Fix empty partial activity tail recency

CHANGELOG.md

@@ -3,6 +3,14 @@
 
 ## [Unreleased]
 
+### Fixed
+
+- Empty partial activity rows preserved from cancelled turns no longer define
+  sidebar recency, anchor the initial paginated message window, or get restored
+  after newer completed turns. Long sessions with old activity-only partials
+  after recent replies now stay grouped by their latest real message and open on
+  the recent readable transcript. (#3057)
+
 ## [v0.51.152] — 2026-05-28 — Release DX (stage-batch34 — single-PR optional gateway-backed browser chat)
 
 ### Added

api/models.py

@@ -369,12 +369,35 @@ def _message_timestamp(message):
         return None
 
 
+def _is_empty_partial_activity_message(message):
+    """Return True for cancelled/recovered activity rows with no reply text."""
+    if not isinstance(message, dict):
+        return False
+    if message.get('role') != 'assistant' or not message.get('_partial'):
+        return False
+    content = message.get('content', '')
+    if isinstance(content, str):
+        return not content.strip()
+    if isinstance(content, list):
+        for part in content:
+            if isinstance(part, dict):
+                if part.get('type') == 'text' and str(part.get('text') or part.get('content') or '').strip():
+                    return False
+                continue
+            if str(part or '').strip():
+                return False
+        return True
+    return not str(content or '').strip()
+
+
 def _last_message_timestamp(messages):
     if not isinstance(messages, list):
         return None
     for message in reversed(messages):
         if isinstance(message, dict) and message.get('role') == 'tool':
             continue
+        if _is_empty_partial_activity_message(message):
+            continue
         ts = _message_timestamp(message)
         if ts:
             return ts

api/routes.py

@@ -2167,13 +2167,15 @@ def _message_counts_as_renderable_for_window(message) -> bool:
     """Return true when a paginated window should include this transcript row.
 
     Tool result rows are rendered through their assistant anchor or hidden as raw
-    tool output. A tail page containing only tool rows makes the frontend set
-    ``S.messages`` to a non-empty array while the visible transcript and topbar
-    count stay empty. Anchor small tail windows on the newest non-tool row so
-    long sessions do not open to a blank chat with only transient metadata.
+    tool output. Empty partial activity rows can be preserved after cancellation
+    to keep thinking/tool details inspectable, but they are not reply text. A
+    tail page containing only transient metadata makes the frontend open to
+    collapsed activity while newer real replies sit behind "load older messages".
     """
     if not isinstance(message, dict):
         return False
+    if _is_empty_partial_activity_message(message):
+        return False
     role = str(message.get("role") or "").strip().lower()
     return bool(role and role != "tool")
 
@@ -2551,6 +2553,7 @@ from api.models import (
     get_state_db_session_summary,
     merge_session_messages_append_only,
     _session_message_merge_key,
+    _is_empty_partial_activity_message,
     prune_session_from_index,
     ensure_cron_project,
     is_cron_session,

api/streaming.py

@@ -43,7 +43,11 @@ from api.metering import meter
 from api.run_journal import RunJournalWriter
 from api.turn_journal import append_turn_journal_event_for_stream
 from api.usage import prompt_cache_hit_percent
-from api.models import get_state_db_session_messages, reconciled_state_db_messages_for_session
+from api.models import (
+    _is_empty_partial_activity_message,
+    get_state_db_session_messages,
+    reconciled_state_db_messages_for_session,
+)
 
 # Global lock for os.environ writes. Per-session locks (_agent_lock) prevent
 # concurrent runs of the SAME session, but two DIFFERENT sessions can still
@@ -2448,6 +2452,8 @@ def _restore_display_reasoning_metadata(previous_messages, updated_messages):
     safe_indices = {idx for idx, _ in prev_safe}
     inserted_reasoning_only = 0
     for prev_idx, prev_msg in enumerate(previous_messages):
+        if _is_empty_partial_activity_message(prev_msg):
+            continue
         if prev_idx in safe_indices or not _is_reasoning_only_assistant_message(prev_msg):
             continue
         safe_pos = sum(1 for idx, _ in prev_safe if idx < prev_idx) + inserted_reasoning_only

