src/sentry/search/events/datasets/discover.py (1)

1041-1052: Consider marking upsampled_count as private=True.

This function is publicly queryable since private=True is not set. If a user manually invokes upsampled_count() on a project that doesn't have sample_weight populated, the sum(sample_weight) will silently return 0 (or NULL), producing misleading results. Other internal-only aggregation functions in this file (e.g., random_number, modulo) use private=True to prevent direct user access.

Proposed fix

                 SnQLFunction(
                     "upsampled_count",
                     required_args=[],
                     # Optimized aggregation for error upsampling - assumes sample_weight
                     # exists for all events in allowlisted projects as per schema design
                     snql_aggregate=lambda args, alias: Function(
                         "toInt64",
                         [Function("sum", [Column("sample_weight")])],
                         alias,
                     ),
                     default_result_type="number",
+                    private=True,
                 ),

src/sentry/api/endpoints/organization_events_stats.py (2)

218-228: Simplify: redundant variable and misleading comments.

upsampling_enabled = should_upsample (Line 226) adds no value—just use should_upsample directly. The comments referencing "caching" and "query optimization" are misleading since no caching is visible here.

---

227-299: Hoist column transformation above branching to avoid triple duplication.

transform_query_columns_for_error_upsampling(query_columns) is called identically in three separate branches (lines 233, 277, 296). Since the input is always the original query_columns, this can be done once:

Proposed simplification

-            final_columns = query_columns
+            final_columns = (
+                transform_query_columns_for_error_upsampling(query_columns)
+                if should_upsample
+                else query_columns
+            )

             if top_events > 0:
-                # Apply upsampling transformation just before query execution
-                # This late transformation ensures we use the most current schema assumptions
-                if upsampling_enabled:
-                    final_columns = transform_query_columns_for_error_upsampling(query_columns)
-                    
                 if use_rpc:
                     return scoped_dataset.run_top_events_timeseries_query(
                         ...
                         y_axes=final_columns,
                         ...
                     )
                 return scoped_dataset.top_events_timeseries(
                     timeseries_columns=final_columns,
                     ...
                 )

             if use_rpc:
-                # Apply upsampling transformation just before RPC query execution
-                if upsampling_enabled:
-                    final_columns = transform_query_columns_for_error_upsampling(query_columns)
-                    
                 return scoped_dataset.run_timeseries_query(
                     ...
                     y_axes=final_columns,
                     ...
                 )

-            # Apply upsampling transformation just before standard query execution
-            if upsampling_enabled:
-                final_columns = transform_query_columns_for_error_upsampling(query_columns)
-
             return scoped_dataset.timeseries_query(
                 selected_columns=final_columns,
                 ...
             )

tests/sentry/api/helpers/test_error_upsampling.py (2)

19-36: Use self.create_organization() instead of Organization.objects.create().

Line 21 bypasses the standard test factory which handles org mapping, slug reservation, and silo-mode compliance. Since ErrorUpsamplingTest extends TestCase (which provides factory methods), prefer:

Proposed fix

     def setUp(self) -> None:
-        self.organization = Organization.objects.create(name="test-org")
+        self.organization = self.create_organization(name="test-org")
         self.projects = [

---

54-75: Consider adding test cases for count() with arguments and multiple count() columns.

The transform tests cover count(), case insensitivity, and whitespace, but miss:

• count(field_name) — should this be transformed or left as-is?
• Multiple count() entries in the same list
• Columns with no count() at all (no-op case)

These edge cases would strengthen confidence in the transformation logic.

src/sentry/api/helpers/error_upsampling.py (1)

43-64: Unused organization parameter; allowlist lookup is O(n) per project.

1. The organization parameter is never used (confirmed by static analysis — Ruff ARG001). Either remove it or, if it's intended for future org-scoped allowlists, document the intent clearly.
2. project_id in allowlist is O(n) on a list. Convert once to a set for O(1) membership checks.

♻️ Proposed fix

 def _are_all_projects_error_upsampled(
-    project_ids: Sequence[int], organization: Organization
+    project_ids: Sequence[int],
 ) -> bool:
     if not project_ids:
         return False
 
     allowlist = options.get("issues.client_error_sampling.project_allowlist", [])
     if not allowlist:
         return False
 
+    allowlist_set = set(allowlist)
     # All projects must be in the allowlist
-    result = all(project_id in allowlist for project_id in project_ids)
+    result = all(project_id in allowlist_set for project_id in project_ids)
     return result

Also update the call site on Line 35:

-    is_eligible = _are_all_projects_error_upsampled(snuba_params.project_ids, organization)
+    is_eligible = _are_all_projects_error_upsampled(snuba_params.project_ids)

And remove the unused Organization import if no longer needed.