Fix multi-condition flags with static cohorts returning wrong variants

When a feature flag has multiple conditions and one contains a static
cohort, the SDK now correctly falls back to the API instead of
evaluating subsequent conditions locally and returning incorrect variants.

Introduce RequiresServerEvaluation exception to distinguish between:
- Missing server-side data (static cohorts) → immediate API fallback
- Evaluation errors (bad regex, missing properties) → try next condition

Changes:
- Add RequiresServerEvaluation exception class
- Update match_cohort() to throw RequiresServerEvaluation for static cohorts
- Update match_property_group() to propagate RequiresServerEvaluation
- Update match_feature_flag_properties() to handle both exception types
- Update client.py to catch both exceptions for API fallback
- Export RequiresServerEvaluation in __init__.py
- Add test for multi-condition static cohort scenario

All 84 feature flag tests pass.

Author: haacked

posthog/__init__.py

@@ -11,6 +11,7 @@
     set_context_session as inner_set_context_session,
     identify_context as inner_identify_context,
 )
+from posthog.feature_flags import InconclusiveMatchError, RequiresServerEvaluation
 from posthog.types import FeatureFlag, FlagsAndPayloads, FeatureFlagResult
 from posthog.version import VERSION
 

posthog/client.py

@@ -20,7 +20,11 @@
     exception_is_already_captured,
     mark_exception_as_captured,
 )
-from posthog.feature_flags import InconclusiveMatchError, match_feature_flag_properties
+from posthog.feature_flags import (
+    InconclusiveMatchError,
+    RequiresServerEvaluation,
+    match_feature_flag_properties,
+)
 from posthog.poller import Poller
 from posthog.request import (
     DEFAULT_HOST,
@@ -1583,7 +1587,7 @@ def _locally_evaluate_flag(
                     self.log.debug(
                         f"Successfully computed flag locally: {key} -> {response}"
                     )
-                except InconclusiveMatchError as e:
+                except (RequiresServerEvaluation, InconclusiveMatchError) as e:
                     self.log.debug(f"Failed to compute flag {key} locally: {e}")
                 except Exception as e:
                     self.log.exception(

posthog/feature_flags.py

@@ -22,6 +22,18 @@ class InconclusiveMatchError(Exception):
     pass
 
 
+class RequiresServerEvaluation(Exception):
+    """
+    Raised when feature flag evaluation requires server-side data that is not
+    available locally (e.g., static cohorts, experience continuity).
+
+    This error should propagate immediately to trigger API fallback, unlike
+    InconclusiveMatchError which allows trying other conditions.
+    """
+
+    pass
+
+
 # This function takes a distinct_id and a feature flag key and returns a float between 0 and 1.
 # Given the same distinct_id and key, it'll always return the same float. These floats are
 # uniformly distributed between 0 and 1, so if we want to show this feature to 20% of traffic
@@ -239,7 +251,12 @@ def match_feature_flag_properties(
                 else:
                     variant = get_matching_variant(flag, distinct_id)
                 return variant or True
+        except RequiresServerEvaluation:
+            # Static cohort or other missing server-side data - must fallback to API
+            raise
         except InconclusiveMatchError:
+            # Evaluation error (bad regex, invalid date, missing property, etc.)
+            # Track that we had an inconclusive match, but try other conditions
             is_inconclusive = True
 
     if is_inconclusive:
@@ -449,8 +466,8 @@ def match_cohort(
     # }
     cohort_id = str(property.get("value"))
     if cohort_id not in cohort_properties:
-        raise InconclusiveMatchError(
-            "can't match cohort without a given cohort property value"
+        raise RequiresServerEvaluation(
+            f"cohort {cohort_id} not found in local cohorts - likely a static cohort that requires server evaluation"
         )
 
     property_group = cohort_properties[cohort_id]
@@ -503,6 +520,9 @@ def match_property_group(
                     # OR group
                     if matches:
                         return True
+            except RequiresServerEvaluation:
+                # Immediately propagate - this condition requires server-side data
+                raise
             except InconclusiveMatchError as e:
                 log.debug(f"Failed to compute property {prop} locally: {e}")
                 error_matching_locally = True
@@ -552,6 +572,9 @@ def match_property_group(
                         return True
                     if not matches and negation:
                         return True
+            except RequiresServerEvaluation:
+                # Immediately propagate - this condition requires server-side data
+                raise
             except InconclusiveMatchError as e:
                 log.debug(f"Failed to compute property {prop} locally: {e}")
                 error_matching_locally = True

posthog/test/test_feature_flags.py

@@ -3013,6 +3013,75 @@ def test_multivariate_feature_flag_payloads(self, patch_flags):
         )
         self.assertEqual(patch_flags.call_count, 0)
 
+    @mock.patch("posthog.client.flags")
+    @mock.patch("posthog.client.get")
+    def test_fallback_to_api_when_flag_has_static_cohort_in_multi_condition(
+        self, patch_get, patch_flags
+    ):
+        """
+        When a flag has multiple conditions and one contains a static cohort,
+        the SDK should fallback to API for the entire flag, not just skip that
+        condition and evaluate the next one locally.
+
+        This prevents returning wrong variants when later conditions could match
+        locally but the user is actually in the static cohort.
+        """
+        client = Client(FAKE_TEST_API_KEY, personal_api_key=FAKE_TEST_API_KEY)
+
+        # Mock the local flags response - cohort 999 is NOT in cohorts map (static cohort)
+        client.feature_flags = [
+            {
+                "id": 1,
+                "key": "multi-condition-flag",
+                "active": True,
+                "filters": {
+                    "groups": [
+                        {
+                            "properties": [
+                                {"key": "id", "value": 999, "type": "cohort"}
+                            ],
+                            "rollout_percentage": 100,
+                            "variant": "set-1",
+                        },
+                        {
+                            "properties": [
+                                {
+                                    "key": "$geoip_country_code",
+                                    "operator": "exact",
+                                    "value": ["DE"],
+                                    "type": "person",
+                                }
+                            ],
+                            "rollout_percentage": 100,
+                            "variant": "set-8",
+                        },
+                    ],
+                    "multivariate": {
+                        "variants": [
+                            {"key": "set-1", "rollout_percentage": 50},
+                            {"key": "set-8", "rollout_percentage": 50},
+                        ]
+                    },
+                },
+            }
+        ]
+        client.cohorts = {}  # Note: cohort 999 is NOT here - it's a static cohort
+
+        # Mock the API response - user is in the static cohort
+        patch_flags.return_value = {"featureFlags": {"multi-condition-flag": "set-1"}}
+
+        result = client.get_feature_flag(
+            "multi-condition-flag",
+            "test-distinct-id",
+            person_properties={"$geoip_country_code": "DE"},
+        )
+
+        # Should return the API result (set-1), not local evaluation (set-8)
+        self.assertEqual(result, "set-1")
+
+        # Verify API was called (fallback occurred)
+        self.assertEqual(patch_flags.call_count, 1)
+
 
 class TestMatchProperties(unittest.TestCase):
     def property(self, key, value, operator=None):