Appendix A: Live Trading Logging Standards

In the practice of quantitative trading, raw trading logs are the only bridge connecting "simulation" and "live trading." This appendix details the minimum log checklist that a professional quant system must record, aimed at helping you collect high-quality data that can directly feed back into strategy training (especially reinforcement learning RL).

A1.1 Core Data Structure

A complete trading record should cover at least three dimensions of information to ensure every trade is traceable.

A1.1.1 Order Level Information

Record details of every instruction sent to the exchange:

• Identifiers: order_id (globally unique ID), symbol (trading code).
• Order Parameters: side (buy/sell), order_type (market/limit), order_price (order price), order_qty (order quantity).
• Lifecycle: submit_ts (order timestamp), cancel_ts (cancel timestamp, if applicable).

A1.1.2 Fill Level Information

Since one order may correspond to multiple fills, every specific fill detail must be recorded:

• Association Identifiers: fill_id, order_id.
• Fill Details: fill_price (fill unit price), fill_qty (fill quantity), fill_ts (millisecond-precision fill timestamp).

A1.1.3 Derived Key Metrics

This is the most critical part for feeding back to strategy, calculated from the above base data:

• Expected Price (expected_price): Theoretical price when signal triggered.
• Slippage Cost (slippage): Deviation between actual average fill price and theoretical price.
• Execution Latency (latency_ms): Time from order submission to first fill.
• Fill Rate (fill_ratio): Ratio of actual filled quantity to planned quantity.

A1.2 Cost Accounting and Market Snapshot

A1.2.1 Financial P&L and Friction Costs

• Explicit Costs: commission (trading fees), tax (stamp duty/regulatory fees).
• Final Returns: realized_pnl (realized profit and loss).

A1.2.2 Market State at Decision Time

Record market background at moment of order placement, for analyzing environment's impact on execution:

• Bar Data: bar_ts, bar_open/high/low/close, bar_vwap.
• Volatility Indicators: atr_5min.
• (Optional) Liquidity Depth: Recommend recording at least top-of-book bid/ask (bid1/ask1).

A1.3 Agent Decision Metadata (RL-Specific)

To enable RL models to learn true causal relationships, decision context must be recorded:

• Version Identifier: agent_id / version.
• Action Details: action (buy/sell/hold), target_position (target position).
• Confidence Measure: confidence or model output prediction score.

A1.4 Summary: Why These Fields Are Essential

A standardized log checklist is not just for financial reconciliation; its core value lies in:

1. Revealing slippage truth: Precisely distinguish "signal error" from "execution deviation."
2. Training execution logic: Provide real Reward signals for execution-type RL (such as execution delay and fill rate).
3. Breaking data dependency: Without expensive Level-2 data, rely solely on live logs to build your own execution simulator.

Core Principle: If you don't completely record the full process of "order -> fill -> delay -> failure," your live data has no value for strategy evolution.