[davior]

Pixabay answers GAM's page fetch with a 403, and its terms rule out automated access. GAM does not disguise itself to get past that. Instead, the person does the part that needs a browser, and GAM never contacts the site:

1. Save the page in the browser.
2. Download the wanted tracks with the site's own Download button.
3. Choose both under Add → "Or use a saved page".

Reading the saved page (page_tracks.extract_saved, no network)

• Page address: read from the canonical link, og:url, or Chrome's <!-- saved from url=… --> comment, because adapters key on host and path.

• Pixabay adapter: one rendered row per track. Class names carry build hashes (audioRow--et3ac), so only their stable prefixes are matched. Each row gives:

  • title, artist and track page;
  • duration, genre and tags;
  • the numeric track ID;
  • the listing's /mood/<x>/ (or /genre/) applied to every track.

  On the real saved "mysterious" page, all 20 tracks extract with every field.

• Other saved pages reuse the existing adapters without the network, so a saved Mixkit page works. A saved incompetech page says to use the link instead, because that page only points to pieces.json.

Matching (page_tracks.match_files, mirrored in utils/matchFiles.ts)

• By the site's ID in the filename first. Pixabay names downloads artist-title-604094.mp3.
• Then by title contained in the filename, longest title first ("Tension Riser" before "Riser").
• One file per track.
• Shown before upload, so a wrong match is visible while it can still be fixed.

Importing (page_import.import_saved)

POST /api/assets/import-page/saved/preview and POST /api/assets/import-page/saved are multipart and synchronous, like an upload.

• Every file is stored as music, so it is never transcribed, and is queued for tempo/key/energy analysis.
• Matched files get the page's details, stamped embedded.
• Unmatched files are still imported and reported as "not on the page".
• Checksum duplicates aren't stored again; the page only fills blanks on the existing copy.
• Non-audio files are refused, and an import takes at most 50 files.

PageTrack.audio_url is now optional, for saved pages with no audio link, and match_id is new. A key property replaces audio_url as the identity in the duplicate helpers. Live imports still require an audio URL.

Testing

• pytest -q: 1397 passed. The new tests/test_saved_page_import.py (its fixture is a hand-trimmed copy of the Pixabay row structure) covers:
  • extraction and finding the page's address;
  • Mixkit, and incompetech's note;
  • matching by ID, by title, longest title first, and one-to-one;
  • the full import, including that the file is music, isn't transcribed, and is queued for analysis;
  • unmatched and duplicate files;
  • non-audio files and the file limit.
• npm run format:check, npm run lint, npm test (429 passed, including SavedPageImport.test.tsx) and npm run build all clean.

---

Generated by Claude Code