From 6ca2c9f884df9f4541ac535b43d666bcd80aef25 Mon Sep 17 00:00:00 2001
From: Rue <343175018+thatsrue@users.noreply.github.com>
Date: Fri, 30 Oct 2026 21:35:11 -0400
Subject: [PATCH] fix: name a linked track's output by the link, not its target

---
 tests/test_batch.py                 | 42 +++++++++++++++++++++++++++++
 transcript_combobulator/pipeline.py | 16 ++++++-----
 2 files changed, 51 insertions(+), 7 deletions(-)

diff --git a/tests/test_batch.py b/tests/test_batch.py
index e89aaf4..4611ccb 100644
--- a/tests/test_batch.py
+++ b/tests/test_batch.py
@@ -739,6 +739,48 @@ def test_symlink_is_placed_by_the_link_not_its_target(self, tmp_path):
         assert out.call_args.args == (link, None)
         assert ts.call_args.args[1] == tmp_path / "out"
 
+    def test_link_named_unlike_its_target_writes_the_vtt_combine_reads(
+        self, tmp_path, monkeypatch
+    ):
+        from transcript_combobulator import config
+        from transcript_combobulator.pipeline import process_file
+
+        monkeypatch.setattr(config, "OUTPUT_DIR", tmp_path / "output")
+        real = tmp_path / "archive" / "4-pikel_123.wav"
+        real.parent.mkdir()
+        real.write_bytes(b"audio")
+        link = tmp_path / "s11" / "4-pikel.wav"
+        link.parent.mkdir()
+        link.symlink_to(real)
+
+        def fake_transcribe(audio_path, output_dir, **kwargs):
+            # Names its files from the audio it is handed, as the real one does.
+            vtt = output_dir / config.vtt_name_for_stem(audio_path.stem)
+            vtt.write_text("WEBVTT\n")
+            return {
+                "vtt_file": str(vtt),
+                "json_file": str(output_dir / f"{audio_path.stem}_transcription.json"),
+                "mapping_file": str(output_dir / f"{audio_path.stem}_mapping.json"),
+                "metrics": {"chunk_count": 0, "failed_chunk_count": 0, "chunks": []},
+            }
+
+        with (
+            patch(
+                "transcript_combobulator.pipeline.needs_conversion", return_value=False
+            ),
+            patch(
+                "transcript_combobulator.pipeline.process_audio",
+                return_value=(tmp_path, []),
+            ),
+            patch(
+                "transcript_combobulator.pipeline.transcribe_segments",
+                side_effect=fake_transcribe,
+            ),
+            patch("transcript_combobulator.pipeline.write_pipeline_manifest"),
+        ):
+            process_file(str(link), session="s11")
+        assert config.vtt_path_for_input(link, "s11").exists()
+
     def test_session_is_passed_through(self, tmp_path):
         track = tmp_path / "1-cadderly.wav"
         track.touch()
diff --git a/transcript_combobulator/pipeline.py b/transcript_combobulator/pipeline.py
index 957afd5..cb70931 100644
--- a/transcript_combobulator/pipeline.py
+++ b/transcript_combobulator/pipeline.py
@@ -57,8 +57,10 @@ def process_file(
     metrics: Optional[dict[str, Any]] = None,
     session: Optional[str] = None,
 ) -> dict[str, Any]:
-    # Read through symlinks, but place output by the path as given.
+    # Read through symlinks, but place and name output by the path as given,
+    # so the VTT lands where vtt_path_for_input looks for it.
     input_file = Path(input_path).resolve()
+    stem = Path(input_path).stem
     setup_logging()
     file_metrics = metrics if metrics is not None else {}
     file_metrics.clear()
@@ -100,8 +102,8 @@ def _progress_callback(phase: str, current: int, total: int) -> None:
         with _timed_stage(file_metrics, "setup"):
             output_dir = get_output_path_for_input(Path(input_path), session)
             output_dir.mkdir(parents=True, exist_ok=True)
-            output_file = output_dir / f"{input_file.stem}.wav"
-            manifest_path = get_manifest_path(output_dir, input_file.stem)
+            output_file = output_dir / f"{stem}.wav"
+            manifest_path = get_manifest_path(output_dir, stem)
             file_metrics["output_dir"] = str(output_dir)
 
         with _timed_stage(file_metrics, "cache_check"):
@@ -114,10 +116,10 @@ def _progress_callback(phase: str, current: int, total: int) -> None:
             return file_metrics
 
         fingerprints = build_stage_fingerprints(input_file)
-        progress_path = get_progress_path(output_dir, input_file.stem)
-        mapping_path = output_dir / f"{input_file.stem}_mapping.json"
-        json_path = output_dir / f"{input_file.stem}_transcription.json"
-        vtt_path = output_dir / vtt_name_for_stem(input_file.stem)
+        progress_path = get_progress_path(output_dir, stem)
+        mapping_path = output_dir / f"{stem}_mapping.json"
+        json_path = output_dir / f"{stem}_transcription.json"
+        vtt_path = output_dir / vtt_name_for_stem(stem)
         with _timed_stage(file_metrics, "cache_invalidation"):
             if force:
                 # A forced run trusts nothing it finds.