# Experiment 001A — Exploratory AI Conclusion v1.1

**Exploratory AI Result: Inconclusive**  
**Preregistered Human Annotation: Pending**

AI-A classified 76/510 M items (14.90%) and 60/510 Controls (11.76%) as HUMAN: absolute risk difference +3.14 percentage points, 95% Newcombe CI −1.05 to +7.33; one-sided Fisher exact p=0.0835 (two-sided p=0.1669); supplemental OR 1.31 (95% Wald CI 0.91–1.89).

AI-B classified 79/510 M items (15.49%) and 61/510 Controls (11.96%) as HUMAN: absolute risk difference +3.53 percentage points, 95% Newcombe CI −0.71 to +7.77; one-sided Fisher exact p=0.0608 (two-sided p=0.1217); supplemental OR 1.35 (95% Wald CI 0.94–1.93).

The preregistered/OPS evidence-family analysis retained the positive direction but remained imprecise: AI-A 61/385 versus 59/475, RD +3.42 pp, 95% CI −1.23 to +8.22, one-sided p=0.0901; AI-B 62/385 versus 60/475, RD +3.47 pp, 95% CI −1.21 to +8.30, one-sided p=0.0883.

Both frozen passes agree in direction and approximate magnitude, but every primary and family-level 95% CI includes zero and every directional Fisher test has p > 0.05. Under the requested three-level vocabulary—Supported, Not Supported, or Inconclusive—the exploratory evidence is therefore **Inconclusive**. It is compatible with a modest positive association but is not sufficiently precise to classify as Supported, and it does not run against the prediction strongly enough to classify as Not Supported.

The immutable v1.0 report used the preregistration's five-level phrase **Tentatively Supported**. That historical record and its checksums are preserved; v1.1 records the current three-level classification without changing any frozen input, annotation, analysis output, or prior checksum.

The restricted analysis key and reusable item-level M/Control identity mapping remain private. Experiment 001B was not started.

---

# Experiment 001A — 探索性 AI 结论 v1.1

**探索性 AI 结果：无法判断（Inconclusive）**  
**预注册真人标注：待进行**

AI-A 将 M 组 76/510（14.90%）和对照组 60/510（11.76%）标为 HUMAN：绝对风险差 +3.14 个百分点，95% Newcombe 置信区间 −1.05 至 +7.33；单侧 Fisher 精确检验 p=0.0835（双侧 p=0.1669）；补充 OR 1.31（95% Wald CI 0.91–1.89）。

AI-B 将 M 组 79/510（15.49%）和对照组 61/510（11.96%）标为 HUMAN：绝对风险差 +3.53 个百分点，95% Newcombe 置信区间 −0.71 至 +7.77；单侧 Fisher 精确检验 p=0.0608（双侧 p=0.1217）；补充 OR 1.35（95% Wald CI 0.94–1.93）。

预注册／OPS 指定的证据家族分析保持正向，但精度仍不足：AI-A 为 61/385 对 59/475，风险差 +3.42 个百分点，95% CI −1.23 至 +8.22，单侧 p=0.0901；AI-B 为 62/385 对 60/475，风险差 +3.47 个百分点，95% CI −1.21 至 +8.30，单侧 p=0.0883。

两份冻结标注在方向和幅度上基本一致，但所有主要分析及家族层分析的 95% 置信区间都包含 0，且所有单侧 Fisher 检验均为 p > 0.05。因此，按本次要求的三级结论词汇——Supported、Not Supported 或 Inconclusive——探索性证据应归为 **Inconclusive（无法判断）**。结果与较小的正向关联相容，但精度不足以判为 Supported；同时也没有稳定地指向预测反面，因此不判为 Not Supported。

不可变的 v1.0 报告采用了预注册五级词汇中的 **Tentatively Supported（初步支持）**。该历史记录及其校验和保持不变；v1.1 只记录当前三级分类，不修改任何冻结输入、标注、分析输出或既有校验和。

restricted analysis key 与可复用的逐项 M/Control 身份映射继续保密。Experiment 001B 未启动。