Fix 3 engine bugs: swing-point NaN false positives, ROC div-by-zero, ADX epsilon

Ported from feature/eurusd-support (e6e09c5), pair-neutral fixes:
- _find_swing_points: all(<empty generator>) is vacuously True, so a value
  surrounded by NaNs was reported as a swing point; now requires at least
  one valid neighbor on each side
- ROC: guard against zero close at the lookback positions (inf scores)
- ADX: epsilon denominator instead of NaN-replace to avoid NaN cascade

Adds 8 regression tests (Module 9). 150/150 pass.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

engine.py

@@ -492,7 +492,7 @@ class TechnicalEngine:
         plus_di = 100 * (plus_dm.ewm(alpha=1/c.adx_period, min_periods=c.adx_period, adjust=False).mean() / atr)
         minus_di = 100 * (minus_dm.ewm(alpha=1/c.adx_period, min_periods=c.adx_period, adjust=False).mean() / atr)
 
-        dx = 100 * (plus_di - minus_di).abs() / (plus_di + minus_di).replace(0, np.nan)
+        dx = 100 * (plus_di - minus_di).abs() / np.maximum(plus_di + minus_di, 1e-10)
         adx = dx.ewm(alpha=1/c.adx_period, min_periods=c.adx_period, adjust=False).mean()
 
         adx_val = adx.iloc[-1] if not adx.empty and not pd.isna(adx.iloc[-1]) else 0.0
@@ -602,8 +602,12 @@ class TechnicalEngine:
         if len(close) < n + 5:
             return
 
-        roc_val = ((close.iloc[-1] - close.iloc[-n - 1]) / close.iloc[-n - 1]) * 100
+        denom1 = close.iloc[-n - 1]
-        roc_prev = ((close.iloc[-2] - close.iloc[-n - 2]) / close.iloc[-n - 2]) * 100
+        denom2 = close.iloc[-n - 2]
+        if denom1 == 0 or denom2 == 0:
+            return
+        roc_val = ((close.iloc[-1] - denom1) / denom1) * 100
+        roc_prev = ((close.iloc[-2] - denom2) / denom2) * 100
 
         if pd.isna(roc_val) or pd.isna(roc_prev):
             return
@@ -1154,12 +1158,14 @@ class TechnicalEngine:
         for i in range(window, len(vals) - window):
             if np.isnan(vals[i]):
                 continue
+            left_vals = [vals[i-j] for j in range(1, window+1) if not np.isnan(vals[i-j])]
+            right_vals = [vals[i+j] for j in range(1, window+1) if i+j < len(vals) and not np.isnan(vals[i+j])]
+            if not left_vals or not right_vals:
+                continue
             if mode == "low":
-                if all(vals[i] <= vals[i-j] for j in range(1, window+1) if not np.isnan(vals[i-j])) and \
+                if all(vals[i] <= v for v in left_vals) and all(vals[i] <= v for v in right_vals):
-                   all(vals[i] <= vals[i+j] for j in range(1, window+1) if i+j < len(vals) and not np.isnan(vals[i+j])):
                     points.append((i, vals[i]))
             else:
-                if all(vals[i] >= vals[i-j] for j in range(1, window+1) if not np.isnan(vals[i-j])) and \
+                if all(vals[i] >= v for v in left_vals) and all(vals[i] >= v for v in right_vals):
-                   all(vals[i] >= vals[i+j] for j in range(1, window+1) if i+j < len(vals) and not np.isnan(vals[i+j])):
                     points.append((i, vals[i]))
         return points

test_all.py

@@ -1761,6 +1761,85 @@ def _():
     assert r1.market_regime == r2.market_regime
 
 
+print("\n═══ Module 9: Engine Bug Fix Regressions ═══")
+
+# ── 9a. Swing point NaN fix ──────────────────────────────────────────
+
+@test("Swing points: all-NaN neighbors produces no false positives")
+def _():
+    series = pd.Series([np.nan, np.nan, np.nan, 5.0, np.nan, np.nan, np.nan])
+    lows = TechnicalEngine._find_swing_points(series, mode="low", window=3)
+    highs = TechnicalEngine._find_swing_points(series, mode="high", window=3)
+    assert lows == [], f"Expected no lows with all-NaN neighbors, got {lows}"
+    assert highs == [], f"Expected no highs with all-NaN neighbors, got {highs}"
+
+
+@test("Swing points: partial NaN neighbors still detect valid swings")
+def _():
+    series = pd.Series([10.0, 8.0, np.nan, 4.0, np.nan, 8.0, 10.0])
+    lows = TechnicalEngine._find_swing_points(series, mode="low", window=2)
+    low_vals = [v for _, v in lows]
+    assert 4.0 in low_vals, f"Expected 4.0 as swing low, got {low_vals}"
+
+
+@test("Swing points: empty series returns empty list")
+def _():
+    series = pd.Series([], dtype=float)
+    assert TechnicalEngine._find_swing_points(series, mode="low") == []
+    assert TechnicalEngine._find_swing_points(series, mode="high") == []
+
+
+@test("Swing points: monotonic increasing has no interior lows")
+def _():
+    series = pd.Series([float(i) for i in range(20)])
+    lows = TechnicalEngine._find_swing_points(series, mode="low", window=2)
+    assert len(lows) == 0, f"Monotonic increase should have no interior lows, got {lows}"
+
+
+@test("Swing points: V-shape detects low at bottom")
+def _():
+    series = pd.Series([10.0, 8.0, 6.0, 4.0, 6.0, 8.0, 10.0])
+    lows = TechnicalEngine._find_swing_points(series, mode="low", window=2)
+    low_vals = [v for _, v in lows]
+    assert 4.0 in low_vals, f"V-shape should detect 4.0 as low, got {low_vals}"
+
+
+@test("Swing points: inverted V-shape detects high at top")
+def _():
+    series = pd.Series([4.0, 6.0, 8.0, 10.0, 8.0, 6.0, 4.0])
+    highs = TechnicalEngine._find_swing_points(series, mode="high", window=2)
+    high_vals = [v for _, v in highs]
+    assert 10.0 in high_vals, f"Inverted V should detect 10.0 as high, got {high_vals}"
+
+
+# ── 9b. ROC div-by-zero fix ──────────────────────────────────────────
+
+@test("ROC: zero close at lookback position does not produce inf")
+def _():
+    df = make_ohlcv(n=100, seed=42)
+    df.iloc[-11, df.columns.get_loc("Close")] = 0.0
+    engine = TechnicalEngine(PlatformConfig().technical)
+    result = engine.analyze(df, "EUR/USD", Timeframe.H1)
+    if "roc" in result.indicators:
+        assert np.isfinite(result.indicators["roc"].value), "ROC should not be inf"
+
+
+# ── 9c. ADX epsilon fix ──────────────────────────────────────────────
+
+@test("ADX: flat data produces finite ADX value")
+def _():
+    n = 100
+    dates = pd.date_range(end=pd.Timestamp.now(tz="UTC"), periods=n, freq="h")
+    df = pd.DataFrame({
+        "Open": [100.0]*n, "High": [100.01]*n, "Low": [99.99]*n,
+        "Close": [100.0]*n, "Volume": [1000.0]*n,
+    }, index=dates)
+    engine = TechnicalEngine(PlatformConfig().technical)
+    result = engine.analyze(df, "EUR/USD", Timeframe.H1)
+    if "adx" in result.indicators:
+        assert np.isfinite(result.indicators["adx"].value), "ADX should be finite on flat data"
+
+
 # ══════════════════════════════════════════════════════════════════════
 #  SUMMARY
 # ══════════════════════════════════════════════════════════════════════