diff --git a/predicators/approaches/__init__.py b/predicators/approaches/__init__.py
index 502e46736b..ee5ad6fa52 100644
--- a/predicators/approaches/__init__.py
+++ b/predicators/approaches/__init__.py
@@ -17,17 +17,17 @@
 
 
 def _get_approach_cls_from_name(name: str) -> TypingType[BaseApproach]:
-    for cls in utils.get_all_subclasses(BaseApproach):
-        if not cls.__abstractmethods__ and cls.get_name() == name:
-            return cls
-    raise NotImplementedError(f"Unknown approach: {name}")
+    cls = utils.get_registered_subclass(BaseApproach, name)
+    if cls is None:
+        raise NotImplementedError(f"Unknown approach: {name}")
+    return cls
 
 
 def _get_wrapper_cls_from_name(name: str) -> TypingType[BaseApproachWrapper]:
-    for cls in utils.get_all_subclasses(BaseApproachWrapper):
-        if not cls.__abstractmethods__ and cls.get_name() == name:
-            return cls
-    raise NotImplementedError(f"Unknown wrapper approach: {name}")
+    cls = utils.get_registered_subclass(BaseApproachWrapper, name)
+    if cls is None:
+        raise NotImplementedError(f"Unknown wrapper approach: {name}")
+    return cls
 
 
 def create_approach(name: str, initial_predicates: Set[Predicate],
diff --git a/predicators/approaches/human_low_level_control_approach.py b/predicators/approaches/human_low_level_control_approach.py
index 54ed56a8f2..3d9fb52280 100644
--- a/predicators/approaches/human_low_level_control_approach.py
+++ b/predicators/approaches/human_low_level_control_approach.py
@@ -354,21 +354,12 @@ def _get_shadow_robot_for_env() -> SingleArmPyBulletRobot:
     ground plane, and ask the env class for a robot.
     """
     # pylint: disable=import-outside-toplevel
-    from predicators.envs.base_env import BaseEnv
     from predicators.envs.pybullet_env import PyBulletEnv
 
     # pylint: enable=import-outside-toplevel
 
     env_name = CFG.env
-    env_cls = None
-    for cls in utils.get_all_subclasses(BaseEnv):
-        if cls.__abstractmethods__:
-            continue
-        if not issubclass(cls, PyBulletEnv):
-            continue
-        if cls.get_name() == env_name:
-            env_cls = cls
-            break
+    env_cls = utils.get_registered_subclass(PyBulletEnv, env_name)
     if env_cls is None:
         raise NotImplementedError(
             f"human_low_level_control: no PyBulletEnv subclass registered "
diff --git a/predicators/bridge_policies/__init__.py b/predicators/bridge_policies/__init__.py
index d73ab03441..de3a3fdf9b 100644
--- a/predicators/bridge_policies/__init__.py
+++ b/predicators/bridge_policies/__init__.py
@@ -18,10 +18,7 @@ def create_bridge_policy(name: str, types: Set[Type],
                          options: Set[ParameterizedOption],
                          nsrts: Set[NSRT]) -> BaseBridgePolicy:
     """Create a bridge policy given its name."""
-    for cls in utils.get_all_subclasses(BaseBridgePolicy):
-        if not cls.__abstractmethods__ and cls.get_name() == name:
-            bridge_policy = cls(types, predicates, options, nsrts)
-            break
-    else:
+    cls = utils.get_registered_subclass(BaseBridgePolicy, name)
+    if cls is None:
         raise NotImplementedError(f"Unknown bridge policy: {name}")
-    return bridge_policy
+    return cls(types, predicates, options, nsrts)
diff --git a/predicators/competence_models.py b/predicators/competence_models.py
index 6bd99959fa..1a27f6d2dd 100644
--- a/predicators/competence_models.py
+++ b/predicators/competence_models.py
@@ -240,10 +240,10 @@ def _run_map_inference(self, betas: List[BetaRV]) -> List[float]:
 
 def _get_competence_model_cls_from_name(
         name: str) -> TypingType[SkillCompetenceModel]:
-    for cls in utils.get_all_subclasses(SkillCompetenceModel):
-        if not cls.__abstractmethods__ and cls.get_name() == name:
-            return cls
-    raise NotImplementedError(f"Unknown competence model: {name}")
+    cls = utils.get_registered_subclass(SkillCompetenceModel, name)
+    if cls is None:
+        raise NotImplementedError(f"Unknown competence model: {name}")
+    return cls
 
