches exposure across every desk and can halt one desk, or all of them, at any time.
-
Enforces limits per order, per desk and per day in code. Models cannot negotiate with it.
-
A stop that runs apart from execution and cannot be switched off by any strategy.
-
Manages the capital allocation assigned to the gold market within its limits, running approved strategies in simulation.
-
Manages the capital allocation assigned to the currency market within its limits, running approved strategies in simulation.
-
Manages the capital allocation assigned to the equity index market within its limits, running approved strategies in simulation.
-
Manages the capital allocation assigned to the digital assets market within its limits, running approved strategies in simulation.
-
Mirrors every simulated position on separate infrastructure, so the state of each desk can always be checked independently.
-
Keeps an ordered, tamper-evident record of every event: who decided what, when and why.
Decision cycle
From a written hypothesis to a recorded decision.
- 01
Data
Versioned, point-in-time data enters the desk.
- 02
Research
A researcher writes the hypothesis down before any test.
- 03
Validation
The protocol tests it out of sample, with realistic costs. Most ideas stop here.
- 04
Committee
Three members vote on the proposal and its capital allocation. A majority is required.
- 05
Risk Control
Limits are checked in code before anything moves.
- 06
Desk Managers
The desk manager administers its allocation within limits, in simulation only.
- 07
Digital Twin
An independent replica mirrors the position.
- 08
Record
Every step is written to an ordered record.
In development. Every agent and service on this page is in development or planned. Execution exists only in simulation, and there is no connection to the live market.