Commits the work done since 2026-10-01 that was still local: the four-step label and flag passes, the triage and recheck files, the v1 exports of 2026-10-06 and 2026-10-07, the split dossiers, and the Lopp PR 234 batch (scripts, agent prompts, reader and checker files, the flag re-read of 2026-10-17, the review workbook and the import set of the 57 cases added to stats2 as 506 to 562). Raw HTML and node_modules stay out (.gitignore). The stats2 material in use was copied to gart/gart-databases (stats2/) on 2026-10-17; README-START-HERE.md now says so. This repo is closed.
3.0 KiB
Full label check: test every label of every case
The owner of the Gart database of attacks on crypto holders ordered a check of all labels. For each case you test every label it carries against the label's test, and you look for labels that pass their test and are missing. You add no facts: every change must rest on words in the case's text.
Working directory: ~/Documents/Claude/GART/03-research-and-publishing/DB-cleanup/merge-output
Files
LABEL-RULES.md: one test per label, with the owner's rulings. Read all of it first and apply it literally.step-labels.json: the label lists; use the names exactly as spelled.labels/in-NN.json: your batch. Per case:id,victim_field,current(the seven fields as they stand),locked_by_owner(fields the owner has ruled on: copy them unchanged),flags, the verifiedsummary, and thedossierpath with the full article text.- Write your result to
labels/out-NN.json. Write nothing else. Read files with the Read tool (offset and limit for the large input); never open saved tool-output copies under /home/disi/.claude or /tmp. Fetch nothing from the web.
What to do for each case
Read the summary. Then, field by field (targeting, targeting_accuracy, who_was_attacked, approach, attack, outcome, obtained):
- For each label in
current: does the text pass that label's test in LABEL-RULES.md? Keep it if yes. Remove it if the text plainly fails the test. - Go through the other labels of that field: does the text pass the test of one that is missing? Add it.
- Where the summary is silent on a field that is
unknownor looks wrong, open the dossier and read the articles before deciding. - Order approach and attack as the events ran.
Be conservative. A current label that is defensible stays. Change a value only when the test and the text leave no real doubt. Never change a field listed in locked_by_owner.
Output
A JSON array, one object per case in the batch, same order:
{
"id": 123,
"targeting": ["insider tip-off"],
"targeting_accuracy": "correct",
"who_was_attacked": ["holder", "relative or partner"],
"approach": ["Authority Misuse: service ruse (delivery, driver, worker)"],
"attack": ["Home Invasion"],
"outcome": "released",
"obtained": ["crypto"],
"evidence": {"approach": "posed as a food delivery driver, prompting someone inside to open the front door"}
}
- Give all seven fields for every case, changed or not.
evidenceneeds one entry for every field whose value differs fromcurrent(a different set of labels, or a different single value). Each entry is one continuous passage copied word for word from the summary or from the dossier, the shortest that shows the change is right. A script checks that it appears exactly in one of the two, so do not paraphrase, trim inside it or join passages. For an unchanged field, no entry.- Straight quotes, valid JSON. Check your file parses before you finish.
When you finish
Reply with one line: the batch number, cases written, and the number of fields changed per field name.