mirror of
https://github.com/RomySaputraSihananda/ares.git
synced 2026-08-16 20:28:07 +00:00
romysaputrasihanandaandClaude Sonnet 4.6
a78dd26cf0
Vendor domain and mt5-client from hermes into crates/ so the repo can be pushed to GitHub without external path dependencies. Also includes all bug fixes from autonomous session: - impulse SL fix (SL at impulse candle low/high, not zone edge) - pip_size fix for 5-decimal pairs - fill_ok premature cancellation fix - TIMEOUT_CANDLES, MIN_FVG_PIPS, MIN_SL_PIPS env vars Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
171 lines
5.0 KiB
Markdown
171 lines
5.0 KiB
Markdown
# Autonomous Backtest Session — 2026-06-09
|
||
|
||
Dikerjakan saat user tidur. Ringkasan semua bug fix, backtest, dan temuan.
|
||
|
||
---
|
||
|
||
## Bug Fixes
|
||
|
||
### 1. `!fill_ok` premature FVG cancellation (kritikal)
|
||
|
||
**Masalah:** Saat zona FVG disentuh candle tapi entry midpoint belum tercapai, kode langsung
|
||
membatalkan (`pending_fvg = None`). Akibatnya banyak setup yang valid ikut dibuang.
|
||
|
||
**Fix:** Saat `is_touched && !fill_ok`, FVG tetap aktif dan dicoba di candle berikutnya.
|
||
FVG hanya dibatalkan saat:
|
||
- Expiry tercapai (`i >= expiry_idx`) → `missed_fills += 1`
|
||
- EMA filter tidak sesuai
|
||
- SL distance terlalu kecil
|
||
|
||
**Dampak:**
|
||
```
|
||
Sebelum fix: 993 trades, WR=48.3%, PF=1.33, Return=+314.7%
|
||
Sesudah fix: 1268 trades, WR=50.3%, PF=1.43, Return=+1183.3%
|
||
```
|
||
|
||
### 2. `missed_fills` counter dipindah ke expiry
|
||
|
||
Sebelumnya counter dihitung setiap candle zona disentuh tanpa fill. Sekarang hanya dihitung
|
||
sekali saat setup expired tanpa pernah terisi.
|
||
|
||
---
|
||
|
||
## Fitur Baru
|
||
|
||
### 3. `TIMEOUT_CANDLES` (env var)
|
||
|
||
Sama seperti Hermes — force close trade setelah N candles. Default = 0 (disabled).
|
||
|
||
**Test result:** Hampir tidak berpengaruh di XAUUSDm:
|
||
```
|
||
TIMEOUT=0 → PF=1.43, Return=1183%
|
||
TIMEOUT=24 → PF=1.42, Return=1088%
|
||
TIMEOUT=48 → PF=1.43, Return=1132%
|
||
```
|
||
Artinya trade XAU selalu resolve (SL/TP) dalam <24 M5 candles (2 jam).
|
||
|
||
---
|
||
|
||
## Parameter Sweep XAUUSDm
|
||
|
||
### BODY_PCT_MIN (body/range minimum)
|
||
```
|
||
0.50 → Trades=1351, PF=1.50, Return=1808%
|
||
0.55 → Trades=1320, PF=1.46, Return=1509%
|
||
0.60 → Trades=1268, PF=1.43, Return=1183% ← default
|
||
0.65 → Trades=1191, PF=1.43, Return=854%
|
||
0.70 → Trades=1102, PF=1.40, Return=575%
|
||
```
|
||
Lower = more trades, slightly better PF. 0.60 adalah titik keseimbangan.
|
||
|
||
### MIN_RR
|
||
```
|
||
1.2 → WR=55.7%, PF=1.39, Return=784%
|
||
1.5 → WR=50.3%, PF=1.43, Return=1183% ← default
|
||
2.0 → WR=41.2%, PF=1.32, Return=853%
|
||
2.5 → WR=35.2%, PF=1.27, Return=688%
|
||
3.0 → WR=32.1%, PF=1.33, Return=1289%
|
||
```
|
||
MIN_RR=1.5 memberikan balance terbaik. MIN_RR=3.0 menarik tapi WR 32% beresiko tinggi.
|
||
|
||
### FVG_EXPIRY_CANDLES
|
||
```
|
||
3 → PF=1.21, Return=170%
|
||
5 → PF=1.31, Return=401%
|
||
10 → PF=1.43, Return=1183% ← default
|
||
15 → PF=1.44, Return=972%
|
||
20 → PF=1.45, Return=796%
|
||
```
|
||
10 adalah sweet spot — cukup waktu untuk fill, tidak terlalu lama jadi stale.
|
||
|
||
### EMA_PERIOD (KRITIKAL)
|
||
```
|
||
0 → WR=41.2%, PF=0.97, Return=-18% ← RUGI tanpa filter!
