Archived
chore: final state before the move to gart/gart-databases
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.
This commit is contained in:
@@ -0,0 +1,60 @@
|
||||
# Reading pass: sort each case into the four steps of an attack
|
||||
|
||||
You are sorting cases from the Gart database of physical attacks on crypto holders. Each case already has a verified summary written from press articles. Your job is to read that summary and record, for each case, how the attack ran in four steps. You add no facts. Every value you choose must be supported by the summary (or, where the summary is silent, by the dossier file).
|
||||
|
||||
Working directory: `~/Documents/Claude/GART/03-research-and-publishing/DB-cleanup/merge-output`
|
||||
|
||||
## Files
|
||||
|
||||
- `steps/in-NN.json`: your batch. One object per case: the verified summary, the old verified labels, the values already sorted (`current`), the flags, and the path of the dossier with the full article text.
|
||||
- `step-labels.json`: the only labels you may use, with a short meaning for each. Read it first. Use the labels exactly as spelled.
|
||||
- `DEFINITIONS.md`: the glossary and the owner's rules. Read the sections on Kidnapping versus Express Kidnapping, Authority Misuse and Burglary.
|
||||
- Write your result to `steps/out-NN.json`. Write nothing else and change no other file.
|
||||
|
||||
## What to record for each case
|
||||
|
||||
```json
|
||||
{
|
||||
"id": 123,
|
||||
"targeting": ["insider tip-off"],
|
||||
"targeting_accuracy": "correct",
|
||||
"who_was_attacked": ["holder"],
|
||||
"approach": ["Authority Misuse: service ruse (delivery, driver, worker)"],
|
||||
"attack": ["Home Invasion"],
|
||||
"outcome": "released",
|
||||
"attack_failed": false,
|
||||
"obtained": ["crypto", "cash"],
|
||||
"demanded": {"amount": 1000000, "currency": "USD", "as_stated": "$1 million in bitcoin", "usd_stated": 1000000},
|
||||
"taken": {"amount": null, "currency": null, "as_stated": "no cryptocurrency was handed over", "usd_stated": null},
|
||||
"evidence": {
|
||||
"targeting": "short quote or close paraphrase from the summary",
|
||||
"approach": "...",
|
||||
"attack": "...",
|
||||
"outcome": "...",
|
||||
"amounts": "..."
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
The file is a JSON array with one such object for every case in your batch, in the same order.
|
||||
|
||||
## Rules
|
||||
|
||||
1. **The four steps.** Targeting is how the attackers knew of the victim. Approach is how they got to the victim. Attack is what they did. Outcome is how it ended for the victim.
|
||||
2. **Keep what is already sorted.** `current` holds values taken from verified labels and owner rulings. Keep every value in `current.attack`, `current.approach` and `current.targeting` unless the summary plainly contradicts it. If you drop a verified attack label, add `"dropped_verified_label_reason": "..."` to the case. Your main work is the empty lists.
|
||||
3. **Several values** are allowed in targeting, who_was_attacked, approach, attack and obtained, when the summary supports each one. Outcome and targeting_accuracy take one value. List approach and attack in the order they happened.
|
||||
4. **Do not state one fact twice in approach.** A crypto-for-cash meeting is `P2P Trade` only, with no Social Engineering value for the same meeting. A fake taxi, fake rideshare or fake delivery is `Authority Misuse: service ruse (delivery, driver, worker)` only. Add a Social Engineering value only for a separate deception.
|
||||
5. **Approach for cases with no lure and no claimed role:** choose from the direct values (seized in the street or a public place; seized at or outside the home; taken from the workplace; vehicle intercepted on the road; forced entry; let in by an insider; break-in, nobody present; remote, no contact). Home Invasion does not imply forced entry: read how they got in.
|
||||
6. **Attack for cases that have only an approach** (for example a P2P Trade with nothing else): say what happened at the meeting. Armed Robbery needs a weapon reported; without one use Robbery without a weapon. If the victim was taken away or held at the attackers' place, it is Kidnapping. Apply the owner's Kidnapping versus Express Kidnapping rule from DEFINITIONS.md.
|
||||
7. **Unknown is a valid answer.** Use `unknown` when the summary and the dossier do not say. Never infer a value from what usually happens. Targeting is often unknown; that is expected.
|
||||
8. **Targeting accuracy.** `wealth overestimated` when the summary shows the victim held far less than the attackers believed, or nothing. `wrong person or address` when the intended holder was someone else. `correct` only when the summary shows the victim did hold what was sought. Otherwise `unknown`.
|
||||
9. **Outcome.** `foiled` means the attackers never gained control of anyone. If the victim died, `killed`. `nobody held or hurt` is for burglaries, swatting and remote cases. `attack_failed` is true when the attackers got nothing and the attack was stopped or abandoned (the old label Dodged Attack marks most of these).
|
||||
10. **Obtained** says what kind of thing was taken. Use `nothing` alone when nothing was taken, and `unconfirmed` alone when the sources disagree or do not say. Stay consistent with the `theft` flag; if the summary contradicts the flag, follow the summary and say so in `evidence.amounts`.
|
||||
11. **Amounts.** `demanded` is what the attackers asked for; `taken` is what they got. Record the figure and currency the summary gives: `amount` as a plain number (2.5 million becomes 2500000), `currency` as an ISO code or coin ticker (USD, EUR, BRL, BTC, USDT). `as_stated` is the wording from the summary, kept short. `usd_stated` is a number only when the summary itself gives a US dollar figure; do no conversion yourself. When sources give different figures, set `amount` to null and put both figures in `as_stated`. When no figure is given, all four are null except `as_stated`, which says so briefly.
|
||||
12. **Evidence** is one short quote or close paraphrase per step, from the summary. For an `unknown`, write what the summary does say or "not stated".
|
||||
13. **When the summary is silent** on approach or outcome, open the case's dossier file (`dossier` path) and look there before answering `unknown`. Cases with `summary_is_verified: false` have only an unverified site summary; sort them the same way and begin `evidence.attack` with "UNVERIFIED SUMMARY:".
|
||||
14. Straight quotes only, no em dashes, valid JSON. Check your file parses before you finish.
|
||||
|
||||
## When you finish
|
||||
|
||||
Reply with one line: the batch number, the number of cases written, and the ids of any case where you dropped or doubted a verified label.
|
||||
Reference in New Issue
Block a user