tests/test_empty_partial_activity_restore.py

@@ -0,0 +1,53 @@
+from api.streaming import _restore_display_reasoning_metadata
+
+
+def test_restore_display_reasoning_skips_empty_partial_activity_rows():
+    previous = [
+        {"role": "user", "content": "old turn", "timestamp": 1},
+        {
+            "role": "assistant",
+            "content": "",
+            "_partial": True,
+            "timestamp": 2,
+            "reasoning": "cancelled thinking",
+            "_partial_tool_calls": [{"name": "terminal", "done": True}],
+        },
+        {"role": "user", "content": "new turn", "timestamp": 3},
+        {"role": "assistant", "content": "new answer", "timestamp": 4},
+    ]
+    updated = [
+        {"role": "user", "content": "old turn"},
+        {"role": "user", "content": "new turn"},
+        {"role": "assistant", "content": "new answer"},
+    ]
+
+    restored = _restore_display_reasoning_metadata(previous, updated)
+
+    assert [m.get("content") for m in restored] == [
+        "old turn",
+        "new turn",
+        "new answer",
+    ]
+    assert not any(m.get("_partial") for m in restored)
+
+
+def test_restore_display_reasoning_keeps_non_partial_thinking_rows():
+    previous = [
+        {"role": "user", "content": "old turn", "timestamp": 1},
+        {
+            "role": "assistant",
+            "content": "",
+            "timestamp": 2,
+            "reasoning": "visible thinking card",
+        },
+        {"role": "assistant", "content": "old answer", "timestamp": 3},
+    ]
+    updated = [
+        {"role": "user", "content": "old turn"},
+        {"role": "assistant", "content": "old answer"},
+    ]
+
+    restored = _restore_display_reasoning_metadata(previous, updated)
+
+    assert restored[1]["reasoning"] == "visible thinking card"
+    assert restored[2]["content"] == "old answer"

tests/test_session_index.py

@@ -83,6 +83,31 @@ def test_compact_exposes_last_message_at_from_message_timestamp():
     assert compact["last_message_at"] == 200.0
 
 
+def test_compact_ignores_empty_partial_activity_for_last_message_at():
+    s = Session(
+        session_id="sess_partial_tail",
+        title="Partial tail",
+        updated_at=300.0,
+        messages=[
+            {"role": "user", "content": "today question", "timestamp": 200.0},
+            {"role": "assistant", "content": "today answer", "timestamp": 201.0},
+            {
+                "role": "assistant",
+                "content": "",
+                "_partial": True,
+                "timestamp": 100.0,
+                "reasoning": "old cancelled thinking",
+                "_partial_tool_calls": [{"name": "terminal", "done": True}],
+            },
+        ],
+    )
+
+    compact = s.compact()
+
+    assert compact["updated_at"] == 300.0
+    assert compact["last_message_at"] == 201.0
+
+
 def test_session_load_allows_hyphenated_safe_ids_but_rejects_traversal():
     sid = "api-182894de593468b6"
     s = _make_session(sid, "API session", updated_at=100)

tests/test_session_message_window_renderable_tail.py

@@ -16,6 +16,28 @@ def test_initial_msg_limit_skips_trailing_tool_only_rows():
     assert offset == 0
 
 
+def test_initial_msg_limit_skips_trailing_empty_partial_activity_rows():
+    messages = [
+        {"role": "user", "content": "today question", "timestamp": 200},
+        {"role": "assistant", "content": "today answer", "timestamp": 201},
+    ] + [
+        {
+            "role": "assistant",
+            "content": "",
+            "_partial": True,
+            "timestamp": 100,
+            "reasoning": f"old cancelled thinking {idx}",
+            "_partial_tool_calls": [{"name": "terminal", "done": True}],
+        }
+        for idx in range(40)
+    ]
+
+    window, offset = _message_window_for_display(messages, msg_limit=5)
+
+    assert [m["content"] for m in window] == ["today question", "today answer"]
+    assert offset == 0
+
+
 def test_msg_limit_keeps_raw_tail_when_it_has_renderable_rows():
     messages = [
         {"role": "user", "content": f"u{idx}"} if idx % 2 == 0 else {"role": "assistant", "content": f"a{idx}"}