|
||
10 → WR=53.5%, PF=1.70, Return=1723%
|
||
15 → WR=50.1%, PF=1.46, Return=880%
|
||
20 → WR=50.3%, PF=1.43, Return=1183% ← default
|
||
30 → WR=47.6%, PF=1.26, Return=499%
|
||
50 → WR=44.4%, PF=1.11, Return=107%
|
||
```
|
||
**EMA filter sangat kritikal** — tanpanya strategi rugi. EMA10 terlihat terbaik tapi
|
||
lebih beresiko overfitting. EMA20 adalah pilihan konservatif yang terbukti profitable
|
||
di EMA range 10-50.
|
||
|
||
### SL_BUFFER
|
||
```
|
||
0 → WR=50.3%, PF=1.43, Return=1183% ← default (terbaik)
|
||
0.5 → WR=48.0%, PF=1.29, Return=482%
|
||
1.0 → WR=47.0%, PF=1.27, Return=348%
|
||
2.0 → WR=45.2%, PF=1.20, Return=120%
|
||
```
|
||
Buffer di luar impulse candle malah mengurangi performa. SL tepat di impulse low/high adalah optimal.
|
||
|
||
---
|
||
|
||
## Hermes Validation
|
||
|
||
Build dan backtest berhasil. BREAKEVEN_SL_1R=true bekerja dengan benar:
|
||
- GBPUSDm H1: 102 BE-SL exits, PF=4.27 — fitur berjalan normal.
|
||
|
||
---
|
||
|
||
## Recommended Config (XAUUSDm)
|
||
|
||
```env
|
||
SYMBOL=XAUUSDm
|
||
TIMEFRAME=M5
|
||
BACKTEST_BALANCE=600
|
||
BACKTEST_CANDLES=50000
|
||
RISK_PCT=0.01
|
||
|
||
BODY_PCT_MIN=0.6
|
||
CLOSE_PCT_MIN=0.8
|
||
FVG_EXPIRY_CANDLES=10
|
||
MIN_FVG_PIPS=1
|
||
MIN_SL_PIPS=5
|
||
SL_BUFFER=0
|
||
MIN_RR=1.5
|
||
TIMEOUT_CANDLES=0
|
||
|
||
EMA_PERIOD=20 # kritikal — jangan disable
|
||
|
||
COMMISSION_PER_LOT=7
|
||
SLIPPAGE_POINTS=5
|
||
SPREAD_OVERRIDE=0
|
||
```
|
||
|
||
**Hasil:** 1268 trades, WR=50.3%, PF=1.43, Return=+1183%, Max DD=-$550 (dari peak ~$8250 = 6.7%)
|
||
|
||
---
|
||
|
||
## Peringatan / Risiko
|
||
|
||
1. **Data hanya Sep 2025–Jun 2026** — XAU naik 65% dalam periode ini (bull run sangat kuat).
|
||
Strategy dengan EMA filter otomatis bias LONG. Performance di bear market atau ranging market
|
||
belum diketahui.
|
||
|
||
2. **EMA adalah kunci** — tanpa EMA, strategi rugi. Ini menunjukkan edge bukan dari FVG pattern
|
||
saja, tapi kombinasi FVG + trend momentum. Hati-hati jika trend berbalik.
|
||
|
||
3. **Compounding effect besar** — return +1183% karena compounding 1268 trades. Real-world
|
||
perlu ditest dengan RISK_PCT lebih kecil (0.005) di awal.
|
||
|
||
4. **Post-fix multi-pair results (setelah fill_ok bug fix):**
|
||
```
|
||
GBPJPYm → WR=50.5%, PF=1.03, Return=+13.3% (was +10.3%)
|
||
EURUSDm → WR=49.5%, PF=1.15, Return=+9.0% (was -5.6% → NOW POSITIVE!)
|
||
GBPUSDm → WR=42.3%, PF=0.85, Return=-19.8% (still negative)
|
||
XAUUSDm → WR=50.3%, PF=1.43, Return=+1183.3% (flagship)
|
||
```
|
||
Bug fix meningkatkan EURUSDm dari negatif ke positif!
|
||
|
||
---
|
||
|
||
## Next Steps (untuk user)
|
||
|
||
1. Live micro-lot test di XAUUSDm dengan RISK_PCT=0.005
|
||
2. Cari data XAU 2023–2024 untuk out-of-sample validation
|
||
3. Explore apakah EURUSDm bisa profitable dengan EMA=10 (mungkin lebih baik dari EMA=20)
|
||
4. GBPUSDm masih negatif (PF=0.85) — tidak direkomendasikan untuk sekarang
|
||
5. Monitor performa 2 minggu pertama live dengan DD limit manual
|