diff --git a/app/strategies/quant_algos/mean_reversion_algos.py b/app/strategies/quant_algos/mean_reversion_algos.py
index 062608a..7e870ec 100644
--- a/app/strategies/quant_algos/mean_reversion_algos.py
+++ b/app/strategies/quant_algos/mean_reversion_algos.py
@@ -88,9 +88,13 @@ def stochastic_signal(
     if high == low:
         return 0.0, "flat range"
     k = 100 * (close - low) / (high - low)
-    # Simplified %D = SMA of %K
+    # Simplified %D = SMA of only the trailing %K values it consumes.
     k_vals = []
-    for i in range(len(candles) - k_period, len(candles)):
+    if d_period <= k_period:
+        first_d_index = len(candles) - d_period
+    else:
+        first_d_index = len(candles)
+    for i in range(first_d_index, len(candles)):
         h = max(c["high"] for c in candles[i - k_period + 1 : i + 1])
         l_ = min(c["low"] for c in candles[i - k_period + 1 : i + 1])
         c_ = candles[i]["close"]
diff --git a/shared/tests/test_mean_reversion_algos.py b/shared/tests/test_mean_reversion_algos.py
new file mode 100644
index 0000000..ccc37c5
--- /dev/null
+++ b/shared/tests/test_mean_reversion_algos.py
@@ -0,0 +1,65 @@
+from importlib.util import module_from_spec, spec_from_file_location
+from pathlib import Path
+
+
+def load_module():
+    root = Path(__file__).resolve().parents[2]
+    module_path = root / "app" / "strategies" / "quant_algos" / "mean_reversion_algos.py"
+    spec = spec_from_file_location("mean_reversion_algos", module_path)
+    module = module_from_spec(spec)
+    spec.loader.exec_module(module)
+    return module
+
+
+def reference_stochastic_signal(candles, k_period=14, d_period=3):
+    if not candles or len(candles) < k_period:
+        return 0.0, f"warming up ({len(candles)}/{k_period})"
+    high = max(c["high"] for c in candles[-k_period:])
+    low = min(c["low"] for c in candles[-k_period:])
+    close = candles[-1]["close"]
+    if high == low:
+        return 0.0, "flat range"
+    k = 100 * (close - low) / (high - low)
+    k_vals = []
+    for i in range(len(candles) - k_period, len(candles)):
+        h = max(c["high"] for c in candles[i - k_period + 1 : i + 1])
+        l_ = min(c["low"] for c in candles[i - k_period + 1 : i + 1])
+        c_ = candles[i]["close"]
+        k_vals.append(100 * (c_ - l_) / (h - l_) if h != l_ else 50)
+    d = sum(k_vals[-d_period:]) / d_period if len(k_vals) >= d_period else k
+    if k <= 20:
+        sig = 0.5
+    elif k >= 80:
+        sig = -0.5
+    else:
+        sig = (50 - k) / 50
+    return max(-1, min(1, sig)), f"%K={k:.1f} %D={d:.1f}"
+
+
+def make_candles(count=120):
+    return [
+        {
+            "high": 100 + (i % 17) + i * 0.01,
+            "low": 95 + (i % 11) + i * 0.01,
+            "close": 97 + (i % 13) + i * 0.01,
+        }
+        for i in range(count)
+    ]
+
+
+def test_stochastic_signal_matches_previous_d_value():
+    module = load_module()
+    candles = make_candles()
+
+    assert module.stochastic_signal(candles, k_period=14, d_period=3) == reference_stochastic_signal(
+        candles, k_period=14, d_period=3
+    )
+
+
+def test_stochastic_signal_keeps_large_d_period_fallback():
+    module = load_module()
+    candles = make_candles()
+
+    assert module.stochastic_signal(candles, k_period=14, d_period=20) == reference_stochastic_signal(
+        candles, k_period=14, d_period=20
+    )