the synthesis of trading systems architecture execution and feedback
A trading system is not a strategy. A strategy is just an idea. A trading system is the complete architecture that turns ideas into executed trades — signal generation, risk filtering, order dispatch, execution monitoring, and feedback loops.
This note synthesizes the Turtle Wiki’s trading systems hub into a unified architecture framework.
see also: gn21-three-layer-trading-system · gn30-setup-patterns · gn29-regime-identification · gn38-backtest-science-validation
the 7-layer trading system architecture
The full system, from data to execution:
Layer 1 — Data: Clean, normalized, point-in-time data. No look-ahead, no survivorship bias. Includes bid/ask, volume, open interest. Garbage in = garbage out applies most severely here.
Layer 2 — Strategy: Signal generation based on your edge. Entries, exits, filters. This is what most traders call “the strategy” — it’s actually just one layer.
Layer 3 — Confidence/Reliability: Regime filter. Assess whether current market conditions are favorable for your strategy. Don’t trade mean reversion in a trend. Don’t trade breakouts in a range.
Layer 4 — Risk: Position sizing. How much to bet given current account equity, volatility, and strategy confidence. This layer prevents ruin.
Layer 5 — Coordination (Meta): Decides which signals to execute when multiple strategies fire simultaneously. Resolves conflicts between strategies. This is where AI fits best — context-dependent decision making.
Layer 6 — Execution Bridge: Routes orders to the broker. Manages order types (market, limit, stop), slippage control, and execution timing.
Layer 7 — Monitoring: Tracks performance, slippage, execution quality, and system health. Feeds back to all other layers.
Most traders build layer 2 and stop. That’s like designing a car engine and calling it a vehicle.
the meta-layer: where AI belongs
The meta-layer (layer 5) is the best place to put AI/LLMs in your trading system. Not predicting prices — but coordinating:
- Assessing current regime (trending/ranging/transitioning)
- Adjusting confidence levels across different strategies
- Deciding which strategy gets priority when signals conflict
- Modifying position sizes based on context
AI is terrible at predicting market direction from unstructured data. It’s good at pattern classification and context-dependent decision making in structured environments. The meta-layer is a structured environment.
signal generation approaches
Rule-based (discretionary): Human reads the chart and decides. Flexible but inconsistent. Best for traders with deep experience.
Mechanical (automated): Fixed rules programmed into software. Consistent and testable. Best for strategies where edge is clearly defined.
Discretionary-with-system: The framework is fixed (risk rules, entry filters, position sizing) but the final entry decision is human. This hybrid often outperforms pure discretion or pure automation.
the feedback loop
A trading system must monitor itself:
- Track slippage vs modeled slippage quarterly
- Compare forward test results to backtest projections
- Measure strategy correlation (are all your strategies trending at once?)
- Check regime fit (is the strategy operating where it was designed to?)
When slippage exceeds modeled slippage by 20%, investigate. When forward test deviates from backtest by more than the Monte Carlo range, stop and review. When all strategies draw down simultaneously, assess macro regime — not individual strategy performance.
my take
I run a three-strategy portfolio: trend-following, mean reversion, and carry. Each has its own layer 1-4 logic. The meta-layer (me, with a spreadsheet) decides which gets priority based on the regime filter. I review the system quarterly.
The biggest lesson: system architecture matters more than any individual strategy. A mediocre strategy in a good system outperforms a great strategy in a bad system. Build the vehicle before obsessing about the engine.
linkage
- [[gn21-three-layer-trading-system]]
- [[gn30-setup-patterns]]
- [[gn29-regime-identification]]
- [[gn38-backtest-science-validation]]
- [[gn37-risk-management-deep-dive]]
ending questions
what is your complete trading system architecture? if you can’t list all 7 layers, you have a strategy, not a system.