Multi-agent architecture.

This chapter states the architecture. The full blueprint — with proposal schema, kill-switch conditions, seven-stage deployment roadmap and the exact obligations of each role — is published as a companion piece at agentteam.co.uk/share-trading.html. Read the two together.

Section ISeparation of powers

The organisational design deliberately prevents one agent from researching, sizing, approving and executing its own idea. The research and control departments contribute to the same proposal, but the risk function retains an independent veto and the broker connection is isolated.

Why separation matters:

• Language models can produce persuasive explanations even when the underlying evidence is weak.
• Several agents may repeat the same assumption, so a majority vote is not a reliable safety mechanism.
• Risk limits must remain enforceable even when models, prompts or data feeds fail.
• Broker credentials should never be available to research agents.
• Every decision should be reproducible from stored inputs, calculations, policies and approvals.

Agents recommend. Code calculates. Risk controls veto. Humans authorise. The execution service trades.

Section IIThe eleven roles

Section IIIThe controlled decision workflow

Every proposed order passes through the same auditable sequence. A failed or missing stage stops the process rather than being silently ignored.

1. Define permitted universe.
2. Validate and timestamp data.
3. Run independent analyses.
4. Red-team the thesis.
5. Create formal trade proposal.
6. Backtest and validate.
7. Assess portfolio impact.
8. Risk approval or veto.
9. Human approval.
10. Execute and reconcile.
11. Monitor and audit.

Section IVFormal trade proposal

Agents exchange structured records. The following fields are mandatory so that risk, execution and audit services can validate the proposal automatically.

{
  "instrument": {"symbol": "ABC", "exchange": "LSE", "ccy": "GBP"},
  "action": "BUY",
  "strategy_id": "MOMENTUM_01_V3",
  "thesis": "...",
  "evidence": ["...", "..."],
  "catalyst": "...",
  "holding_period": {"min_days": 5, "max_days": 20},
  "entry_condition": "price <= 12.45 GBP",
  "invalidation_condition": "close > 13.20 for 2 sessions",
  "exit_plan": ["take_profit @ 13.80", "stop_loss @ 11.90", "time_stop 20 sessions"],
  "proposed_allocation": {"capital": 5000, "max_loss": 250},
  "portfolio_impact": {"sector_delta": "+1.2% cons_disc", "correlation": 0.34},
  "residual_risks": ["..."],
  "confidence": 0.62,
  "data_timestamp": "2026-07-13T09:15:00Z",
  "approvals": {"research": "R-2018", "validation": "V-0912", "risk": "RISK-18267", "human": "HUM-90411"}
}

Section VRead the full blueprint

---

Next: Risk & governance →