Sofia/DB-cleanup
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Activity
Files
main
DB-cleanup/merge-output/sofia-worklist.md
T
SofiaandClaude Opus 4.8 ba7529ea50 K&R database cleanup: handoff bundle 
Self-contained bundle to continue the case-summary enrichment pass (70/256 done). sources/ holds the 364 .txt dossiers; scripts use relative paths.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018FJRciSWZc9HftS2edbzBf
2026-09-02 19:49:46 -03:00

22 KiB
Raw Permalink Blame History

Sofia — manual source worklist

Every source the automated pipeline could not retrieve. Open each in your browser, copy the article text, and paste it back to me with the case number. I file it verbatim.

Ordered by urgency: the cases at the top have NO other working source — losing them means the case has zero evidence. Cases lower down already have other articles on file, so their missing source is a nice-to-have, not a gap.

Verdict meaning: DEAD = link 404/gone (try the Wayback Machine). THIN = the page needs a browser to render (JavaScript). BLOCKED = the site refused an automated request but will open normally for you.

NO OTHER SOURCE — priority

has 1 other source(s)

has 2 other source(s)

has 3 other source(s)

has 4 other source(s)

Special: La Provence, paywalled (serves cases 457 AND 458)

We hold only the opening 25%. The rest is behind the paywall and unreachable by any tool. If you can read the full article, paste it and I'll file it into both case files. The missing portion likely contains the Gignac-la-Nerthe detail (458) and any answer on the plate question. https://www.laprovence.com/article/faits-divers-justice/2193885092828214/quatre-hommes-arretes-en-plein-braquage-sur-fond-de-cryptomonnaies-pres-de-marseille

  • Case 391 (Chinese national rescue, 2 arrested) — its only linked source (abs-cbn.com) is JS-rendered and returned no article body. Needs a manual/browser fetch or a replacement source. No usable article on file otherwise.

Round-up articles needing per-case passage isolation (Claude to do, your review)

One article covers several different cases; each case file must keep only its own passage.

  • Known set: 68, 315, 422, 445, 450.
  • 375 (Trincity, Trinidad) — its cryptonews article bundles several separate Trinidad incidents (a St. Joseph car theft at para 17 is a different victim) then goes global. Left UNTRIMMED pending isolation.
  • Newly found: 423 and 424 — both link the SAME San Francisco Chronicle article (byte-identical), a multi-victim wrench-attack round-up. Left UNTRIMMED pending isolation so the two files don't cross-contaminate.

Wrong-source / mismatched cases (the linked article does not describe this record)

  • Case 248 — the linked leparisien.fr article is a DIFFERENT incident (a woman assaulted at home in Suresnes, France), NOT the Oslo Norwegian-family home invasion this record describes. That Suresnes event is case 242, which already has the correct source — so the misfiled article belongs to 242's story. Find a correct Oslo source for 248. (248's summary was written from its record fields only.)
  • Case 441 — its fetched sources (url_2, url_3) actually describe case 438 (same Montreal home-invasion cluster, wrong incident). Decide: drop them, or move to 438.
  • Case 295 — the Forbes source (url_3) predates the event it is filed under, so it cannot be reporting on it. Decide: drop or replace.
  • Case 358 — the DB url points to an unrelated thethaiger dog-in-floods story, not the Bangkok kidnapping. Correct the URL and find the real source (also in the source-gap list — it is sourceless).

Probable duplicate record

  • Case 323 ≡ Case 365 — almost certainly the same incident: three teenagers, a crypto-event host kidnapped at gunpoint on the way home, driven to a remote desert (~1 hr out, White Hills AZ), ~$4M in crypto + NFTs, released in the desert, victim walked ~5 miles to a gas station, a 4th person on speakerphone. 323 (8News, court records) names the charged teens (Belal Ashraf, Austin Fletcher, both 16, Pasco County FL) and puts the event in Downtown Las Vegas — so 365's notes "Downtown LA" is a typo. Decide whether to merge 365 into 323. (365's summary drops the city to avoid asserting the error.)

