DM2S/NX v9.9: Chapter 6 Checkpoint

Product: "DM2S/NX - The Next Generation Document and Media Management System v9.9" Session: 2026-09-28 19:16 → 20:12 AEST | Andrew.human & Claude Builds on: checkpoint_mod_and_chatlog_now_closed_v9_8.md (Chapter 5, kept, not overwritten). Its §0 standing policies, §5 machines/procedures/paths and §4 corrections all still apply.

---

1. Done this session

• 2419 removed from audio_store.json on Clone. 59 → 58 items. The backup is /var/www/html/ISE_Data/audio_store.json.pre2419.
• /bin scripts confirmed installed on Clone by md5:
  • /bin/json_push_audio_Clone_to_Live v2.2.0: 9db83c50e05113fdaf2ebe04a4f5392e
  • /bin/mirrorISE_Clone: 56866dc59a56622a3895b357c02eb939
• First --force push done. Result: push OK (md5 match) | mirror OK. The push record now exists, so later pushes don't need --force.
  • Store md5 on Clone, on Live, and last pushed: all b54fcaa1b536d9fadb1df36c2e652fd9.
  • A second push correctly reported "Live already has exactly this store -- no push needed".
• Verified on Live with bpm:120-122: 13 hits, down from 14, with no 2419.
  • 2610 is debounced correctly. Its changes are 1:00, 16:00, 23:30, 27:00, 57:00, 1:00:00, 1:03:30.
• Clone AudioProcessor 1.1.1 run: 57 of 57 audio items done, 0 to do.
• Andrew disabled the auto-push on Clone. The audio store reaches Live only when /bin/json_push_audio_Clone_to_Live is run by hand. The mirrors exclude audio_store.json*, so they can't overwrite it.
• Andrew is running both mirrors (Clone and Dev).

2. Found this session

• media_store.json is stale on all three machines. It lacks 6982 and 6998 and still lists 2419.
  • Cause: Live's copy came from Dev, which never saw them (Ch5 §1.1 v1.7.1). Clone's mirror then pulled that stale copy over the one MediaIndexer had updated on Clone.
  • Search is unaffected: MediaSearch 1.7.1 gets file_hash from the DB.
  • The orphan check (audio-store IDs not in media_store) returned ['6982', '6998'] on Clone.
• 6982 still has pre-debounce track changes. It shows ~13 tracks, with changes 0:30, 1:30, 3:30, 5:00, 26:30, 27:00, 31:00, 38:00, 39:00, 52:30, 56:00, 56:30, several of them under 120 s apart. It was measured before audio_features 1.0.1 and never re-measured.
• 3236 "Shelter Me Chip Remix" reads 80-120.8 BPM. A 10th percentile of 80 against a median of 120.8 suggests some windows were read at about ⅔ tempo. The single value is right, but bpm:80 would wrongly match it.
• Junk file a is in Live's /var/www/html/The_ISE_Project/. It holds the saved bpm:120-122 search output. The mirror pulled it to Clone, and Dev gets it on its next sync.

3. Open, in order

1. Re-measure 6982 on Clone, then push:

   cd / && /root/venv-audio-test/bin/python /var/www/html/The_ISE_Project/AudioProcessor.py --features-only --verbose
   /bin/json_push_audio_Clone_to_Live

   If it reports 0 to do, then 1.0.1 didn't bump features_version. Add --force in that case; all 58 items take about 0.1 s per window.

2. Decide where media_store.json is built. One option is Clone → Live → Dev, the same as the audio store. That would need media_store.json* added to the mirror excludes; otherwise the next mirror undoes any re-index.
3. Remove 2419 from media_store.json and add 6982/6998. Either happens with item 2 through a rebuild or re-index from the DB.
4. Delete a from Live's ISE folder, then from Clone and Dev, or let the next mirror do it.
5. PDFsearch 7.7.0 (AND): install on Live, then rsync to Clone and Dev. This is Ch5 §3 item 2.
6. Torch uses one core on Clone. Still waiting on the BLAS/thread output (Ch5 §3 item 3).
7. Known search bugs: a bare /sort newest crashes posts search, and ISEmega's "Show all" does nothing (qf_MegaSearch_bridge.php ignores &all=1).
8. QAT: install ise_qat.py and ise_forge.py in /bin on all three machines, then run the baselines.
9. Genre tags are weak on house music, plus the Ch4 tag review.
10. Duplicates: the Sade copies 6428/6433/6440 (still NO TOPIC FOUND, still in bpm: results) plus the Ch4 list.
11. AudioProcessor wishes:
    • sec_per_window 19.6 → 8.3
    • single-instance lock
    • PID on log lines
    • SIGTERM summary
    • new: the indexer's progress line prints bpm_range instead of the rounded bpm
    • new: the half-tempo low end in bpm_range (3236)
12. clap_diag.py 1.1.0 is on none of the machines (Ch5 §3 item 9).
13. Ideas awaiting Andrew's decision and Chapter 4 leftovers: unchanged, see Ch5 §3 items 10–11.

---

DM2S/NX v9.9 Chapter 6 checkpoint, 2026-09-28 20:12 AEST