@@ -968,8 +968,13 @@ void score_general_ce_nonanalog(Particle& p, int i_tally, int start_index,
                   const DelayedGroupFilter& filt {
                     *dynamic_cast<DelayedGroupFilter*>(
                       model::tally_filters[i_dg_filt].get())};
-                  score_fission_delayed_dg(i_tally, delayed_groups[0] - 1,
-                    score, score_index, p.filter_matches());
+                  for (int d_bin = 0; d_bin < filt.n_bins(); ++d_bin) {
+                    if (filt.groups()[d_bin] == delayed_groups[0]) {
+                      score_fission_delayed_dg(
+                        i_tally, d_bin, score, score_index, p.filter_matches());
+                      break;
+                    }
+                  }
                   continue;
                 }
               }
 
@@ -0,0 +1,97 @@
+"""Test that delayed group filter bins follow the groups they name rather than
+their position in the filter's list of bins."""
+
+import pytest
+import openmc
+
+
+GROUPS_ASCENDING = [1, 2, 3, 4, 5, 6]
+GROUPS_SHUFFLED = [4, 1, 6, 3, 5, 2]
+GROUPS_SUBSET = [2, 4]
+
+SCORES = ["delayed-nu-fission", "ifp-beta-numerator"]
+
+
+@pytest.fixture(scope="module")
+def geometry():
+    openmc.reset_auto_ids()
+    material = openmc.Material()
+    material.add_nuclide("U235", 1.0)
+    material.set_density("g/cm3", 16.0)
+    sphere = openmc.Sphere(r=10.0, boundary_type="vacuum")
+    cell = openmc.Cell(region=-sphere, fill=material)
+    return openmc.Geometry([cell])
+
+
+def build_model(geometry, score):
+    """Model tallying one score four ways: three delayed group filters that
+    differ only in how the bins are listed, plus an unfiltered reference."""
+    model = openmc.Model(geometry=geometry)
+    model.settings.particles = 5000
+    model.settings.batches = 20
+    model.settings.inactive = 5
+    model.settings.ifp_n_generation = 5
+
+    for name, groups in [
+        ("ascending", GROUPS_ASCENDING),
+        ("shuffled", GROUPS_SHUFFLED),
+        ("subset", GROUPS_SUBSET),
+    ]:
+        tally = openmc.Tally(name=name)
+        tally.scores = [score]
+        tally.filters = [openmc.DelayedGroupFilter(groups)]
+        model.tallies.append(tally)
+
+    tally = openmc.Tally(name="unfiltered")
+    tally.scores = [score]
+    model.tallies.append(tally)
+
+    return model
+
+
+def means_by_group(statepoint, name):
+    """Map delayed group number to tallied mean.
+
+    The group order is read back from the filter stored in the statepoint, so
+    a mismatch between the reported bin labels and the values fails here
+    instead of being hidden by assuming an order.
+    """
+    tally = statepoint.get_tally(name=name)
+    delayed_group_filter = tally.find_filter(openmc.DelayedGroupFilter)
+    means = tally.mean.ravel()
+    assert len(means) == len(delayed_group_filter.bins)
+    return dict(zip(delayed_group_filter.bins, means))
+
+
+@pytest.mark.parametrize("score", SCORES)
+def test_delayed_group_bins(run_in_tmpdir, geometry, score):
+    """Delayed group labels determine values independently of bin order.
+
+    All tallies come from a single run, so they see the same histories and the
+    same scoring events. Corresponding groups must therefore agree within
+    statistics, and all selected groups must sum to the unfiltered result.
+    """
+    model = build_model(geometry, score)
+    sp_file = model.run()
+    with openmc.StatePoint(sp_file) as sp:
+        ascending = means_by_group(sp, "ascending")
+        shuffled = means_by_group(sp, "shuffled")
+        subset = means_by_group(sp, "subset")
+        unfiltered = sp.get_tally(name="unfiltered").mean.ravel()[0]
+
+    # Guard against a vacuous pass if the run were too short to populate bins.
+    assert all(value > 0.0 for value in ascending.values())
+
+    # Listing groups out of order must not move the corresponding results.
+    assert sorted(shuffled) == sorted(ascending)
+    for group in GROUPS_ASCENDING:
+        assert shuffled[group] == pytest.approx(ascending[group], rel=1e-6)
+
+    # A partial selection must reproduce those groups from the full list. A
+    # group-to-bin index assumption would write out of bounds for group 4.
+    assert sorted(subset) == sorted(GROUPS_SUBSET)
+    for group in GROUPS_SUBSET:
+        assert subset[group] == pytest.approx(ascending[group], rel=1e-6)
+
+    # Every delayed neutron lands in exactly one bin and none are lost.
+    assert sum(ascending.values()) == pytest.approx(unfiltered, rel=1e-6)