fix(tests): Fix flaky AI conversations tests exceeding Snuba retention (#112612)

Author: mchen-sentry

tests/sentry/api/endpoints/test_organization_ai_conversations.py

@@ -637,7 +637,7 @@ def test_complete_conversation_data_across_time_range(self) -> None:
 
     def test_first_input_last_output(self) -> None:
         """Test firstInput and lastOutput are correctly populated from ai_client spans"""
-        now = before_now(days=90).replace(microsecond=0)
+        now = before_now(days=11).replace(microsecond=0)
         conversation_id = uuid4().hex
         trace_id = uuid4().hex
 
@@ -699,7 +699,7 @@ def test_first_input_last_output(self) -> None:
 
     def test_no_ai_client_spans_filtered_out(self) -> None:
         """Test conversations without input/output are filtered out"""
-        now = before_now(days=91).replace(microsecond=0)
+        now = before_now(days=12).replace(microsecond=0)
         conversation_id = uuid4().hex
         trace_id = uuid4().hex
 
@@ -732,7 +732,7 @@ def test_no_ai_client_spans_filtered_out(self) -> None:
 
     def test_query_filter(self) -> None:
         """Test that query parameter filters conversations"""
-        now = before_now(days=30).replace(microsecond=0)
+        now = before_now(days=25).replace(microsecond=0)
         conversation_id_1 = uuid4().hex
         conversation_id_2 = uuid4().hex
 
@@ -770,7 +770,7 @@ def test_query_filter(self) -> None:
 
     def test_conversation_with_user_data(self) -> None:
         """Test that user data is extracted from spans and returned in the response"""
-        now = before_now(days=100).replace(microsecond=0)
+        now = before_now(days=13).replace(microsecond=0)
         conversation_id = uuid4().hex
         trace_id = uuid4().hex
 
@@ -822,7 +822,7 @@ def test_conversation_with_user_data(self) -> None:
 
     def test_conversation_with_partial_user_data(self) -> None:
         """Test that user is returned even with partial user data"""
-        now = before_now(days=101).replace(microsecond=0)
+        now = before_now(days=14).replace(microsecond=0)
         conversation_id = uuid4().hex
         trace_id = uuid4().hex
 
@@ -857,7 +857,7 @@ def test_conversation_with_partial_user_data(self) -> None:
 
     def test_new_format_input_output_messages(self) -> None:
         """Test that new format gen_ai.input.messages and gen_ai.output.messages are parsed correctly"""
-        now = before_now(days=102).replace(microsecond=0)
+        now = before_now(days=16).replace(microsecond=0)
         conversation_id = uuid4().hex
         trace_id = uuid4().hex
 
@@ -904,7 +904,7 @@ def test_new_format_input_output_messages(self) -> None:
 
     def test_new_format_with_multiple_text_parts(self) -> None:
         """Test that multiple text parts are concatenated correctly"""
-        now = before_now(days=103).replace(microsecond=0)
+        now = before_now(days=17).replace(microsecond=0)
         conversation_id = uuid4().hex
         trace_id = uuid4().hex
 
@@ -955,7 +955,7 @@ def test_new_format_with_multiple_text_parts(self) -> None:
 
     def test_new_format_priority_over_old_format(self) -> None:
         """Test that new format attributes take priority over old format when both exist"""
-        now = before_now(days=104).replace(microsecond=0)
+        now = before_now(days=18).replace(microsecond=0)
         conversation_id = uuid4().hex
         trace_id = uuid4().hex
 
@@ -1010,7 +1010,7 @@ def test_new_format_priority_over_old_format(self) -> None:
 
     def test_new_format_parts_structure(self) -> None:
         """Test that new format with parts structure works correctly"""
-        now = before_now(days=105).replace(microsecond=0)
+        now = before_now(days=19).replace(microsecond=0)
         conversation_id = uuid4().hex
         trace_id = uuid4().hex
 
@@ -1055,7 +1055,7 @@ def test_new_format_parts_structure(self) -> None:
 
     def test_tool_names_populated(self) -> None:
         """Test that toolNames is populated with distinct tool names from tool spans"""
-        now = before_now(days=106).replace(microsecond=0)
+        now = before_now(days=21).replace(microsecond=0)
         conversation_id = uuid4().hex
         trace_id = uuid4().hex
 
@@ -1128,7 +1128,7 @@ def test_tool_names_populated(self) -> None:
 
     def test_tool_errors_counted(self) -> None:
         """Test that toolErrors counts only failed tool spans"""
-        now = before_now(days=107).replace(microsecond=0)
+        now = before_now(days=35).replace(microsecond=0)
         conversation_id = uuid4().hex
         trace_id = uuid4().hex
 
@@ -1212,7 +1212,7 @@ def test_tool_errors_counted(self) -> None:
 
     def test_empty_tool_names_when_no_tool_calls(self) -> None:
         """Test that toolNames is empty when there are no tool calls"""
-        now = before_now(days=108).replace(microsecond=0)
+        now = before_now(days=23).replace(microsecond=0)
         conversation_id = uuid4().hex
         trace_id = uuid4().hex
 
@@ -1257,7 +1257,7 @@ def test_tokens_only_counted_from_ai_client_spans(self) -> None:
         This prevents double counting when both agent spans (invoke_agent) and their
         child ai_client spans have token/cost data.
         """
-        now = before_now(days=109).replace(microsecond=0)
+        now = before_now(days=24).replace(microsecond=0)
         conversation_id = uuid4().hex
         trace_id = uuid4().hex