Trading Strategy Reverse Engineering Tool
Trade examples · Sierra timestamps must be UTC
Loading the May–June 2026 TradeActivityLog…
Preparing trade examples…
Held-out entry ranking · 213 trades
What the model actually recovers
Random orderingreference
Top 5% recovered5.00%
Mean rank50.00%
Average signatureprice + clock
Top 5% recovered9.39%
Mean rank39.83%
Price-only pairwise29 of 213 entries
Top 5% recovered13.62%
Mean rank34.03%
Clock-only shortcutinvalid test design
Top 5% recovered99.06%
Mean rank1.30%
Top 5% recovered means the true entry landed among the highest-scored 5% of bars. Mean rank is its average percentile; 50% is random. The clock-only result is crossed out because each test window ended at the known entry, letting time identify the last bar without learning price logic.
Actual vs model entry and exit — every held-out trade
actual entry / exit model entry / exitshading = actual holding window
Open to load trade cards.
Actual vs learned
What changed?
| Market | Actual trades | Actual hold | Actual move | Learned matches | Learned hold | Learned move | Best loss | Setup |
|---|
Recurrence cadence
Median-gap extrapolation
Matched occurrences
| Signal | Market | Frame | Price | Loss | RSI | Rel. volume | Pattern |
|---|
Notes
- Export the Trade Activity Log in UTC so fills align with market bars. The browser submits only the latest 1,000 closed trades; the raw file is never uploaded.
- Window features, multi-timeframe bars, candlestick labels, rolling levels, trend slopes, outcome evaluation, and loss ranking run on the server. Price loss compares observable entry setups; timing loss compares learned entry and exit times of day. Outcomes are displayed but never used to select a match.
- Projected dates extrapolate the median interval between the closest matches. They are rough cadence estimates—not calibrated probabilities, future-price forecasts, or trade signals.