Upload Quasar_axrvi_ranker.py

Quasar_axrvi_ranker.py

@@ -378,7 +378,7 @@ SYMBOL_MAP = {
     "1HZ100V":   "V100_1s",
     "CRASH500":  "CRASH500",
     "CRASH1000": "CRASH1000",
-    "stpRNG2":   "STEP200",   # CONFIRMED: live Deriv WS API returns "stpRNG2" (STPRNG200 does not exist)
+    "stpRNG2":   "STEP200",   # CONFIRMED: live Deriv WS API returns "stpRNG2"
 }
 SYMBOL_MAP_REVERSE = {v: k for k, v in SYMBOL_MAP.items()}
 
@@ -408,7 +408,7 @@ ASSET_REGISTRY: Dict[str, dict] = {
     "V100_1s":  {"symbol": "1HZ100V",   "base_vol": 100.0, "max_pos": 0.002},
     "CRASH500": {"symbol": "CRASH500",  "base_vol": 50.0,  "max_pos": 0.003},
     "CRASH1000":{"symbol": "CRASH1000", "base_vol": 100.0, "max_pos": 0.002},
-    "STEP200":  {"symbol": "stpRNG2",   "base_vol": 200.0, "max_pos": 0.002},  # CONFIRMED: stpRNG2 is the live Deriv symbol
+    "STEP200":  {"symbol": "stpRNG2",   "base_vol": 200.0, "max_pos": 0.002},  # CONFIRMED live symbol
 }
 
 # ── Per-asset MULTUP/MULTDOWN multipliers (BROKER-VALIDATED ACCEPTABLE RANGES) ──
@@ -468,16 +468,16 @@ ASSET_ACCEPTABLE_MULTIPLIERS: Dict[str, List[int]] = {
 # Stop-loss as fraction of stake per asset (capped to protect $6 account)
 # e.g. 0.50 = close when $0.50 of the $1 stake is lost
 ASSET_STOP_LOSS_FRAC: Dict[str, float] = {
-    "V25":      0.60,   # slow mover — give it room
+    "V25":      0.60,
     "V30_1s":   0.55,
-    "V50":      0.55,   # FIX 2: was missing — added explicit entry
+    "V50":      0.55,
     "V50_1s":   0.55,
     "V75":      0.50,
     "V75_1s":   0.45,
-    "JD100":    0.50,   # FIX 2: was missing — added explicit entry
-    "V100":     0.50,   # FIX 2: was missing — added explicit entry
+    "JD100":    0.50,
+    "V100":     0.50,
     "V100_1s":  0.40,
-    "CRASH500": 0.50,   # hard spike cap — bail at 50%
+    "CRASH500": 0.50,
     "CRASH1000":0.50,
     "STEP200":  0.55,
 }
@@ -486,12 +486,12 @@ ASSET_STOP_LOSS_FRAC: Dict[str, float] = {
 ASSET_TAKE_PROFIT_FRAC: Dict[str, float] = {
     "V25":      1.00,
     "V30_1s":   0.90,
-    "V50":      0.90,   # FIX 3: was missing — added explicit entry
+    "V50":      0.90,
     "V50_1s":   0.90,
     "V75":      0.80,
     "V75_1s":   0.75,
-    "JD100":    0.80,   # FIX 3: was missing — added explicit entry
-    "V100":     0.80,   # FIX 3: was missing — added explicit entry
+    "JD100":    0.80,
+    "V100":     0.80,
     "V100_1s":  0.70,
     "CRASH500": 0.80,
     "CRASH1000":0.80,
@@ -521,8 +521,8 @@ GAMMA           = 0.99
 LAMBDA_RANK     = 0.4
 LAMBDA_RISK     = 0.3
 REPLAY_CAPACITY = 10_000
-TRAIN_BATCH     = 2    # FIX 4: Lowered to 2 — trains after only 2 closed trades (faster warmup with 12 assets)
-TRAIN_EVERY_N   = 2   # FIX 4: Check buffer every 2 rank cycles for faster training start
+TRAIN_BATCH     = 2    # FIX: Lowered to 2 — trains after 2 closed trades
+TRAIN_EVERY_N   = 2   # FIX: Check every 2 rank cycles
 
 # Connection
 WS_RECONNECT_DELAY   = 5
@@ -829,15 +829,29 @@ class AssetSnapshot:
             self.training_steps = int(training.get("training_steps", self.training_steps))
 
         if voting:
-            raw_signal = voting.get("dominant_signal", self.dominant_signal)
+            # ── CORE FIX 6: prefer latest_signal (per-message, always fresh) ──
+            # dominant_signal = accumulated ALL-TIME vote majority → stale.
+            # With 7.5K historical SELL votes, a fresh BUY is invisible.
+            # latest_signal = set by hub on EVERY voting message received →
+            # always reflects the engine's most recent inference output.
+            _raw_latest   = voting.get("latest_signal",   "")
+            _raw_dominant = voting.get("dominant_signal", self.dominant_signal)
+            # Use latest_signal if present and valid, else fall back to dominant
+            _preferred = _raw_latest if _raw_latest.upper() in {"BUY", "SELL", "NEUTRAL"} else _raw_dominant
             self.dominant_signal = (
-                raw_signal.upper()
-                if isinstance(raw_signal, str)
-                   and raw_signal.upper() in {"BUY", "SELL", "NEUTRAL"}
+                _preferred.upper()
+                if isinstance(_preferred, str)
+                   and _preferred.upper() in {"BUY", "SELL", "NEUTRAL"}
                 else "NEUTRAL"
             )
             self.buy_count  = int(voting.get("buy_count",  self.buy_count))
             self.sell_count = int(voting.get("sell_count", self.sell_count))
+            # When using latest_signal, confidence = 1.0 (direct model output,
+            # not a vote ratio). Override only if we actually got latest_signal.
+            if _raw_latest.upper() in {"BUY", "SELL"}:
+                self._latest_signal_confidence = 1.0
+            else:
+                self._latest_signal_confidence = None  # use vote ratio
 
         self.last_updated = snapshot.get("last_updated", time.time())
 
