qc-order-forensics（质量控制命令取证）

QC Order Forensics

Turn raw backtest order data into actionable diagnostic reports. Feed the output directly to an LLM for strategy improvement decisions.

When to use

Use this skill when the user has completed a backtest and wants to understand why the strategy performed the way it did. Typical triggers:

• "Analyze my backtest results"
• "Why did my strategy lose money?"
• "Show me the trade quality report"
• "Diagnose this backtest"

How it works

The OrderForensics class takes an orders.csv (and optionally a result.json) and produces a multi-section diagnostic report:

from forensics import OrderForensics

forensics = OrderForensics("path/to/orders.csv", "path/to/result.json")
report = forensics.full_diagnosis()
print(report)  # LLM-readable text report

Or from CLI:

python3 -c "
from forensics import OrderForensics
f = OrderForensics('orders.csv', 'result.json')
print(f.full_diagnosis())
"

Report Sections

1. Key Statistics

Extracted from result.json: Net Profit, Sharpe Ratio, Drawdown, Win Rate, Expectancy, Alpha, Beta, Total Fees.

2. Trade Quality Detection

• Total order count (options vs equity)
• Buy/sell split with average fill prices
• Number of unique underlyings traded
• Zero rate: % of options sold at <= $0.05 (expired worthless)
• Windfall trades: Count of trades exceeding 100%, 400%, 1000% ROI

3. ROI Breakdown by Contract

• Per-symbol: Cost, Return, Net Profit, ROI%
• Top 5 winners and Top 5 losers
• Overall option ROI, winning vs losing symbol counts
• Average winner ROI vs average loser ROI

4. Monthly Cashflow

• Net cash flow per month (buys + sells)
• Cumulative cashflow curve
• Trade count and buy/sell density per month
• Identifies bleeding months vs profitable months

5. Drawdown Death Causes

• Detects consecutive loss streaks exceeding $1,000
• Reports start/end month and cumulative loss for each streak
• Sorted by severity (worst first)

6. Yearly Breakdown

• Annual net cashflow, trade count, and ticker diversity
• Identifies which years were profitable vs destructive

Input Format

orders.csv columns (standard QC export):

orderId, symbol, type, direction, quantity, fillPrice, fillQty, fee, status, submitTime, fillTime, tag

The parser automatically separates options from equity (QQQ) orders by extracting the underlying from the symbol field.

Output

A structured plain-text report using emoji markers for quick visual scanning. Designed to be directly consumed by LLMs for follow-up analysis and strategy iteration decisions.

Key Metrics to Watch

Rules

• Input CSV must be in standard QuantConnect export format with columns: orderId, symbol, type, direction, quantity, fillPrice, fillQty, fee, status, submitTime, fillTime, tag. Missing columns will cause silent misclassification.
• Do not manually edit the orders CSV before analysis. Changing fill prices, removing rows, or reordering columns will produce misleading diagnostics.
• The equity_symbol parameter (default "QQQ") must match your strategy's anchor equity. If your strategy holds SPY instead of QQQ, pass --equity-symbol SPY or the equity/options split will be wrong.
• Zero rate above 65% is not automatically bad for OTM lottery strategies — it's expected. Interpret in context of the strategy's design, not as an absolute red flag.
• This tool analyzes closed trades only. Open positions at backtest end are excluded from ROI calculations.