Perception
Market context
Market-data adapters, multi-timeframe pulse generation, and regime/volatility labels prepare the evidence packet for decision-making.
Architecture
Audit-first decision machinery.
FFE is organized around a simple rule: every decision must be explainable after the fact. The public map below is intentionally high-level and omits credentials, private infrastructure, and live routing details.
Market data + sentiment
│
▼
Multi-timeframe pulse ──► Regime / volatility classification
│ │
▼ ▼
Decision context ───────► LLM debate / candidate actions
│ │
▼ ▼
Policy scoring / rerank ─► Risk gatekeeper ─► Execution adapter
│ │
└──────────────► Decision store / audit spine ◄──── Trade monitorPublic components
Reasoning
Debate and candidate actions
AI providers can produce candidate actions. Debate metadata and decision origin are preserved so later reports can separate judged decisions from skip or fallback paths.
Control
Risk gatekeeper
Risk checks, legal-action enforcement, stale-data protection, and audit logging constrain the decision before any execution adapter is considered.
Memory
Persistence and evaluation
Decision artifacts, trade outcomes, and experiment metadata support replay, regression checks, and public-safe summary generation.
Audit spine invariants
- Decision origin, regime, confidence bucket, volatility bucket, and action family remain reconstructable.
- Policy experiments record whether a feature gate fired, what pocket matched, and how scores changed.
- Risk gates and legal-action controls are not weakened by public-showcase or M3 experiment work.
- Public diagrams never publish live routing endpoints, credentials, positions, or private host details.