Universal Trading Brain
Multi-armed decision engine with 40+ indicator sockets, SPA regime selection, RL action optimization, and ML-Bridge risk shaping — all flowing through a single unified pipeline. Click any architecture card below for full system documentation.
API Health
—
checking...
WebSocket
—
—
LLM Orchestrator
—
—
Broker Port
Paper
dummy adapter active
System Architecture — click any card for full documentation
◇
Indicator Registry
35 Python indicators (5 structure + 30 signals) with typed feature maps and weighted socket aggregation — every vote now scaled by that engine's learned per-regime reliability.
click for full indicator breakdown →
⚡
SPA Engine
4-arm bandit with Thompson/UCB/ε-greedy/Softmax selection, trauma compression, adaptive DataSocket normalization — shadow P&L now pays real trading costs.
click for theory, usage, configuration →
◎
RL Agent
Q-learning + policy gradient + actor-critic + ensemble with TD(λ), cost-aware rewards, plus an SB3 deep-RL pipeline (PPO/A2C/SAC) training in the real orchestrator gym env.
click for theory, usage, configuration →
⬡
ML-SPA Bridge
Signal translation, Regime Decoder reconciliation, external socket aggregation, and 7-factor risk computation.
click for theory, usage, configuration →
▣
Order Management
Market/Limit/Stop/Trailing/Bracket orders with RiskGate, PositionTracker, and multi-broker routing.
click for theory, usage, configuration →
⬢
Universal Ports
Adaptive broker adapters (Paper default; connect your own broker — Alpaca supported) and a data registry where Yahoo works out of the box: your Datasets → your APIs → Yahoo → Synthetic.
click for theory, usage, configuration →
◆
Strategy Orchestration
Regime Decoder, reliability-weighted Confluence, Execution Engine with the exit trio (breakeven trail · regime-flip · time-stop), seeded Backtester with ⚔️ A/B, and Optimizer/AGE.
click for theory, usage, configuration →
⇗
Signal Evolution & Roadmap
The quality doctrine: what shipped (P0 pack, verified by seeded A/B) and the full upgrade path — meta-labeling, correlation-aware confluence, promotion pipeline, portfolio governors.
click for the shipped pack + every planned upgrade →
✦
Multi-Agent LLM
7 roles under a two-tier command structure — a Director (cloud or best local) writes the brief, an Executor (fast local) carries it out with tools. Ollama/GGUF, memory, voice, Workspace personas.
click for theory, usage, configuration →
system.log
[boot] ✓ FastAPI server on :8787
[boot] ✓ SPA engine initialized (4 arms)
[boot] ✓ RL agent ready (Q-table + policy net)
[boot] ✓ Bridge state initialized
[boot] ◆ DafeRCM indicator registered