Skip to content

Use Case: Finance & Autonomous Trading Governance

The RCT Platform adds a constitutional governance layer to autonomous financial agents. Every trade order, rebalance decision, or risk action passes through FDIA scoring and policy gates before execution.

Why RCT for Finance?

Autonomous trading systems fail when they have no constitutional constraint — flash crashes, runaway algos, and compliance violations all stem from unconstrained execution. RCT enforces:

  1. Intent-aligned scoring — ACCUMULATE-intent agents are scored against desired outcomes, not just immediate P&L
  2. Governance gates — orders above a threshold require human sign-off (Tier-6 consensus)
  3. Audit trail — every decision is cryptographically linked and replayable
  4. Deterministic replay — regulators can re-run any past decision with identical output

Risk-Tiered Order Routing

from core.fdia.fdia import FDIAScorer, FDIAWeights, NPCAction, NPCIntentType

scorer = FDIAScorer(weights=FDIAWeights(
    desire=0.40,
    intent=0.30,
    alignment=0.20,
    governance=0.10,
))

def evaluate_order(
    order_type: str,   # "buy_bond", "buy_crypto", "stop_loss", "hold_cash"
    amount_usd: float,
    portfolio_risk: float,  # 0.0 = low risk, 1.0 = max risk
) -> tuple:
    """Score a financial action against a conservative PROTECT intent."""

    investment_actions = [
        NPCAction("buy_bond",   "trade",   amount=amount_usd,   resource_id="bonds"),
        NPCAction("buy_crypto", "trade",   amount=amount_usd,   resource_id="crypto"),
        NPCAction("stop_loss",  "defend",  amount=amount_usd,   resource_id="portfolio"),
        NPCAction("hold_cash",  "idle",    amount=0.0),
    ]

    # Conservative investor: PROTECT intent
    ranked = scorer.rank_actions(
        agent_intent=NPCIntentType.PROTECT,
        actions=investment_actions,
        world_resources={"bonds": 5_000_000.0, "crypto": 200_000.0, "portfolio": 1_000_000.0},
        agent_reputation=max(0.1, 1.0 - portfolio_risk),
        governance_penalties={
            "buy_crypto": 0.30 if portfolio_risk > 0.7 else 0.0,  # penalise crypto in high-risk env
        } if portfolio_risk > 0.7 else None,
    )
    return ranked

# Example: risky market environment
orders = evaluate_order("buy_bond", 100_000.0, portfolio_risk=0.8)
best, score = orders[0]
print(f"Best action: {best.action_id} (score: {score:.4f})")
# → stop_loss or buy_bond (PROTECT intent + safe instrument = high score)

Governance Gates by Order Size

Order size Tier Approvers Threshold
< $10,000 Tier-S Auto-approve n/a
$10k – $1M Tier-4 3 of 4 75%
$1M – $10M Tier-6 4 of 6 67%
> $10M Tier-8 5 of 8 63% + chairman veto 
from signedai.core.registry import SignedAIRegistry, SignedAITier

def get_tier_for_amount(amount_usd: float) -> SignedAITier:
    if amount_usd < 10_000:
        return SignedAITier.TIER_S
    elif amount_usd < 1_000_000:
        return SignedAITier.TIER_4
    elif amount_usd < 10_000_000:
        return SignedAITier.TIER_6
    else:
        return SignedAITier.TIER_8

tier = get_tier_for_amount(2_500_000.0)   # → TIER_6
calc = SignedAIRegistry.calculate_consensus(tier, votes_for=4, votes_against=2)
print(f"Consensus: {calc.passed}")  # True — 4 ≥ threshold of 4

Policy: Block High-Risk Deploys

from rct_control_plane.default_policies import create_cost_cap_policy, create_risk_escalation_policy
from decimal import Decimal

# Block any automated action costing > $1M without human approval
evaluator.register_policy(create_cost_cap_policy(max_cost_usd=Decimal("1000000.00")))

# Escalate SYSTEMIC risk (e.g., full portfolio rebalance during market crash)
evaluator.register_policy(create_risk_escalation_policy())

Audit & Compliance

Every scored action produces a deterministic record:

rct compile "Buy $2.5M in US 10-year Treasuries" --user-id trader-007
rct evaluate --intent-id <id>
rct audit    <id> --verify  # cryptographic chain verification

The audit chain uses Ed25519-signed JITNA packets. Any past decision can be replayed with:

python -m rct.audit.replay --pillar 4 --verbose
# → Confirms 100% deterministic replay (Pillar 4)

FDIA Score Threshold for Finance

Decision type Minimum FDIA score
Routine order 0.65
Large order (> $1M) 0.70
Systemic risk action 0.80
Emergency stop-loss 0.60 (lower — fast path)

See Also