Fix camera frames in notes and incomplete section caching

Bug 1 — Camera frames appearing in notes:
- _discover_screenshare_lectures() had `except Exception: pass` that
  silently continued when alignment JSON was corrupted, including
  non-screenshare lectures as screenshare. Changed to `continue` so
  corrupted alignments are skipped, not included.

Bug 2 — Incomplete/truncated notes cached and included:
- Empty or too-short LLM drafts (< 100 chars) are no longer written
  to disk. Previously, an empty draft still wrote a heading-only
  file (~30 bytes) that got cached and included in final notes.
- Raised section cache threshold from 50 to 500 bytes. A heading
  alone is ~30 bytes; 50 was too low and let truncated sections
  pass the cache check. 500 bytes ensures meaningful content.
- merge_sections threshold also raised to 500 bytes to match.
- When draft is empty/short, returns a placeholder message telling
  the user to re-run with force, without caching the bad result.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: nodeeeeee

note_generation.py

@@ -777,7 +777,7 @@ def generate_section(
     """Generate (or load from cache) one section and save it to sections_dir."""
     sec_file = _section_path(sections_dir, lec_num, ci, ld.file_idx)
 
-    if not force and sec_file.exists() and sec_file.stat().st_size > 50:
+    if not force and sec_file.exists() and sec_file.stat().st_size > 500:
         if bar:
             fi = f"F{ld.file_idx} " if ld.file_idx > 1 else ""
             bar.set_postfix_str(f"L{lec_num}{fi}§{ci} cached")
@@ -813,8 +813,10 @@ def generate_section(
     draft = _call(NOTE_MODEL, _P("system"), user, _max_tokens(detail))
     tqdm.write(f"     ✓ {len(draft):,} chars  ({_time.monotonic()-_t0:.0f}s)")
 
-    if not draft:
-        tqdm.write(f"  [warn] Empty draft for L{lec_num} §{ci} — skipping")
+    if not draft or len(draft.strip()) < 100:
+        tqdm.write(f"  [warn] Empty or too-short draft for L{lec_num} §{ci} — not caching")
+        heading = f"### {lec_num}.{ci} {_chunk_title(chunk)}"
+        return f"{heading}\n\n*(Section could not be generated — re-run with force to retry.)*", True
 
     if VERIFY_NOTES and draft:
         terms = set()
@@ -1111,7 +1113,7 @@ def merge_sections(
             file_parts: list[str] = []
             for ci in range(1, n_chunks + 1):
                 sec_file = _section_path(sections_dir, ld.num, ci, ld.file_idx)
-                if sec_file.exists() and sec_file.stat().st_size > 50:
+                if sec_file.exists() and sec_file.stat().st_size > 500:
                     file_parts.append(sec_file.read_text(encoding="utf-8"))
                 else:
                     fi = f" F{ld.file_idx}" if multi_file else ""
@@ -1414,15 +1416,15 @@ def _discover_screenshare_lectures(course_dir: Path) -> list[LectureData]:
         # Check for alignment
         align_path = course_dir / "alignment" / f"{fdir.name}.json"
         align = align_path if align_path.exists() else None
-        # Verify alignment source is screenshare
+        # Verify alignment source is screenshare — skip if not
         if align:
             try:
                 with open(align, encoding="utf-8") as f:
                     data = json.load(f)
                 if data.get("source") != "screenshare":
                     continue  # This alignment is from traditional slide matching
             except Exception:
-                pass
+                continue  # Corrupted alignment — skip, don't include
         else:
             continue  # No alignment yet — need frame_extractor to run first
 

test/test_language_and_skip.py

@@ -370,9 +370,10 @@ def test_section_cache_respected(self, tmp_path):
         sections_dir = tmp_path / "sections"
         sections_dir.mkdir()
 
-        # Create a cached section file
+        # Create a cached section file (must be > 500 bytes to pass cache check)
         sec_file = _section_path(sections_dir, 1, 1, 1)
-        sec_file.write_text("### 1.1 Cached Content\n\nThis was previously generated.")
+        content = "### 1.1 Cached Content\n\n" + ("This was previously generated. " * 30)
+        sec_file.write_text(content)
 
         assert sec_file.exists()
         assert sec_file.stat().st_size > 50