rq3 3B set: judge v3 votes, verification records and payloads (labels of record)

rq3/spurious_train200_3b/judge_v3/README.md

@@ -0,0 +1,9 @@
+# Judge v3 records for rq3_spurious_train200_3b (labels of record, 2026-11-01)
+
+- `votes_v3_runA.jsonl`: 479 original-DB matches x 3 judges (Claude Sonnet 5.5, GPT-5.6-sol, Gemini-4-argon), each vote with its
+  `future_update` and the stored execution check (`verification`). Labels = `evaluation/llm_judge/judge.py replay` in
+  github.com/thanhdath/core-db (Spurious iff >= 2 `no` and >= 1 verified update).
+- `votes_v3_runB.jsonl`: independent rerun on a seeded 150-match subset (self-agreement 148/150 labels).
+- `payloads_v3.jsonl`: the exact user message of every judged case (payload_sha256 = SHA-256 of `user`).
+- `labels_v3_runA.jsonl`: replayed labels with the previous (archived) label for comparison.
+Prompt: `evaluation/llm_judge/prompt.py` (SYSTEM SHA-256 422f67eb...). Temperature 0, OpenRouter.