Upload 2 files

app.py

@@ -60,6 +60,7 @@ def load_settings():
         "smart_exit_atr_multiplier": 3.0,
         "intelligent_tsl_pct": 0.2,
         "norm_lookback_years": 1,
+        'use_rolling_benchmark': False,
         "benchmark_rank": 99,
         "use_ma_floor_filter": True,
         "catcher_stop_pct": 0.03,
@@ -587,7 +588,8 @@ def run_backtest(data, params,
                  intelligent_tsl_pct=1.0,
                  analysis_start_date=None,
                  analysis_end_date=None,
-                 benchmark_lookback_years=None):
+                 benchmark_lookback_years=None,
+                 use_rolling_benchmark=True):
 
     df = data.copy()
     required_cols = ['Close']
@@ -642,7 +644,7 @@ def run_backtest(data, params,
     else: 
         df['RSI'], df['Volatility_p'], df['ADX'], df['ATR'] = np.nan, np.nan, np.nan, np.nan
 
-    # --- [NEW] MFI & SUPERTREND CALCULATIONS (Fixed) ---
+# --- [NEW] MFI & SUPERTREND CALCULATIONS (Fixed) ---
     if len(df) >= 14:
         # 1. MFI Calculation
         try:
@@ -750,36 +752,47 @@ def run_backtest(data, params,
         best_markov_setup=markov_setup 
     )
 
-    if long_score_95_percentile is None:
-        # If user specified a lookback (e.g. 1 year), use Rolling Window
-        if benchmark_lookback_years is not None and benchmark_lookback_years > 0:
-            window_days = int(benchmark_lookback_years * 252) # Approx trading days
-            # Calculate Rolling Percentile. 'min_periods' ensures we get data early in the chart.
-            long_95 = raw_long_score.rolling(window=window_days, min_periods=50).quantile(benchmark_rank).fillna(method='bfill')
-        else:
-            # Fallback to old "Global" method if no lookback specified
+    # 1. GLOBAL (STATIC) MODE - "The Crystal Ball"
+    # Used for "Elite Filtering" of open trades. 
+    if not use_rolling_benchmark:
+        if long_score_95_percentile is None:
             long_scores_gt_zero = raw_long_score[raw_long_score > 0]
             val = long_scores_gt_zero.quantile(benchmark_rank) if not long_scores_gt_zero.empty else 1.0
             long_95 = pd.Series(val, index=df.index)
-    else:
-        long_95 = pd.Series(long_score_95_percentile, index=df.index)
-
-    if short_score_95_percentile is None:
-        # Same logic for Short
-        if benchmark_lookback_years is not None and benchmark_lookback_years > 0:
-            window_days = int(benchmark_lookback_years * 252)
-            short_95 = raw_short_score.rolling(window=window_days, min_periods=50).quantile(benchmark_rank).fillna(method='bfill')
         else:
+            long_95 = pd.Series(long_score_95_percentile, index=df.index)
+
+        if short_score_95_percentile is None:
             short_scores_gt_zero = raw_short_score[raw_short_score > 0]
             val = short_scores_gt_zero.quantile(benchmark_rank) if not short_scores_gt_zero.empty else 1.0
-            short_95 = pd.Series(val, index=df.index)
+            short_95 = pd.Series(val, index=df.index) 
+        else:
+            short_95 = pd.Series(short_score_95_percentile, index=df.index)
+
+    # 2. ADAPTIVE (ROLLING) MODE - "Real World"
+    # Used for realistic historical simulation.
     else:
-        short_95 = pd.Series(short_score_95_percentile, index=df.index)
+        if long_score_95_percentile is None:
+            # Default to 1 year if lookback is missing
+            years = benchmark_lookback_years if (benchmark_lookback_years is not None and benchmark_lookback_years > 0) else 1
+            window_days = int(years * 252)
+            # Calculate Rolling Percentile
+            long_95 = raw_long_score.rolling(window=window_days, min_periods=50).quantile(benchmark_rank).fillna(method='bfill')
+        else:
+            long_95 = pd.Series(long_score_95_percentile, index=df.index)
+
+        if short_score_95_percentile is None:
+            years = benchmark_lookback_years if (benchmark_lookback_years is not None and benchmark_lookback_years > 0) else 1
+            window_days = int(years * 252)
+            short_95 = raw_short_score.rolling(window=window_days, min_periods=50).quantile(benchmark_rank).fillna(method='bfill')
+        else:
+            short_95 = pd.Series(short_score_95_percentile, index=df.index)
 
