Every strategy, against real data.
Round 3: 154 backtests · 7 majors · real Pepperstone history (≈5y H1, ≈8y H4) · per-pair spreads · full risk stack + the live 3-bar re-entry cooldown. Round 2 shown for comparison; Round 1 and Round 2 reports are archived in full.
On top of Round 2's SL/TP simulation, the engine now waits 3 bars after any exit before re-entering — matching the live loop's REENTRY_COOLDOWN_BARS. This removed most of Round 2's repeated-stop-out artifact: the churn-heavy strategies improved noticeably (MA Crossover median −43% → −33%, RSI −17% → −8%) while the leaders were unchanged — a good sign that their edges are real, not artifacts.
Three rounds, one consistent story: RevertX has a small real edge; Sentinel is a USDJPY specialist; BreakoutGuard's rare H4 trades are high quality; everything else loses after costs at default parameters. The stability table below tests whether the leaders held up year by year — Lesson 9's sealed-exam idea applied across time.
The scoreboard
| strategy | PF r2→r3 | median r2→r3 | positive runs | trades | verdict |
|---|---|---|---|---|---|
| Revert X | 1.14→1.14 | -0.3%→-0.4% | 3/14 | 292 | PROMISING |
| RSI | 0.90→0.91 | -17.1%→-8.3% | 3/14 | 19,129 | MARGINAL |
| Combined | 0.87→0.88 | -38.0%→-28.8% | 1/14 | 23,752 | NEEDS WORK |
| MA Crossover | 0.87→0.87 | -43.0%→-32.7% | 0/14 | 18,797 | NEEDS WORK |
| Trend Rider | 0.86→0.86 | -1.5%→-1.5% | 4/14 | 2,161 | MARGINAL |
| Sentinel | 0.85→0.85 | -25.1%→-17.9% | 2/14 | 19,507 | SPECIALIST |
| Breakout Guard | 0.84→0.83 | -1.7%→-1.6% | 3/14 | 2,401 | MARGINAL |
| Adaptive Edge | 0.83→0.82 | -37.0%→-25.7% | 0/14 | 21,238 | NEEDS WORK |
| Grid | 0.63→0.60 | -7.7%→-1.9% | 1/14 | 4,410 | NEEDS WORK |
| Pulse Scalper | 0.56→0.55 | -12.4%→-10.1% | 0/14 | 9,648 | NEEDS WORK |
| Scalping | 0.48→0.00 | -0.2%→-0.1% | 3/14 | 48 | NEEDS WORK |
Stability check — the leaders, year by year
Profit factor per calendar year (≥1.00 = profitable year; — = no data, shorter H1 history). A real edge should show up in most years, not one golden stretch — Lesson 8's movie-trailer test.
| run | 2018 | 2019 | 2020 | 2021 | 2022 | 2023 | 2024 | 2025 | 2026 | good yrs |
|---|---|---|---|---|---|---|---|---|---|---|
| revert x · GBPUSD H1 | — | — | — | 9.90 | 0.72 | 0.42 | 6.37 | 2.13 | 9.90 | 4/6 |
| sentinel · USDJPY H1 | — | — | — | 1.34 | 1.06 | 1.12 | 1.23 | 1.03 | 0.78 | 5/6 |
| sentinel · USDJPY H4 | 0.62 | 1.05 | 0.96 | 0.97 | 1.51 | 0.98 | 1.27 | 0.78 | 1.00 | 3/9 |
| breakout guard · USDJPY H4 | 0.00 | 1.95 | 2.54 | 2.75 | 0.62 | 1.15 | 0.83 | 3.98 | 9.90 | 6/9 |
| rsi · EURUSD H4 | 1.99 | 0.60 | 0.93 | 1.01 | 1.08 | 1.64 | 0.87 | 1.17 | 1.08 | 6/9 |
| trend rider · USDCAD H4 | 1.77 | 1.28 | 0.78 | 0.42 | 2.00 | 0.88 | 0.45 | 1.76 | 9.90 | 5/9 |
Top 10 individual runs
| run | return | PF | win% | max DD | trades |
|---|---|---|---|---|---|
| Sentinel · USDJPY H1 | +9.9% | 1.08 | 35% | 5.0% | 1920 |
| Sentinel · USDJPY H4 | +6.3% | 1.07 | 34% | 5.8% | 804 |
| RSI · EURUSD H4 | +5.6% | 1.06 | 59% | 7.8% | 810 |
| RSI · USDCAD H4 | +3.8% | 1.04 | 58% | 11.4% | 751 |
| RSI · GBPUSD H4 | +2.7% | 1.02 | 59% | 17.0% | 769 |
| Breakout Guard · USDJPY H4 | +1.8% | 1.54 | 50% | 0.9% | 46 |
| Combined · NZDUSD H4 | +1.7% | 1.01 | 45% | 12.3% | 977 |
| Breakout Guard · GBPUSD H4 | +1.2% | 1.54 | 48% | 1.0% | 23 |
| Trend Rider · USDCAD H4 | +1.0% | 1.11 | 48% | 2.2% | 87 |
| Revert X · GBPUSD H1 | +0.9% | 1.75 | 64% | 0.6% | 36 |
1) The leaders (RevertX, Sentinel/USDJPY, BreakoutGuard-H4, RSI-H4) go to parameter tuning with the overfitting guard — plateaus, not needles. 2) The rest are benched pending redesign. 3) Survivors graduate to forward testing on demo (Academy Lesson 11's protocol). The Academy's Lessons 8–9 are the manual for reading everything on this page.