DB record / flag corrections (surfaced while writing summaries — apply on the website)

Authoritative correction lists already exist: corrections-register.json (35 fixes, with status) and corrections-for-database.json (per-record: id 321 victim, id 307 description, id 258 archive link, id 460 name/ownership/amount/scene-list/weapons, id 462 weapons note, sheet row 18 scenario, sheet row 252 wrong URL / Phuket case). Additional items found during the summary pass:

  • Case 334 — bulk summary said "Victim deceased"; WRONG. The victim (unnamed São Paulo businessman) survived (thrown from the car, 3 fractured ribs, rescued by a truck driver); life_taken = 2 is two SUSPECTS killed by police. Also clarify whether life_taken counts perpetrator deaths (your earlier rule was "counts victims", under which 2 would be wrong here).
  • Case 353 — money_wanted "100 million rubles" is misattributed (that figure is an unrelated Irkutsk case in the same article's tail); this robbery took 100 GPUs / 4 containers, damage only "could significantly exceed 1 million" (currency unspecified).
  • Case 301 — bulk summary "180 BTC" was wrong: it was 180 bitcoin mining MACHINES (ASICs).
  • Case 299 — money_wanted lists the four amounts as "Euros"; the Guardian source says POUNDS (£).
  • Case 290 — description says "$100,000" but the sums taken were Dhs300,000 + Dhs63,000.
  • Case 298 — description says "$1,000,000" while money_wanted/source say Dh4.12m.
  • Case 219 — description "beaten with a stick" contradicts the article (bag simply snatched from behind, no violence); the flags (violence/weapons = 0) already match the article.
  • Case 218 — (a) notes say the husband fired, but the Daily Star article quotes Amouranth herself ("I believe I shot one of them"); (b) theft = 0 but a MacBook WAS taken (she tracked it via Find My).
  • Case 221 — date = March 23 but the article says March 24 (twice); reconcile the day.
  • Case 252 — description "tortured for 4 days" and "possible mistaken identity" are not in the linked Bitcoin.com article (it says only: held several days, found tied up). Confirm or soften.
  • Case 253 — flags all = 0 but the source shows an abduction, assault (hospitalised), firearm threat and a forced crypto transfer → should be kidnappings 1, violence 1, weapons 1, theft 1.
  • weapons = 0 but a weapon was involved: 321, 323 (gunpoint); 290 (knifepoint/blunt objects); 242 (knives + cleaver seized); 243 (tased); 245 (weapon recovered).
  • theft = 0 but property was taken: 303 (Dh183k + phones/cheques); 297 (Dh1.7m cash); 273 (dog, MacBook, jewellery); and see 218 above.
  • kidnappings = 1 but no abduction (attacked/robbed in place): 293, 271, 273, 298.
  • Case 237 — violence_torture = 0 but the source says he was beaten; theft = 1 though nothing was taken (released once found "broke").
  • Case 228 (NYC SoHo torture) — the linked ABC source withholds the victim's name; the DB victim field carries "Michael V.T. Carturan". Confirm the name against a primary document before relying on it.

Sheet-article furniture trim — deferred / accepted items (from the 2026-08-26 pass)

Most furniture was removed cleanly. A few items were left, deliberately:

  • Salford cluster (162, 169, 307, 354, 355) — these share one Manchester Evening News article (sheet row 164) embedded via the special shared-source path, so the furniture trimmer did not touch it. It carries a few photo captions ("(Image: GMP)", "(From top left to right) - Scott Armstrong…"). Low severity; trim row 164 once, and all five inherit it.
  • Other-case CONTENT (not furniture) left in, to revisit with the round-up bucket: 235 Article 2 (France24 arrest piece names the Balland/Noizat incidents — see case 209), 254 Article 2 (references the Las Vegas case 323 and a France case as context). These are coherent articles about a suspect/trend, so cutting is a judgment call, not furniture.
  • Accepted intra-line furniture (cannot remove without editing inside a line, which would break the verbatim-line guarantee): case 204 (a Thai "read the original" link is appended to the last body sentence), case 218 ("Article continues below"/"View 4 Images" prefixes are fused onto body sentences).
  • Residual furniture the automated scan missed (add to sheet-trim-decisions.json next pass): 362 (sheet row 259: "Share / Article / Suggestions" nav lines); 353 (sheet row 133: a trailing "Подробнее на Bits.media: " line); 244 (sheet row 245, Article 2: a truncated "Read more at " stub).

Source gaps found by the reconciliation sweep (add to manual queue)

  • Case 66 (Exxon Gas Station, East Lansdowne PA, 2020) — DB url is a Facebook video (facebook.com/watch), not an article. Find a text source.
  • Case 260 (Nate Geier, Singapore, Oct 2025) — DB url is the victim's own X post (x.com/nategeier). Not an extractable article; find reporting or treat the post as a primary quote.
  • Case 358 (35-y-o Chinese man, Bangkok kidnap, Nov 2025) — DB url is WRONG: it points to an unrelated thethaiger dog-in-floods story. Correct the URL AND find the real source.
  • Case 399 (Young trader, Dominican Republic, Oct 2025) — no url in the DB at all and no spreadsheet article. Sourceless; find one from scratch.
Reference in New Issue View Git Blame Copy Permalink