route xauusd atr breakout to modal

templates/xauusd-atr-breakout/README.md

@@ -8,6 +8,6 @@ The default gate is intentionally responsive so PPE produces more events than a
 
 - `breakout_window`: `12`
 - `atr_mult`: `0.05`
-- no runtime package dependencies
+- assigned runtime: Modal, because this starter uses `pandas`
 
 If it overtrades, raise `atr_mult` first. If it is still too quiet, shorten `breakout_window`.

templates/xauusd-atr-breakout/quant.config.json

@@ -2,7 +2,7 @@
   "pair": "XAUUSD",
   "timeframe": "15m",
   "model_family": "custom Python",
-  "runtime_target": "edge",
+  "runtime_target": "modal",
   "artifact_format": "python_bundle",
   "parameters": {
     "lookback": 64,
@@ -10,8 +10,14 @@
     "breakout_window": 12,
     "atr_mult": 0.05
   },
-  "training_requirements": [],
+  "training_requirements": [
-  "inference_requirements": [],
+    "numpy",
+    "pandas"
+  ],
+  "inference_requirements": [
+    "numpy",
+    "pandas"
+  ],
   "symbol": "XAUUSD",
   "description": "XAUUSD ATR breakout rule baseline",
   "disclaimer": "Educational template only. Not financial advice."

templates/xauusd-atr-breakout/strategy.py

@@ -1,34 +1,18 @@
-def _clean_candles(candles):
+import numpy as np
-    cleaned = []
+import pandas as pd
-    for candle in candles:
-        try:
-            cleaned.append({
-                "open": float(candle[0]),
-                "high": float(candle[1]),
-                "low": float(candle[2]),
-                "close": float(candle[3]),
-                "volume": float(candle[4]) if len(candle) > 4 else 1.0,
-            })
-        except Exception:
-            continue
-    return cleaned
 
 
-def _atr(candles, n):
+def _df(candles):
-    if len(candles) < 2:
+    df = pd.DataFrame(candles, columns=["open", "high", "low", "close", "volume"])
-        return 0.0
+    for col in df.columns:
-    alpha = 2.0 / (n + 1.0)
+        df[col] = pd.to_numeric(df[col], errors="coerce")
-    value = None
+    return df.dropna().reset_index(drop=True)
-    prev_close = candles[0]["close"]
+
-    for candle in candles[1:]:
+
-        tr = max(
+def _atr(df, n):
-            candle["high"] - candle["low"],
+    prev = df["close"].shift()
-            abs(candle["high"] - prev_close),
+    tr = pd.concat([(df["high"] - df["low"]), (df["high"] - prev).abs(), (df["low"] - prev).abs()], axis=1).max(axis=1)
-            abs(candle["low"] - prev_close),
+    return tr.ewm(span=n, adjust=False).mean()
-        )
-        value = tr if value is None else value + alpha * (tr - value)
-        prev_close = candle["close"]
-    return float(value or 0.0)
 
 
 def train(data, config):
@@ -41,7 +25,7 @@ def train(data, config):
             "atr_mult": float(params.get("atr_mult", 0.05)),
         },
         "name": "xauusd_atr_breakout",
-    }, {"training_bars": int(len(data)), "model": "edge_rule_baseline"}
+    }, {"training_bars": int(len(data)), "model": "modal_python_rule_baseline"}
 
 
 def predict(model, market_data, config):
@@ -50,44 +34,23 @@ def predict(model, market_data, config):
     atr_window = int(params.get("atr_window", 14))
     breakout_window = int(params.get("breakout_window", 12))
     atr_mult = float(params.get("atr_mult", 0.05))
-    candles = _clean_candles(market_data.get("candles", []))
+    candles = market_data.get("candles", [])
-
     if len(candles) < lookback:
-        return {
+        return {"signal": "HOLD", "confidence": 0.0, "metadata": {"reason": "not_enough_candles"}}
-            "signal": "HOLD",
-            "confidence": 0.0,
-            "metadata": {"reason": "not_enough_candles", "got": len(candles), "need": lookback},
-        }
 
-    window = candles[-lookback:]
+    df = _df(candles[-lookback:])
-    previous = window[-breakout_window - 1:-1]
+    atr = float(_atr(df, atr_window).iloc[-1])
-    if len(previous) < breakout_window:
+    close = float(df["close"].iloc[-1])
-        return {"signal": "HOLD", "confidence": 0.0, "metadata": {"reason": "not_enough_breakout_window"}}
+    high = float(df["high"].iloc[-breakout_window:-1].max())
-
+    low = float(df["low"].iloc[-breakout_window:-1].min())
-    atr = _atr(window, atr_window)
-    close = window[-1]["close"]
-    high = max(candle["high"] for candle in previous)
-    low = min(candle["low"] for candle in previous)
     upper = high + atr * atr_mult
     lower = low - atr * atr_mult
 
     if close > upper:
         edge = (close - upper) / max(atr, 1e-9)
-        return {
+        return {"signal": "UP", "confidence": round(min(0.55 + edge, 0.92), 4), "metadata": {"breakout": "high", "atr": atr, "upper": upper}}
-            "signal": "UP",
-            "confidence": round(min(0.55 + edge, 0.92), 4),
-            "metadata": {"breakout": "high", "atr": round(atr, 6), "upper": round(upper, 6), "close": round(close, 6)},
-        }
     if close < lower:
         edge = (lower - close) / max(atr, 1e-9)
-        return {
+        return {"signal": "DOWN", "confidence": round(min(0.55 + edge, 0.92), 4), "metadata": {"breakout": "low", "atr": atr, "lower": lower}}
-            "signal": "DOWN",
-            "confidence": round(min(0.55 + edge, 0.92), 4),
-            "metadata": {"breakout": "low", "atr": round(atr, 6), "lower": round(lower, 6), "close": round(close, 6)},
-        }
 
-    return {
+    return {"signal": "HOLD", "confidence": 0.2, "metadata": {"high": high, "low": low, "atr": atr, "upper": upper, "lower": lower}}
-        "signal": "HOLD",
-        "confidence": 0.2,
-        "metadata": {"high": round(high, 6), "low": round(low, 6), "atr": round(atr, 6), "upper": round(upper, 6), "lower": round(lower, 6)},
-    }