Upload Nemotron optimization response

optimization_cpu_nemotron/20260824T124918Z/nemotron_cycle_02.json

@@ -0,0 +1,47 @@
+{
+  "pure": {
+    "params": {
+      "lookback": 20,
+      "confirm_buffer": 0.03,
+      "min_sweep_fraction": 0.015,
+      "ema_fast": 20,
+      "ema_slow": 50,
+      "ema_slope_bars": 5,
+      "rsi_buy_max": 0.52,
+      "rsi_sell_min": 0.48,
+      "min_trend_score": 1.0,
+      "min_mtf_ratio": 0.65,
+      "require_ema_alignment": true,
+      "require_structure_break": false,
+      "require_mtf_alignment": true,
+      "require_m5_alignment": false
+    },
+    "reason": "Tighten confirm_buffer to 0.03 and add min_sweep_fraction 0.015 to filter noise while keeping 54k+ trade sample. Raise min_trend_score to 1.0 and min_mtf_ratio to 0.65 to align with profit_hit clusters showing stronger trend/MTF agreement. Enable require_ema_alignment and require_mtf_alignment (both false in top_pure) because profit_hits show higher weighted_mtf_ratio and ema_alignment consistency. Narrow RSI window to 0.52/0.48 to avoid extreme mean-reversion entries that appear in stop_losses. Keep structure_break false as it adds no edge in evidence."
+  },
+  "deep": {
+    "params": {
+      "min_sweep_depth": 0.035,
+      "max_sweep_depth": 0.65,
+      "min_close_back": 0.3,
+      "min_rejection_wick": 0.28,
+      "min_range_quality": 0.65,
+      "max_range_quality": 3.5,
+      "max_c1_body_ratio": 0.85,
+      "require_purity": true,
+      "require_m5_confirmation": true,
+      "ema_fast": 21,
+      "ema_slow": 51,
+      "ema_slope_bars": 9,
+      "rsi_buy_max": 0.54,
+      "rsi_sell_min": 0.46,
+      "min_trend_score": 1.5,
+      "min_mtf_ratio": 0.7,
+      "require_ema_alignment": true,
+      "require_structure_break": false,
+      "require_mtf_alignment": true,
+      "require_m5_alignment": true
+    },
+    "reason": "Deepen CRT quality filters: raise min_sweep_depth to 0.035, lower max_sweep_depth to 0.65, increase min_close_back to 0.3 and min_rejection_wick to 0.28 to match profit_hit rejection patterns. Tighten range_quality to 0.65-3.5 and max_c1_body_ratio to 0.85 to exclude messy candles prevalent in stop_losses. Enable require_purity and require_m5_confirmation (both false in top_deep) because selection shows negative expectancy without them. Raise min_trend_score to 1.5 and min_mtf_ratio to 0.7, enable all alignment booleans to force H1/H4/M5 agreement seen in profitable deep trades. Narrow RSI to 0.54/0.46 for directional clarity."
+  },
+  "overall_reason": "Pure strategy has robust 54k-trade backtest with positive expectancy across all R:R ratios; changes aim to improve selection consistency by adding moderate filters that align with profit_hit feature distributions (better MAE/MFE, higher MTF ratio, EMA alignment). Deep strategy suffers from small sample (7k backtest, 863 selection) and negative selection expectancy; changes significantly tighten CRT quality gates and add confirmation requirements to shift toward higher-quality setups evidenced in profit_hit trades. Both strategies move toward requiring multi-timeframe alignment and trend agreement, which the trade evidence shows separates profit_hits from stop_losses more cleanly than CRT parameters alone."
+}