     # Safety: Ensure we don't divide by zero
     long_95 = long_95.replace(0, 1.0)
     short_95 = short_95.replace(0, 1.0)
-
+    
+    # Final Calculation (Removing 'if' check to avoid Series Ambiguity Error)
     df['long_confidence_score'] = (raw_long_score / long_95 * 100).clip(0, 100).fillna(0.0)
     df['short_confidence_score'] = (raw_short_score / short_95 * 100).clip(0, 100).fillna(0.0)
 
@@ -3728,6 +3741,13 @@ def main():
                 "• **1-2 Years:** Adaptive. Judges trades only against recent market volatility (good for changing regimes)."
                 "• **Default is set to 1 year as we are usually most interested in the recent year.")
     )
+    use_rolling_benchmark = st.sidebar.checkbox(
+        "Use Rolling Benchmark (Adaptive)", 
+        value=st.session_state.get('use_rolling_benchmark', True), 
+        key='widget_use_rolling_benchmark',
+        on_change=update_state,
+        help="CHECKED: Adaptive Mode. Benchmarks adapt to recent history. Realistic backtest.\nUNCHECKED: Global Mode (Crystal Ball). Uses ALL history to set one strict benchmark. Good for filtering Open Trades."
+    )
 
     # 2. Grade Benchmark - HIDDEN IN PRODUCTION MODE
     if not production_mode:
@@ -4332,6 +4352,7 @@ def main():
                 "smart_exit_atr_multiplier": st.session_state.get('smart_exit_atr_multiplier', 3.0),
                 "intelligent_tsl_pct": st.session_state.intelligent_tsl_pct / 100.0,
                 "norm_lookback_years": norm_lookback_years,
+                "use_rolling_benchmark": use_rolling_benchmark,
                 "benchmark_rank": benchmark_percentile_setting,
                 "use_vol": st.session_state.use_vol, "vol_w": st.session_state.vol_w,
                 "use_trend": st.session_state.use_trend, "trend_w": st.session_state.trend_w,
@@ -4592,9 +4613,10 @@ def main():
                             smart_exit_atr_multiplier=smart_atr_m,
                             intelligent_tsl_pct=intelligent_tsl,
                             benchmark_rank=benchmark_percentile_setting / 100.0,
-                            benchmark_lookback_years=norm_lookback_years, # <--- NEW PASSED VALUE
                             analysis_start_date=user_start_date,
-                            analysis_end_date=end_date
+                            analysis_end_date=end_date,
+                            benchmark_lookback_years=norm_lookback_years,
+                            use_rolling_benchmark=use_rolling_benchmark
                         )
                         st.session_state.single_ticker_results = {"long_pnl": long_pnl, "short_pnl": short_pnl, "avg_long_trade": avg_long_trade, "avg_short_trade": avg_short_trade, "results_df": results_df, "trades": trades}
                         st.session_state.open_trades_df = pd.DataFrame(open_trades) if open_trades else pd.DataFrame()
@@ -4659,9 +4681,10 @@ def main():
                                 smart_trailing_stop_pct=smart_trailing_stop, smart_exit_atr_period=smart_atr_p,
                                 smart_exit_atr_multiplier=smart_atr_m, intelligent_tsl_pct=intelligent_tsl,
                                 benchmark_rank=benchmark_percentile_setting / 100.0,
-                                benchmark_lookback_years=norm_lookback_years, # <--- NEW PASSED VALUE
                                 analysis_start_date=user_start_date,
-                                analysis_end_date=end_date
+                                analysis_end_date=end_date,
+                                benchmark_lookback_years=norm_lookback_years,
+                                use_rolling_benchmark=use_rolling_benchmark
                             )
 
                             if abs(long_pnl) > PROFIT_THRESHOLD or abs(short_pnl) > PROFIT_THRESHOLD or \

config.json

@@ -17,12 +17,14 @@
     "primary_driver": "Bollinger Bands",
     "exit_logic_type": "Intelligent (ADX/MACD/ATR)",
     "use_ma_floor_filter": true,
+    "catcher_stop_pct": 0.03,
     "exit_confidence_threshold": 40,
     "smart_trailing_stop_pct": 5.0,
     "smart_exit_atr_period": 14,
     "smart_exit_atr_multiplier": 3.0,
     "intelligent_tsl_pct": 0.2,
     "norm_lookback_years": 1,
+    "use_rolling_benchmark": false,
     "benchmark_rank": 99,
     "use_vol": false,
     "vol_w": 0.5,