diff --git a/exotic/api/elca.py b/exotic/api/elca.py
index f0fdb421..fd46d680 100644
--- a/exotic/api/elca.py
+++ b/exotic/api/elca.py
@@ -374,9 +374,17 @@ def transit_duration(values):
     if not np.isfinite(chord_sq) or chord_sq <= 0 or not np.isfinite(impact_scale) or impact_scale <= 0:
         return np.nan
 
-    argument = np.sqrt(chord_sq) / (impact_scale * sin_inc)
+    # The arcsin argument is normalized by a*sin(i) (NOT by the eccentricity-
+    # scaled planet-star separation), and the eccentric orbital-speed factor
+    # sqrt(1-e^2)/(1+e*sin(omega)) multiplies the whole duration — the same
+    # correction issue #1383 / PR #1384 applied to the three copies of this
+    # formula in exotic.py and transit QC; this fourth copy was missed. The
+    # broken form biased the ns duration prior, the QC duration score, and the
+    # exposure-smearing window for every eccentric target.
+    argument = np.sqrt(chord_sq) / (ars * sin_inc)
     argument = float(np.clip(argument, -1.0, 1.0))
-    duration = (period / np.pi) * np.arcsin(argument)
+    eccentric_speed_factor = np.sqrt(1.0 - ecc ** 2) / max(np.finfo(float).eps, 1.0 + ecc * np.sin(omega))
+    duration = (period / np.pi) * np.arcsin(argument) * eccentric_speed_factor
     return float(duration) if np.isfinite(duration) and duration > 0 else np.nan
 
 
diff --git a/tests/test_ephemeris_validation.py b/tests/test_ephemeris_validation.py
index 9154efe8..97b08308 100644
--- a/tests/test_ephemeris_validation.py
+++ b/tests/test_ephemeris_validation.py
@@ -90,3 +90,24 @@ def archive_lookup():
             {'pName': 'Example b', 'pPer': None, 'midT': 2460000.25},
             archive_lookup=archive_lookup,
         )
+
+
+def test_elca_transit_duration_matches_model_for_eccentric_orbits():
+    """Regression for the fourth site of issue #1383: elca.transit_duration
+    retained the inverted-eccentricity formula after PR #1384 fixed the three
+    copies in exotic.py. The analytic duration must match the transit model's
+    own first-to-fourth-contact duration for eccentric geometries (it was off
+    by 1.95x at e=0.3/omega=90 and 3.48x at e=0.5/omega=90)."""
+    import numpy as np
+    from exotic.api.elca import transit, transit_duration
+
+    for ecc, omega in [(0.0, 90.0), (0.3, 90.0), (0.3, 270.0), (0.5, 90.0)]:
+        values = dict(u0=0.5, u1=0.1, u2=0.3, u3=-0.1, rprs=0.1, per=3.0,
+                      ars=8.0, tmid=1.0, ecc=ecc, omega=omega, inc=88.0)
+        times = 1.0 + np.linspace(-0.35, 0.35, 300001)
+        flux = transit(times, values)
+        in_transit = np.flatnonzero(flux < 1 - 1e-9)
+        measured = times[in_transit[-1]] - times[in_transit[0]]
+        analytic = transit_duration(values)
+        assert analytic == pytest.approx(measured, rel=0.01), (
+            f"e={ecc} omega={omega}: analytic {analytic*24:.4f} h vs model {measured*24:.4f} h")