Remove 29 dead videos from the data

[banool]

First cleanup pass of filter_dead_media.py (see #239), run standalone against the committed data with extra-patient re-checks (5-minute round spacing), then independently verified ~40 minutes later by verify_removed_media.py (4 rounds, 2-minute spacing, different code path).

Result

• 29 dead video URLs removed (0.52% of 5537 unique URLs), dropping 145 references.
• No dictionary content lost: 0 entries and 0 sub_entries removed — every affected sub_entry has other videos that still work.
• Every removed URL: 5/5 authentic Swift 404s in the filter (status + x-trans-id/x-openstack-request-id + exact not-found body), then 4/4 in the verifier: 29/29 confirmed, 0 false positives, 0 unconfirmed. All 30 sampled kept URLs alive.
• All 29 are also 404 on the R2 mirror (they were never mirrorable), so no user was reaching these via fallback.
• The removed set exactly matches the 29 paths archive_orphaned_media.py reports as referenced-but-missing from the mirror — two independent signals agreeing.

These 29 URLs are what has been driving video_resolve_failed/http_404 in ~5.6% of Auslan sessions (they're heavily shared Definition/Gloss videos). After this merges and clients refresh their data, that metric should fall toward zero.

The 29 removed URLs and the entries that referenced them

Path	Referenced by
/mp4video/Definition/00/60.mp4	accommodate, accommodation, billet, boarder, board (live at), dormitory, hostel, hotel …
/mp4video/Definition/00/83.mp4	act, acting, actor, perform
/mp4video/Definition/19/19233.mp4	eliminate, elimination, exclude, exclusion, exempt, exemption, get rid of, reject …
/mp4video/Definition/20/20442.mp4	run (nose), runny nose
/mp4video/Definition/20/20443.mp4	cure, cured, heal, healed, recovery
/mp4video/Definition/20/20446.mp4	mince, mincemeat
/mp4video/Definition/20/20450.mp4	include, included, inclusion, incorporate, incorporation, involve, involved, involvement …
/mp4video/Definition/20/20451.mp4	alcoholic, alcoholism
/mp4video/Definition/20/20452.mp4	bain marie
/mp4video/Definition/20/20453.mp4	bain marie
/mp4video/Definition/20/20495.mp4	health, healthy, How are you, well (healthy)
/mp4video/Definition/20/20501.mp4	panic, petrified, petrify, scare, scared, terrified, terror
/mp4video/Definition/20/20502.mp4	block (stop), defence, defend, guard, prevent, prevention, protect, protection …
/mp4video/Definition/20/20503.mp4	rape, rapist
/mp4video/Definition/20/20504.mp4	rape, fuck (sexual act)
/mp4video/Definition/20/20510.mp4	doubtful, hazardous, perilous, risk, risky, venture, vulnerable
/mp4video/Definition/20/20514.mp4	chase, pursue, pursuit
/mp4video/Definition/20/20515.mp4	hit, slap
/mp4video/Definition/20/20516.mp4	make love, copulate, copulation, intercourse, sex, sexual, sexual intercourse
/mp4video/Definition/20/20518.mp4	critical, criticise, criticism
/mp4video/Definition/20/20530.mp4	eyewitness, saw (see), seen, witness
/mp4video/Definition/51/5108.mp4	bodyguard, guard, supervise, watch over
/mp4video/Definition/57/5773.mp4	describe, description, explain, explanation, inform, information
/mp4video/Definition/59/5923.mp4	include, included, inclusion, incorporate, incorporation, involve, involved, involvement …
/mp4video/Definition/91/9184.mp4	eliminate, elimination, exclude, exclusion, exempt, exemption, get rid of, reject …
/mp4video/Definition/91/9185.mp4	bond, connected, connection, link, linked, related, relationship
/mp4video/Definition/96/9661.mp4	afraid, fear, fright, frighten, frightened, scare, scared
/mp4video/Gloss/79/7944.mp4	intersectionality
/mp4video/Gloss/82/8271.mp4	opinion

Full per-attempt evidence (headers, body hashes, timings) is in the filter/verify reports; ask if you want them attached.

🤖 Generated with Claude Code

https://claude.ai/code/session_017bum5SFYEZD4gqX8PZhT6i

[banool]

Superseded by #241 — same branch, rebased onto master now that #239 has merged. This PR couldn't be reopened because its base branch banool/filter-dead-media was deleted on merge.

---

Generated by Claude Code