diff --git a/claude_code_log/converter.py b/claude_code_log/converter.py
index 74c4307b..aa409b7e 100644
--- a/claude_code_log/converter.py
+++ b/claude_code_log/converter.py
@@ -4113,6 +4113,9 @@ def convert_jsonl_to(
     shown_sessions: Optional[set[str]] = None
     window_rows: Optional[Dict[str, SessionCacheData]] = None
     record_window = True
+    if input_path.is_dir() and cache_manager is None and generate_individual_sessions:
+        # The session writer needs rows even when no render cache was requested.
+        window_rows = _build_session_data_from_messages(loaded)
     if windowed:
         assert output_path is not None
         if input_path.is_dir():
@@ -5540,10 +5543,12 @@ def _generate_individual_session_files(
                     or not session_file_path.exists()
                 )
             else:
-                # Fallback: no cache, or a format that tracks its own
-                # freshness (JSON) rather than the shared version marker.
+                # Without a cache there is no source fingerprint for an existing
+                # session page; the renderer's version check alone misses appends.
+                # JSON tracks its own version field when a cache is present.
                 should_regenerate_session = (
-                    renderer.is_outdated(session_file_path)
+                    cache_manager is None
+                    or renderer.is_outdated(session_file_path)
                     or from_date is not None
                     or to_date is not None
                     or not session_file_path.exists()
diff --git a/test/test_date_filtered_runs.py b/test/test_date_filtered_runs.py
index e8e9a97b..3e0fd78e 100644
--- a/test/test_date_filtered_runs.py
+++ b/test/test_date_filtered_runs.py
@@ -182,6 +182,42 @@ def _missing(project: Path, shape: str) -> List[str]:
     return missing
 
 
+def test_no_cache_hierarchy_writes_session_pages_linked_from_index(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """An unfiltered no-cache run must not link to absent session pages."""
+    project = _build_project(tmp_path)
+    monkeypatch.setenv("CLAUDE_CODE_LOG_CACHE_PATH", str(tmp_path / "cache.db"))
+
+    process_projects_hierarchy(project.parent, jobs=1, use_cache=False)
+
+    index = (project.parent / "index.html").read_text(encoding="utf-8")
+    for sid in SESSIONS:
+        page = project / f"session-{sid}.html"
+        assert f"session-{sid}.html" in index
+        assert page.is_file(), f"index links to missing {page.name}"
+        assert f"first {sid}" in page.read_text(encoding="utf-8")
+    assert not (tmp_path / "cache.db").exists()
+
+
+def test_no_cache_hierarchy_refreshes_existing_session_page_after_append(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """An uncached rerun must refresh a session page when its source grows."""
+    project = _build_project(tmp_path)
+    monkeypatch.setenv("CLAUDE_CODE_LOG_CACHE_PATH", str(tmp_path / "cache.db"))
+    process_projects_hierarchy(project.parent, jobs=1, use_cache=False)
+    page = project / "session-sess-alpha.html"
+    late = "LATE-UNCACHED-TURN"
+    assert late not in page.read_text(encoding="utf-8")
+
+    _append(project / "sess-alpha.jsonl", _turn("sess-alpha", 2, 2, late))
+    process_projects_hierarchy(project.parent, jobs=1, use_cache=False)
+
+    assert late in page.read_text(encoding="utf-8")
+    assert not (tmp_path / "cache.db").exists()
+
+
 # The hierarchy pass converts only a project with a changed file, so it
 # reaches a filtered run through an append; a project converted on its own
 # reaches it with or without one.
diff --git a/test/test_output_explicit.py b/test/test_output_explicit.py
index 211e1b26..d84caf73 100644
--- a/test/test_output_explicit.py
+++ b/test/test_output_explicit.py
@@ -367,7 +367,11 @@ def test_grown_session_regenerates_without_cache(self, tmp_path: Path):
 
         r2 = runner.invoke(main, [str(proj), "--no-cache"])
         assert r2.exit_code == 0, r2.output
-        assert "skipping regeneration" not in r2.output
+        # Session pages may be current; only the combined page must regenerate.
+        assert (
+            "combined_transcripts.html is current, skipping regeneration"
+            not in r2.output
+        )
         second = combined.read_text(encoding="utf-8")
         assert "SECOND_dir_marker" in second, (
             "stale combined HTML served on a --no-cache directory re-run"