 
 def create_competence_model(model_name: str,
diff --git a/predicators/envs/__init__.py b/predicators/envs/__init__.py
index 2510edd608..22a414ca58 100644
--- a/predicators/envs/__init__.py
+++ b/predicators/envs/__init__.py
@@ -22,12 +22,10 @@ def create_new_env(name: str,
     If do_cache is True, then cache this env instance so that it can
     later be loaded using get_or_create_env().
     """
-    for cls in utils.get_all_subclasses(BaseEnv):
-        if not cls.__abstractmethods__ and cls.get_name() == name:
-            env = cls(use_gui, **kwargs)
-            break
-    else:
+    cls = utils.get_registered_subclass(BaseEnv, name)
+    if cls is None:
         raise NotImplementedError(f"Unknown env: {name}")
+    env = cls(use_gui, **kwargs)
     if do_cache:
         _MOST_RECENT_ENV_INSTANCE[name] = env
     return env
diff --git a/predicators/envs/pybullet_domino_real.py b/predicators/envs/pybullet_domino_real.py
index cb4237e7eb..fb50213c39 100644
--- a/predicators/envs/pybullet_domino_real.py
+++ b/predicators/envs/pybullet_domino_real.py
@@ -123,11 +123,9 @@ class RealSceneGeometryMixin:
     perceived scene, while :class:`PyBulletDominoRealGeometryEnv` pairs it
     with the ordinary generated tasks.
 
-    Deliberately NOT a ``BaseEnv`` subclass. ``create_new_env`` resolves an
-    env by scanning ``get_all_subclasses(BaseEnv)`` for a matching
-    ``get_name()``, so an intermediate env class here would inherit
-    ``PyBulletDominoEnv.get_name()`` and shadow the real
-    ``pybullet_domino``. A plain mixin never enters that scan.
+    Deliberately NOT a ``BaseEnv`` subclass: it adds geometry to whichever
+    domino env it is mixed into and has no env name or tasks of its own,
+    so it stays out of the env registry that ``create_new_env`` reads.
     """
 
     # Annotations only, no values: these ClassVars belong to the domino env
diff --git a/predicators/execution_monitoring/__init__.py b/predicators/execution_monitoring/__init__.py
index d73bf94ca3..004ce78097 100644
--- a/predicators/execution_monitoring/__init__.py
+++ b/predicators/execution_monitoring/__init__.py
@@ -12,10 +12,7 @@
 
