Introduction

Are your discretionary overrides helping, or quietly taxing your P&L through slippage, late exits, and size creep? This guide gives you a clean, repeatable review process. Follow it weekly or bi-weekly to quantify performance, isolate root causes, and decide—objectively—when to lean on the algo and when human context truly adds value.

Core Framework (Step-By-Step)

1. Set the goal (before you start)
   • Choose what you’re optimizing: net P&L, max drawdown, profit factor, or consistency (expectancy per trade).
   • Fix a review window: weekly / bi-weekly / monthly.
2. Pull clean data
   • Export fills from broker: timestamp, symbol, side, qty, entry/exit, fees, slippage.
   • Tag each trade: ALGO, MANUAL, or ALGO+OVERRIDE. No untagged trades.
3. Normalize for fairness
   • Convert results to R (risk units) or % of capital.
   • Remove outliers from corporate actions/errors.
4. Compute core metrics (per bucket: ALGO vs MANUAL)
   • Win Rate, Avg Win, Avg Loss, Expectancy = (Win% × Avg Win) – (Loss% × Avg Loss)
   • Profit Factor = Gross Wins ÷ Gross Losses
   • Max Drawdown, Time in Market, Avg Hold Time, Slippage/Fees per trade
   • Risk-adjusted return: simple version = Net P&L ÷ Max DD (or use Sharpe)
5. Compare vs a passive benchmark
   • Use Nifty buy-and-hold or a sector ETF over the same period.
   • If both buckets underperform the benchmark after costs → pause and reassess.
6. Apply the 20% Gap Rule
   • If MANUAL underperforms ALGO by ≥ 20% for 2 consecutive review cycles → stop intervening, fix execution discipline (overrides, late exits, size creep).
   • If MANUAL outperforms ALGO by ≥ 20% → don’t rip out the system; extract the nuance (entry/exit/context filter), prototype it, and A/B test in a shadow algo before promotion.
7. Run a root-cause autopsy (fast)
   • Entry: was the trigger valid? late/missed?
   • Exit: premature cuts? profit give-back?
   • Sizing: over/undersized vs plan?
   • Regime: traded outside defined regime (trend/range/vol-shock)?
   • Costs: where are slippage/fees bleeding?
8. Execution audit
   • Latency signal → order, order type (limit/market), partial fills, plan vs impulse.
   • If slippage > planned by X bps, change order tactics/time windows.
9. Regime & timeframe sanity check
   • Align signals across primary timeframe and allowed higher/lower frames.
   • If conflicts are frequent, add a no-trade filter or reduce size.
10. Codify changes → test → promote
    • Convert useful manual insight into explicit rules.
    • Paper test for N trades (e.g., 50–100) → compare to live.
    • Promote only what wins; kill rules that fail.
11. Journal the learning (2 lines per trade)
    • “Why in?” “Why out?” “What will I repeat/avoid?”
    • Tag emotional overrides; aim to reduce them next cycle.
12. Decide & act (every review cycle)
    • Scale what works (increase size gradually).
    • Cut what doesn’t (disable, fix, or shelve).
    • If mix is worse than either alone, choose one and commit.

Quick Metrics Reference

⚠️ Risks & Common Mistakes

• Comparing raw P&L without normalizing for size, risk, or time in market.
• Letting one large outlier distort conclusions (clean those first).
• Confusing context edge with impulse overrides. If it’s not a rule, it’s noise.
• Promoting untested tweaks to production without a paper A/B test.

✅ Actionable Takeaways

• Tag everything: ALGO / MANUAL / ALGO+OVERRIDE.
• Review on a fixed cadence; compare buckets against a passive benchmark.
• Use the 20% Gap Rule for two consecutive cycles to decide intervene vs. automate.
• Codify any discretionary “edge” as a rule; paper test 50–100 trades before promoting.

Quick Heuristics to Remember

• If your edge comes from system rules, protect them: fewer overrides, more consistency.
• If your edge comes from context (news/structure), capture it as explicit filters—don’t rely on memory.
• Consistency > brilliance: a small positive expectancy done 100× beats one lucky hero trade.

Conclusion

Edge survives only what you can measure and repeat. Use this checklist to separate true skill from variance, scale what works, and cut what doesn’t. Want a plug-and-play sheet to run this review in minutes?

FAQs

Disclaimer: This content is for educational purposes only and is not investment advice. Markets involve risk. Do your own research and consider consulting a licensed advisor.