@@ -847,6 +861,9 @@ class AssetSnapshot:
 
     @property
     def signal_confidence(self) -> float:
+        # ── CORE FIX 6b: use 1.0 when signal came from latest_signal ──
+        if hasattr(self, "_latest_signal_confidence") and self._latest_signal_confidence is not None:
+            return self._latest_signal_confidence
         if self.total_votes == 0:
             return 0.0
         return max(self.buy_count, self.sell_count) / self.total_votes
@@ -6510,7 +6527,7 @@ class PortfolioRiskManager:
         if self._peak_equity <= 0:
             return 0.0
         raw_dd = (self._peak_equity - self._current_equity) / self._peak_equity
-        return min(1.0, max(0.0, raw_dd))  # FIX 5: clamp to [0,1] — prevents 476%+ when equity goes negative
+        return min(1.0, max(0.0, raw_dd))  # FIX: clamp [0,1]
 
     def _get_max_pos(self, asset_id: str) -> float:
         return self.asset_registry.get(asset_id, {}).get("max_pos", 0.002)
@@ -6796,8 +6813,8 @@ class QuasarAXRVIBridge:
         reward_strategy: str  = "simple",
         hub_ws_url:      str  = os.environ.get("QUASAR_HUB_URL", "ws://localhost:7860/ws/subscribe"),
         enable_logging:  bool = True,
-        checkpoint_dir:  str  = "./Ranker7",   # FIX 6: new fresh folder — no restoring from Ranker6
-        resume:          bool = False,         # FIX 6: fresh start — do NOT resume from old checkpoint
+        checkpoint_dir:  str  = "./Ranker8",   # FIX: fresh folder
+        resume:          bool = False,         # FIX: fresh start
         hf_repo_id:      Optional[str] = "KarlQuant/k1rl-checkpoints",  # HF Dataset repo
     ):
         self.config          = config or AssetRankerConfig()
@@ -9061,7 +9078,7 @@ class RankerCheckpointManager:
 
     def __init__(
         self,
-        checkpoint_dir: str = "./Ranker7",
+        checkpoint_dir: str = "./Ranker8",
         hf_repo_id: Optional[str] = None,
     ):
         self.checkpoint_dir  = checkpoint_dir
@@ -10370,8 +10387,8 @@ async def run_live_trading_system(
     hub_ws_url:      str  = "ws://localhost:7860/ws/subscribe",
     enable_logging:  bool = True,
     shreve_config:   Optional[ShreveConfig] = None,
-    checkpoint_dir:  str  = "./Ranker7",   # FIX 6: new fresh folder
-    resume:          bool = False,   # FIX 6: fresh start — no restore
+    checkpoint_dir:  str  = "./Ranker8",
+    resume:          bool = False,   # FIX: fresh start
     hf_repo_id:      Optional[str] = "KarlQuant/k1rl-checkpoints",  # HF Dataset repo
 ) -> None:
     config = AssetRankerConfig(
@@ -10655,7 +10672,7 @@ def test_components() -> None:
 def _parse_args():
     """Parse CLI args, stripping any Jupyter kernel args."""
     filtered = [a for a in sys.argv[1:] if not a.startswith("-f")]
-    parser   = argparse.ArgumentParser(description="QUASAR AXRVI Ranker v7 — Shreve Framework | Fresh ./Ranker7 folder")
+    parser   = argparse.ArgumentParser(description="QUASAR AXRVI Ranker v7 — Shreve Framework")
     parser.add_argument("--test",   action="store_true",
                         help="Run component tests and exit")
     parser.add_argument("--assets", nargs="+",
@@ -10684,12 +10701,12 @@ def _parse_args():
                         help="[S6/S8] Trade horizon τ in seconds (default 60)")
     parser.add_argument("--martingale-epsilon", type=float, default=0.05,
                         help="[S7] Gate E martingale deviation threshold (default 0.05)")
-    parser.add_argument("--checkpoint-dir", default="./Ranker7",
-                        help="Directory for full-state checkpoints (default ./Ranker7 — fresh folder, no restore)")
+    parser.add_argument("--checkpoint-dir", default="./Ranker8",
+                        help="Directory for full-state checkpoints (default ./Ranker7)")
     parser.add_argument("--no-resume", dest="no_resume", action="store_true", default=True,
-                        help="FIX 7: Default True — always start fresh from ./Ranker7 folder.")
+                        help="Default True — always fresh start.")
     parser.add_argument("--resume", dest="no_resume", action="store_false",
-                        help="Restore from latest Ranker7 checkpoint (overrides --no-resume default)")
+                        help="Restore from latest Ranker7 checkpoint")
     parser.add_argument("--hf-repo", default=None,
                         metavar="OWNER/REPO",
                         help="Hugging Face Dataset repo for checkpoint sync "
@@ -10746,7 +10763,7 @@ if __name__ == "__main__":
         hub_ws_url      = args.hub,
         enable_logging  = not args.no_logs,
         checkpoint_dir  = args.checkpoint_dir,
-        resume          = not args.no_resume,   # FIX 1: default True (always resume)
+        resume          = not args.no_resume,   # default False — always start fresh on Ranker8
         hf_repo_id      = args.hf_repo or "KarlQuant/k1rl-checkpoints",
     )