initial: import DOLPHIN baseline 2026-04-21 from dolphinng5_predict working tree

Includes core prod + GREEN/BLUE subsystems:
- prod/ (BLUE harness, configs, scripts, docs)
- nautilus_dolphin/ (GREEN Nautilus-native impl + dvae/ preserved)
- adaptive_exit/ (AEM engine + models/bucket_assignments.pkl)
- Observability/ (EsoF advisor, TUI, dashboards)
- external_factors/ (EsoF producer)
- mc_forewarning_qlabs_fork/ (MC regime/envelope)

Excludes runtime caches, logs, backups, and reproducible artifacts per .gitignore.

adaptive_exit/train.py

@@ -0,0 +1,75 @@
+"""
+Offline training script.
+
+Run once to build bucket assignments and train continuation models:
+
+    cd /mnt/dolphinng5_predict
+    siloqy-env python adaptive_exit/train.py
+
+Artifacts written:
+    adaptive_exit/models/bucket_assignments.pkl
+    adaptive_exit/models/continuation_models.pkl
+    adaptive_exit/models/training_data.parquet   (optional, for audit)
+"""
+import argparse
+import os
+import sys
+
+sys.path.insert(0, "/mnt/dolphinng5_predict")
+
+from adaptive_exit.bucket_engine import build_buckets
+from adaptive_exit.continuation_model import ContinuationModelBank
+from adaptive_exit.data_pipeline import build_training_data
+
+_MODELS_DIR = os.path.join(os.path.dirname(__file__), "models")
+_TRAIN_DATA_PATH = os.path.join(_MODELS_DIR, "training_data.parquet")
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Train adaptive exit models")
+    parser.add_argument("--k", type=int, default=None, help="Force bucket count (default: auto)")
+    parser.add_argument("--save-data", action="store_true", help="Save training parquet for audit")
+    parser.add_argument("--force-rebuild", action="store_true", help="Rebuild buckets even if cached")
+    parser.add_argument("--vbt-dir", default="/mnt/dolphinng5_predict/vbt_cache",
+                        help="VBT parquet dir for training data generation")
+    parser.add_argument("--klines-dir", default="/mnt/dolphin_training/data/vbt_cache_klines",
+                        help="1m klines dir for asset bucketing")
+    args = parser.parse_args()
+
+    os.makedirs(_MODELS_DIR, exist_ok=True)
+
+    # ── Step 1: Build asset buckets ──────────────────────────────────────────
+    print("\n=== STEP 1: Asset Bucketing ===")
+    bucket_data = build_buckets(
+        klines_dir=args.klines_dir,
+        k_override=args.k,
+        force_rebuild=args.force_rebuild,
+    )
+    print(f"Buckets: {bucket_data['n_buckets']} | Assets: {len(bucket_data['assignments'])}")
+
+    # ── Step 2: Build training data from price series ────────────────────────
+    print("\n=== STEP 2: Generate MAE/MFE Training Data ===")
+    df = build_training_data(
+        bucket_assignments=bucket_data["assignments"],
+        vbt_dir=args.vbt_dir,
+        use_obf_ch=False,   # OBF is live-only (13 days); zero-fill training, bolt on at Phase 2
+    )
+    print(f"Training data shape: {df.shape}")
+    print(f"Bucket distribution:\n{df.groupby('bucket_id').size().describe()}")
+    print(f"Continuation rate: {df['continuation'].mean():.3f}")
+
+    if args.save_data:
+        df.to_parquet(_TRAIN_DATA_PATH)
+        print(f"Training data saved → {_TRAIN_DATA_PATH}")
+
+    # ── Step 3: Train continuation models ────────────────────────────────────
+    print("\n=== STEP 3: Train Continuation Models ===")
+    bank = ContinuationModelBank()
+    bank.train(df)
+    bank.save()
+    print(f"\nModel summary: {bank.summary()}")
+    print("\nDone.")
+
+
+if __name__ == "__main__":
+    main()