Post-Selection Sensitivity in a GHZ-Type Mermin Statistic: An In-Silico Audit (medium-v10) GHZ 型 Mermin 统计量的后选择敏感性:medium-v10 仿真审计(非硬件复刻)

Abstract

[EN] This deposit documents a rigorous in-silico post-selection audit of a GHZ-type Mermin statistic in the repository's medium-v10 wave model (three local phase-locked sources), motivated by textbook/laboratory GHZ narratives—it is not claimed to replay one specific hardware run. Within that simulator class, we show that chasing high F couples to amplitude gating and low retention, and matched-retention random controls fail to reproduce the gated lift—supporting a selection-driven reading of the reported metric inside the model. Using the V10.4 real cost curve and matched-retention random controls, we show that elevated F values in this simulator class are tightly coupled to selection rules and retention trade-offs. We frame this as a method-level sensitivity result inside the model class, not as a hardware-level claim. [中] 本归档记录对 medium-v10 波动模型中 GHZ 型 Mermin 统计量的严格仿真后选择审计(三个局域相位锁定源),动机来自教科书/实验室叙事,不声称复刻某一具体硬件实验。在该模拟器类内,我们展示:追逐高 F 与振幅门控及低保留率绑定,而同保留率随机对照无法复现门控抬升——从而在模型内支持对报告指标的筛选驱动解读。 通过 V10.4 真实代价曲线和同保留率随机对照,我们展示了:在该模拟器类内,较高 F 与筛选规则及保留率权衡紧密耦合。本文将其定位为模型内的方法学敏感性结果,而非硬件层结论。 Reproducibility (repository): from repo root run python scripts/explore/ghz_medium_v10/v10_4_real_cost_curve.py (see script header for output paths). Companion figure in this deposit: figures/V10_4_REAL_COST_CURVE.png. 可复现(配套仓库): 在仓库根目录运行 python scripts/explore/ghz_medium_v10/v10_4_real_cost_curve.py(输出路径见脚本头部)。本归档插图:figures/V10_4_REAL_COST_CURVE.png。