This repository has been archived on 2026-10-19. You can view files and clone it. You cannot open issues or pull requests or push a commit.
Files
DB-cleanup/merge-output/labels/CHECKER-PROMPT.md
StellarCrow 41766c68a2 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.
2026-10-17 10:47:37 +02:00

2.4 KiB

Full label check, second reader: confirm or reject each change

A first reader tested every label of every case against LABEL-RULES.md and changed some values. You judge each changed value. Only confirmed changes reach the database.

Working directory: ~/Documents/Claude/GART/03-research-and-publishing/DB-cleanup/merge-output

Files

  • LABEL-RULES.md: the label tests. Read all of it first.
  • labels/READER-PROMPT.md: the first reader's instructions.
  • labels/in-NN.json: the cases, with current values, summary and dossier path.
  • labels/out-NN.json: the first reader's result. Do not change it.
  • Write your verdicts to labels/check-NN.json. Write nothing else. Read files with the Read tool; never open saved tool-output copies under /home/disi/.claude or /tmp. Fetch nothing from the web.

What to do

  1. For every case and every field where the first reader's value differs from current (compare lists as sets): read the summary, and the dossier if the evidence quote comes from there, and decide whether the new value passes the tests in LABEL-RULES.md better than the current one. Confirm or reject the field as a whole.
  2. Then go through every case once more with LABEL-RULES.md and look for what the first reader left wrong: a current label that plainly fails its test, or a label that plainly passes and is missing. Report those as missed.

Reject when a removed label is supported by the text, an added label rests on someone or something merely mentioned, the change contradicts an owner's rule in LABEL-RULES.md, or the current value was defensible and the change is a matter of taste. When unsure, reject.

Output

A JSON array: one object per changed field, plus one per missed change.

{"id": 123, "field": "approach", "verdict": "confirm"}
{"id": 77, "field": "outcome", "verdict": "reject", "reason": "she was found locked in by police, which is rescued"}
{"id": 301, "field": "who_was_attacked", "verdict": "missed", "proposed": ["holder", "employee or guard"], "quote": "tied up the two clerks", "reason": "staff restrained, not in the list"}

A reject needs a reason. A missed change needs the full proposed value of the field, an exact passage from the text and a reason; it is listed for a person and not applied. Valid JSON, straight quotes.

When you finish

Reply with one line: the batch number, changes confirmed, rejected, missed.