 def create_execution_monitor(name: str, ) -> BaseExecutionMonitor:
     """Create an execution monitor given its name."""
-    for cls in utils.get_all_subclasses(BaseExecutionMonitor):
-        if not cls.__abstractmethods__ and cls.get_name() == name:
-            execution_monitor = cls()
-            break
-    else:
+    cls = utils.get_registered_subclass(BaseExecutionMonitor, name)
+    if cls is None:
         raise NotImplementedError(f"Unrecognized execution monitor: {name}")
-    return execution_monitor
+    return cls()
diff --git a/predicators/explorers/__init__.py b/predicators/explorers/__init__.py
index 3a930dbbe7..924d53a638 100644
--- a/predicators/explorers/__init__.py
+++ b/predicators/explorers/__init__.py
@@ -47,67 +47,59 @@ def create_explorer(
     """Create an explorer given its name."""
     if max_steps_before_termination is None:
         max_steps_before_termination = CFG.max_num_steps_interaction_request
-    for cls in utils.get_all_subclasses(BaseExplorer):
-        if not cls.__abstractmethods__ and cls.get_name() == name:
-            # Special case GLIB because it uses babble predicates and an atom
-            # score function.
-            if name == "glib":
-                assert nsrts is not None
-                assert option_model is not None
-                assert babble_predicates is not None
-                assert atom_score_fn is not None
-                explorer = cls(initial_predicates, initial_options, types,
-                               action_space, train_tasks,
-                               max_steps_before_termination, nsrts,
-                               option_model, babble_predicates, atom_score_fn)
-            # Special case greedy lookahead because it uses a state score
-            # function.
-            elif name == "greedy_lookahead":
-                assert nsrts is not None
-                assert option_model is not None
-                assert state_score_fn is not None
-                explorer = cls(initial_predicates, initial_options, types,
-                               action_space, train_tasks,
-                               max_steps_before_termination, nsrts,
-                               option_model, state_score_fn)
-            # Bilevel planning approaches use NSRTs and an option model.
-            elif issubclass(cls, BilevelPlanningExplorer):
-                assert nsrts is not None
-                assert option_model is not None
-                explorer = cls(initial_predicates, initial_options, types,
-                               action_space, train_tasks,
-                               max_steps_before_termination, nsrts,
-                               option_model)
-            # Random NSRTs explorer uses NSRTs, but not an option model
-            elif name == "random_nsrts":
-                assert nsrts is not None
-                explorer = cls(initial_predicates, initial_options, types,
-                               action_space, train_tasks,
-                               max_steps_before_termination, nsrts)
-            # Active sampler explorer uses ground_op_hist and no option model.
-            elif name == "active_sampler":
-                assert ground_op_hist is not None
-                assert competence_models is not None
-                assert nsrt_to_explorer_sampler is not None
-                assert seen_train_task_idxs is not None
-                assert pursue_task_goal_first is not None
-                explorer = cls(initial_predicates, initial_options, types,
-                               action_space, train_tasks,
-                               max_steps_before_termination, nsrts,
-                               ground_op_hist, competence_models,
-                               nsrt_to_explorer_sampler, seen_train_task_idxs,
-                               pursue_task_goal_first)
-            elif name == "maple_q":
-                assert maple_q_function is not None
-                explorer = cls(initial_predicates, initial_options, types,
-                               action_space, train_tasks,
-                               max_steps_before_termination, nsrts,
-                               maple_q_function)
-            else:
-                explorer = cls(initial_predicates, initial_options, types,
-                               action_space, train_tasks,
-                               max_steps_before_termination)
-            break
-    else:
+    cls = utils.get_registered_subclass(BaseExplorer, name)
+    if cls is None:
         raise NotImplementedError(f"Unrecognized explorer: {name}")
+    # Special case GLIB because it uses babble predicates and an atom
+    # score function.
+    if name == "glib":
+        assert nsrts is not None
+        assert option_model is not None
+        assert babble_predicates is not None
+        assert atom_score_fn is not None
+        explorer = cls(initial_predicates, initial_options, types,
+                       action_space, train_tasks, max_steps_before_termination,
+                       nsrts, option_model, babble_predicates, atom_score_fn)
+    # Special case greedy lookahead because it uses a state score
+    # function.
+    elif name == "greedy_lookahead":
+        assert nsrts is not None
+        assert option_model is not None
+        assert state_score_fn is not None
+        explorer = cls(initial_predicates, initial_options, types,
+                       action_space, train_tasks, max_steps_before_termination,
+                       nsrts, option_model, state_score_fn)
+    # Bilevel planning approaches use NSRTs and an option model.
+    elif issubclass(cls, BilevelPlanningExplorer):
+        assert nsrts is not None
+        assert option_model is not None
+        explorer = cls(initial_predicates, initial_options, types,
+                       action_space, train_tasks, max_steps_before_termination,
+                       nsrts, option_model)
+    # Random NSRTs explorer uses NSRTs, but not an option model
+    elif name == "random_nsrts":
+        assert nsrts is not None
+        explorer = cls(initial_predicates, initial_options, types,
+                       action_space, train_tasks, max_steps_before_termination,
+                       nsrts)
+    # Active sampler explorer uses ground_op_hist and no option model.
+    elif name == "active_sampler":
+        assert ground_op_hist is not None
+        assert competence_models is not None
+        assert nsrt_to_explorer_sampler is not None
+        assert seen_train_task_idxs is not None
+        assert pursue_task_goal_first is not None
+        explorer = cls(initial_predicates, initial_options, types,
+                       action_space, train_tasks, max_steps_before_termination,
+                       nsrts, ground_op_hist, competence_models,
+                       nsrt_to_explorer_sampler, seen_train_task_idxs,
+                       pursue_task_goal_first)
+    elif name == "maple_q":
+        assert maple_q_function is not None
+        explorer = cls(initial_predicates, initial_options, types,
+                       action_space, train_tasks, max_steps_before_termination,
+                       nsrts, maple_q_function)
+    else:
+        explorer = cls(initial_predicates, initial_options, types,
+                       action_space, train_tasks, max_steps_before_termination)
     return explorer
diff --git a/predicators/ground_truth_models/domino/options.py b/predicators/ground_truth_models/domino/options.py
index fa2308ea74..b0f65fa6c7 100644
--- a/predicators/ground_truth_models/domino/options.py
+++ b/predicators/ground_truth_models/domino/options.py
@@ -29,11 +29,9 @@ def _skill_robot_env_cls(env_name: str) -> TypingType[PyBulletEnv]:
     """
     # pylint: disable=import-outside-toplevel  # local: avoid import cycle
     from predicators.envs.base_env import BaseEnv
-    from predicators.utils import get_all_subclasses
-    for c in get_all_subclasses(BaseEnv):
-        if not c.__abstractmethods__ and c.get_name() == env_name:
-            return c  # type: ignore[return-value]
-    return PyBulletDominoEnv
+    from predicators.utils import get_registered_subclass
+    cls = get_registered_subclass(BaseEnv, env_name)
+    return PyBulletDominoEnv if cls is None else cls
 
