The Audit Trilogy: NIST CHSH Denominator Audit + GHZ V10.4 Computed Cost Curve (medium-v10) 审计三部曲:NIST CHSH 分母审计 + GHZ V10.4 计算代价曲线(medium-v10)

[EN] This deposit summarizes two frozen workflows in one methodological bundle.

(1) Bell / CHSH (real public data): Binary CHSH on the NIST complete-blind CSV. Pairing uses tolerances in NIST grid-index units on the exported t axis (not calibrated seconds in this repository). Frozen snapshot: battle_results/nist_completeblind_2015-09-19/battle_result.json (S_strict = 2.336276 at tolerance 0.0 vs S_standard = 2.839387 at 15.0; pair counts 136632 vs 148670). Reproduce the window scan via python scripts/explore/chsh_denominator_audit.py.

(2) GHZ-style statistic (simulation, V10.4 only): The GHZ leg is not tied to ghz_loop_explosion_v19.py. It is anchored to scripts/explore/ghz_medium_v10/v10_4_real_cost_curve.py, which emits a computed_curve (type in V10_4_REAL_COST_CURVE.meta.json): a fixed RNG phase sample (medium_phase_states), soft amplitude gating with parameter gate_k, reported F_gated, mean simultaneous-hit retention R_gated, and a matched-retention random control (F_random_mean, F_random_std). Canonical outputs live under artifacts/ghz_medium_v10/ (V10_4_REAL_COST_CURVE.png, .csv, .meta.json, V10_4_REAL_COST_CURVE_REPORT.md).

Scope statement: This is methodological reproducibility documentation — not a claim that either pipeline certifies a specific hardware Bell or GHZ experiment.

[中] 本归档将两条冻结工作流合并为一条方法论叙述。

(1)贝尔 / CHSH(真实公开数据): 在 NIST complete-blind CSV 上做二元 CHSH。配对容差取导出列 t 上的 NIST 网格指数单位(本仓库不标定为秒)。冻结快照:battle_results/nist_completeblind_2015-09-19/battle_result.json(容差 0.0 时 S_strict = 2.336276,15.0 时 S_standard = 2.839387;配对数 136632 / 148670)。窗口扫描复现:python scripts/explore/chsh_denominator_audit.py。

(2)GHZ 型统计量(仿真,仅 V10.4): GHZ 侧不再绑定 ghz_loop_explosion_v19.py。唯一锚点为 scripts/explore/ghz_medium_v10/v10_4_real_cost_curve.py,产出 computed_curve(见 V10_4_REAL_COST_CURVE.meta.json 中 type):固定 RNG 相位样本(medium_phase_states)、参数为 gate_k 的软振幅门控、报告 F_gated、同时触中平均保留 R_gated,以及同保留率随机对照(F_random_mean、F_random_std)。规范产物位于 artifacts/ghz_medium_v10/(V10_4_REAL_COST_CURVE.png、.csv、.meta.json、V10_4_REAL_COST_CURVE_REPORT.md)。

范围声明: 本文为方法论与可复现性文档——不声称任一脚本对某一具体硬件贝尔或 GHZ 实验作出鉴定。

Reproducibility (repository): Bell: python scripts/explore/chsh_denominator_audit.py (needs data/nist_completeblind_side_streams.csv). GHZ V10.4: python scripts/explore/ghz_medium_v10/v10_4_real_cost_curve.py.

可复现(配套仓库): Bell:python scripts/explore/chsh_denominator_audit.py(需 data/nist_completeblind_side_streams.csv)。GHZ V10.4:python scripts/explore/ghz_medium_v10/v10_4_real_cost_curve.py。