47 lines
1.9 KiB
Markdown
47 lines
1.9 KiB
Markdown
---
|
|
name: tradelab-backtest-interpretation
|
|
description: Use when reviewing TradeLab backtest output so recommendations are grounded in getBacktestResults data and mapped to concrete strategy changes.
|
|
---
|
|
|
|
# TradeLab Backtest Interpretation
|
|
|
|
## Overview
|
|
Use `getBacktestResults` output as the source of truth for strategy feedback. Interpret risk and return metrics first, then propose specific code-level changes through the strategy workflow.
|
|
|
|
## When to Use
|
|
- User asks why strategy performance is good or bad.
|
|
- User asks how to improve a strategy after a backtest.
|
|
- A new `currentBacktestId` is available in context.
|
|
|
|
## Non-Optional Rules
|
|
1. Retrieve metrics via `getBacktestResults` before proposing optimizations.
|
|
2. Anchor every recommendation to returned fields, not guesses.
|
|
3. Separate analysis into: performance, risk, and action plan.
|
|
4. If code changes are needed, route them through full-class `saveStrategy` flow.
|
|
|
|
## Metric Interpretation Baselines
|
|
- `sharpeRatio`: `< 1.0` weak, `1.5-2.0` good, `> 2.0` elite.
|
|
- `profitFactor`: `< 1.2` fragile, `1.5-2.5` robust, `> 3.0` possible overfit.
|
|
- `maxDrawdown`: flag if `> 15%`; suggest volatility controls or regime filters.
|
|
- `winRate < 40%`: verify payoff ratio; add entry-quality filters.
|
|
|
|
## Tool Usage Pattern
|
|
```json
|
|
{
|
|
"backtestId": "optional-explicit-id",
|
|
"strategyId": "optional-fallback-id",
|
|
"includeTrades": false,
|
|
"metricFilter": ["performance", "risk", "ratios"]
|
|
}
|
|
```
|
|
|
|
## Output Pattern
|
|
1. State key facts from tool output.
|
|
2. Explain likely failure mode (entries, exits, regime mismatch, risk sizing).
|
|
3. Propose 2-3 prioritized improvements tied to those facts.
|
|
4. If user approves changes, produce full updated class through `saveStrategy`.
|
|
|
|
## Common Mistakes
|
|
- Recommending changes without calling `getBacktestResults`.
|
|
- Giving generic advice not tied to metric values.
|
|
- Treating high `profitFactor` without checking overfit risk.
|