第9章 偏倚、伦理与隐私(Bias/Ethics/Privacy)


I. 目的与范围(Purpose & Scope)


II. 输入与依赖(Prerequisites & Inputs)


III. 偏倚识别(Bias Identification)


偏倚量化最小集

bias_metrics:

sampling: { stratified_counts: true, ci_method: "wilson" }

measurement: { delta_t_abs_ns: true, delta_tau_ch_ns: true, allan_1s: true }

labeling: { kappa: true, mae: true, dtw: true }

processing: { slice_comparisons: ["low_snr","region_A","device_B"] }


IV. 偏倚缓解(Mitigation)


V. 伦理合规(Ethics Compliance)


VI. 隐私与去标识化(Privacy & De-identification)


VII. 路径量统一口径(Normative Path Forms)

正文显式路径与测度,数据侧记录 delta_form;路径数组 len(gamma_ell)=len(d_ell)=len(n_eff)≥2。


VIII. 质量门与完整性(Gates & Integrity)


IX. 机读策略(Machine-Readable Policies)
A. privacy_policy.yaml

version: "1.0.0"

deid:

techniques: ["hash","mask","generalize"]

k_anonymity: 10

l_diversity: 2

t_closeness: 0.2

access_control:

roles: { reader: ["get"], publisher: ["get","export"], admin: ["get","export","write"] }

retention:

policy_days: 365


B. bias_report.md(提纲)

# Bias Report

- Sampling coverage by strata with CIs

- Measurement bias (δt_abs, Δτ_ch, σ_y(τ), n_eff fit residuals)

- Labeling consistency (κ, MAE, DTW)

- Slices with significant deltas and mitigation plan


C. ethics.md(提纲)

# Ethics Statement

- Consent & purpose limitation

- Data minimization & de-identification

- Governance roles & escalation routes

- Third-party license & redistribution terms


X. 反例与修正(Anti-Patterns & Fixes)


XI. 交叉引用(Cross-References)


XII. 执行勾选清单(Checklist)