 
 class PyBulletDominoGroundTruthOptionFactory(_DominoLegacyOptionsMixin,
diff --git a/predicators/nsrt_learning/strips_learning/__init__.py b/predicators/nsrt_learning/strips_learning/__init__.py
index 84015e2f7e..84bfa771b2 100644
--- a/predicators/nsrt_learning/strips_learning/__init__.py
+++ b/predicators/nsrt_learning/strips_learning/__init__.py
@@ -28,13 +28,9 @@ def learn_strips_operators(trajectories: List[LowLevelTrajectory],
     Return a list of PNADs with op (STRIPSOperator), datastore, and
     option_spec fields filled in (but not sampler).
     """
-    for cls in utils.get_all_subclasses(BaseSTRIPSLearner):
-        if not cls.__abstractmethods__ and \
-           cls.get_name() == CFG.strips_learner:
-            learner = cls(trajectories, train_tasks, predicates,
-                          segmented_trajs, verify_harmlessness, annotations,
-                          verbose, **kwargs)
-            break
-    else:
+    cls = utils.get_registered_subclass(BaseSTRIPSLearner, CFG.strips_learner)
+    if cls is None:
         raise ValueError(f"Unrecognized STRIPS learner: {CFG.strips_learner}")
+    learner = cls(trajectories, train_tasks, predicates, segmented_trajs,
+                  verify_harmlessness, annotations, verbose, **kwargs)
     return learner.learn()
diff --git a/predicators/perception/__init__.py b/predicators/perception/__init__.py
index 7984f70738..edbf3d7af9 100644
--- a/predicators/perception/__init__.py
+++ b/predicators/perception/__init__.py
@@ -11,10 +11,7 @@
 
 def create_perceiver(name: str, ) -> BasePerceiver:
     """Create a perceiver given its name."""
-    for cls in utils.get_all_subclasses(BasePerceiver):
-        if not cls.__abstractmethods__ and cls.get_name() == name:
-            perceiver = cls()
-            break
-    else:
+    cls = utils.get_registered_subclass(BasePerceiver, name)
+    if cls is None:
         raise NotImplementedError(f"Unrecognized perceiver: {name}")
-    return perceiver
+    return cls()
diff --git a/predicators/refinement_estimators/__init__.py b/predicators/refinement_estimators/__init__.py
index 501a1d7637..b9f9c9fee9 100644
--- a/predicators/refinement_estimators/__init__.py
+++ b/predicators/refinement_estimators/__init__.py
@@ -12,10 +12,7 @@
 
 def create_refinement_estimator(name: str) -> BaseRefinementEstimator:
     """Create an approach given its name."""
-    for cls in utils.get_all_subclasses(BaseRefinementEstimator):
-        if not cls.__abstractmethods__ and cls.get_name() == name:
-            estimator = cls()
-            break
-    else:
+    cls = utils.get_registered_subclass(BaseRefinementEstimator, name)
+    if cls is None:
         raise NotImplementedError(f"Unknown refinement cost estimator: {name}")
-    return estimator
+    return cls()
diff --git a/predicators/utils.py b/predicators/utils.py
index c5a871c504..0de16194ab 100644
--- a/predicators/utils.py
+++ b/predicators/utils.py
@@ -5162,6 +5162,34 @@ def get_all_subclasses(cls: Any) -> Set[Any]:
         [s for c in cls.__subclasses__() for s in get_all_subclasses(c)])
 
