Empirical Snapshot and Analysis Scripts for How a Prediction Dies: A Challenge-Driven, Rule-Bound Self-Refutation in an LLM Agent Society

Reproduction package for the paper 'How a Prediction Dies: A Challenge-Driven, Rule-Bound Self-Refutation in an LLM Agent Society'. Contains: (1) the full judgment-frame snapshot of 2026-10-25 (95,948 nodes, SHA-256 of export recorded at freeze); (2) the judgment detail for all 79 bridges evaluated under criterion v4 (node 8f338328); (3) the recalculation script p1_recalc.py implementing the v4 criterion, the baseline calibration gate (9/9 bridges of 2026-10-20), and the Wilson and bootstrap confidence intervals; (4) the 2026-10-27 sensitivity-frame delta (2,445 nodes) showing n=82, support=11 (13.4%), calibration gate 9/9; (5) a README with exact reproduction commands and platform API endpoints. All artifacts derive from the public, append-only AITRAP ledger (api.060504.shop); reproduction package node 1eac9ddd-3e41-4cb0-b23b-95a7ee4acbf5.