fix: a download is marked seen only after it is imported

A run killed between download and import left its files on disk and in the
seen-ledger but never in the library, and every later walk trusted the
ledger. TamadaHeijun's 12PCG post lost 7 of 13 images this way (stranded
run 90402), which read as the duplicates filter.

The ingester now hands phase 3 a mark_seen_after_import hook, called after
the import loop. A file on disk with no ImageRecord at its path is fed to
import, not reconciled into the ledger. Recapture reaches files already
orphaned, because it looks past the ledger to the disk.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LVjrnpQjRgHdvq95rASoiR

backend/app/services/ingest_core.py

@@ -36,6 +36,7 @@ from datetime import UTC, datetime, timedelta
 from sqlalchemy import delete, func, select, text
 from sqlalchemy.dialects.postgresql import insert as pg_insert
 
+from ..models import ImageRecord
 from .gallery_dl import (
     DownloadResult,
     ErrorType,
@@ -261,6 +262,9 @@ class Ingester:
         # source_filehash and (b) link the on-disk image to its Post (#1288) —
         # WITHOUT re-downloading or unlinking the file. Empty outside recapture.
         relink: list[tuple[str, str, str]] = []
+        # Media handed to phase 3 for import. Marked seen by phase 3 once the
+        # import has run (`mark_seen_after_import`), not here — see there.
+        fetched: list[tuple[str, str]] = []
         downloaded = 0
         errors = 0
         quarantined = 0
@@ -313,6 +317,7 @@ class Ingester:
                 written_paths=written,
                 post_record_paths=list(post_records),
                 relink_source_paths=list(relink),
+                mark_seen_after_import=lambda: self._mark_seen(source_id, fetched),
                 stdout="\n".join(log_lines),
                 stderr="",
                 return_code=return_code,
@@ -512,6 +517,13 @@ class Ingester:
                     recapture=recapture,
                 )
 
+                # An on-disk file is only "done" if something imported it. One
+                # with no ImageRecord at its path was written by a run that died
+                # before phase 3 — it goes to import, not to the ledger.
+                imported_paths = self._recorded_paths([
+                    str(o.path) for o in outcomes
+                    if o.status == "skipped_disk" and o.path is not None
+                ])
                 to_mark: list[tuple[str, str]] = []
                 to_clear: list[str] = []  # recovered → drop any dead-letter row
                 to_fail: list[tuple[str, str, str]] = []  # (key, post_id, error)
@@ -523,11 +535,29 @@ class Ingester:
                         downloaded += 1
                         if outcome.path is not None:
                             written.append(str(outcome.path))
-                        to_mark.append((key, media_item.post_id))
+                        fetched.append((key, media_item.post_id))
                         to_clear.append(key)
                         consecutive_seen = 0
+                    elif (
+                        outcome.status == "skipped_disk"
+                        and outcome.path is not None
+                        and str(outcome.path) not in imported_paths
+                    ):
+                        # On disk, never imported: a prior run wrote it and died
+                        # before phase 3. Import it now. Safe to feed to
+                        # attach_in_place because no record owns this path —
+                        # the unlink below is about a file that IS the record.
+                        written.append(str(outcome.path))
+                        fetched.append((key, media_item.post_id))
+                        to_clear.append(key)
+                        skipped_count += 1
+                        consecutive_seen += 1
+                        log_lines.append(
+                            f"  post {media_item.post_id} — on disk but never "
+                            f"imported: {outcome.path.name}"
+                        )
                     elif outcome.status == "skipped_disk":
-                        # Already on disk (a prior run). Reconcile the ledger so a
+                        # Already on disk and imported. Reconcile the ledger so a
                         # later tick skips it at tier-1 without a disk stat, but
                         # do NOT re-feed it to phase 3 — attach_in_place would see
                         # the duplicate sha256 and unlink the on-disk copy.
@@ -887,6 +917,16 @@ class Ingester:
             )
             session.commit()
 
+    def _recorded_paths(self, paths: list[str]) -> set[str]:
+        """Which of `paths` an ImageRecord already points at."""
+        if not paths:
+            return set()
+        with self.session_factory() as session:
+            rows = session.execute(
+                select(ImageRecord.path).where(ImageRecord.path.in_(paths))
+            ).scalars().all()
+        return set(rows)
+
     def _mark_seen(self, source_id: int, items: list[tuple[str, str]]) -> None:
         """Idempotent upsert of (filehash, post_id) seen-ledger rows for a page.