 
+def get_registered_subclass(base: Any, name: str) -> Optional[Any]:
+    """The concrete subclass of ``base`` registered under ``name``.
+
+    The name belongs to the most general concrete subclass whose
+    ``get_name()`` returns it. A subclass that inherits the name, such
+    as a test's instrumented env, never resolves in its place: picking
+    the first match of ``get_all_subclasses`` would follow set order,
+    which differs from process to process. Returns None when no concrete
+    subclass has the name, and raises ValueError when unrelated
+    subclasses claim it.
+    """
+    matches = [
+        cls for cls in get_all_subclasses(base)
+        if not cls.__abstractmethods__ and cls.get_name() == name
+    ]
+    roots = [
+        cls for cls in matches
+        if not any(other is not cls and issubclass(cls, other)
+                   for other in matches)
+    ]
+    if len(roots) > 1:
+        claimants = sorted(f"{cls.__module__}.{cls.__qualname__}"
+                           for cls in roots)
+        raise ValueError(f"Unrelated classes claim the name {name!r}: "
+                         f"{', '.join(claimants)}")
+    return roots[0] if roots else None
+
+
 class _DummyFile(io.StringIO):
     """Dummy file object used by nostdout()."""
 
diff --git a/tests/envs/test_base_env.py b/tests/envs/test_base_env.py
index f1614ff5c0..075168e96c 100644
--- a/tests/envs/test_base_env.py
+++ b/tests/envs/test_base_env.py
@@ -9,6 +9,7 @@
 import predicators.envs
 from predicators import utils
 from predicators.envs import BaseEnv, create_new_env, get_or_create_env
+from predicators.envs.cover import CoverEnv
 from tests.approaches.test_oracle_approach import ENV_NAME_AND_CLS
 
 _MODULE_PATH = predicators.envs.__name__
@@ -42,6 +43,25 @@ def test_env_creation():
         create_new_env("Not a real env")
 
 
+def test_env_creation_ignores_subclasses_that_inherit_the_name():
+    """A test's instrumented env never stands in for the env it extends."""
+
+    class _InstrumentedCoverEnv(CoverEnv):
+        """Inherits the name ``cover``."""
+
+    utils.reset_config({
+        "env": "cover",
+        "num_train_tasks": 1,
+        "num_test_tasks": 1
+    })
+    # The instrumented class first, as set order sometimes has it.
+    with patch.object(utils,
+                      "get_all_subclasses",
+                      return_value=[_InstrumentedCoverEnv, CoverEnv]):
+        env = create_new_env("cover", do_cache=False)
+    assert env.__class__ is CoverEnv
+
+
 @pytest.mark.parametrize("env_name", ("cover", "sandwich"))
 def test_load_task_from_json(env_name):
     """Tests for env._load_task_from_json()."""
diff --git a/tests/test_utils.py b/tests/test_utils.py
index 4cacd0c86d..e15784591d 100644
--- a/tests/test_utils.py
+++ b/tests/test_utils.py
@@ -3810,3 +3810,61 @@ def _always_fails(obj, *args, **kwargs):
     assert "_abc_data" in str(excinfo.value), \
         "a genuinely unpicklable object must still report why"
     assert path.stat().st_size == 0, "a failed dump left a partial file"
+
+
+def test_get_registered_subclass(monkeypatch):
+    """A name resolves to the most general concrete class that claims it."""
+
+    class _Base(abc.ABC):
+
+        @classmethod
+        @abc.abstractmethod
+        def get_name(cls) -> str:
+            """The registry name."""
+
+    class _Named(_Base):
+
+        @classmethod
+        def get_name(cls) -> str:
+            return "named"
+
+    class _Instrumented(_Named):
+        """Inherits the name, as a test's instrumented env does."""
+
+    class _AbstractNamed(_Base):
+
+        @classmethod
+        def get_name(cls) -> str:
+            return "abstract_named"
+
+        @abc.abstractmethod
+        def run(self) -> None:
+            """Left to subclasses."""
+
+    class _Concrete(_AbstractNamed):
+
+        def run(self) -> None:
+            """Nothing to run."""
+
+    class _First(_Base):
+
+        @classmethod
+        def get_name(cls) -> str:
+            return "claimed_twice"
+
+    class _Second(_Base):
+
+        @classmethod
+        def get_name(cls) -> str:
+            return "claimed_twice"
+
+    assert utils.get_registered_subclass(_Base, "named") is _Named
+    assert utils.get_registered_subclass(_Base, "abstract_named") is _Concrete
+    assert utils.get_registered_subclass(_Base, "unknown") is None
+    with pytest.raises(ValueError, match="Unrelated classes claim"):
+        utils.get_registered_subclass(_Base, "claimed_twice")
+    # Set order differs between processes; the subclass coming first must
+    # not change the answer.
+    monkeypatch.setattr(utils, "get_all_subclasses",
+                        lambda _: [_Instrumented, _Named])
+    assert utils.get_registered_subclass(_Base, "named") is _Named