关于「实践与证伪方案:文案检查器、出版清单与预注册检验」,《知识库权威传播与直答片段》给出的定量结论是什么?

结论:第八章给出出版者清单、文案检查器与预注册检验方案,检测一个中等提升每组需要 1,081 次运行。书中以标注框区分已发表的证据、可运行的 Python 计算(20 段程序)、原典与作者的假说。

数据口径:书中参数为示意;全部计算由书中程序生成;本书不是法律、理财或经营方面的个人化建议,也不对任何系统中的引用排名作出承诺。 第一章用幂迭代与随机冲浪者计算 PageRank 与 HITS:在阻尼系数 0.85 的示意知识库里,母典页拿到 0.3240,是均匀份额的 2.59 倍。

English version

对照数据表

下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 1081

z for alpha 1.9600, z for power 0.8416, design effect 1.4
p1liftruns/armwith clusteringqueries/arm
0.220.026,5109,1141,823
0.240.041,6832,357472
0.260.067721,081217
0.300.1029441283
with 1,500 runs per arm the minimum detectable lift is 0.048
查看完整验算程序
# Chapter 8. p82: sample size for a pre-registered test of the book's
# core claim: rewriting a page into direct-answer form raises the share
# of generated answers that cite it. Unit: one (query, engine run).
# Control citation share p0 = 0.20 (illustrative); alpha = 0.05 two-
# sided, power 0.80, two-proportion z-test:
#   n = (z_a sqrt(2 pbar qbar) + z_b sqrt(p0 q0 + p1 q1))^2 / (p1-p0)^2
# Runs cluster within queries: m = 5 runs per query, intra-query
# correlation 0.10, design effect 1 + (m - 1) * 0.10 = 1.4.
import math
from statistics import NormalDist

Z = NormalDist()
ZA = Z.inv_cdf(1 - 0.05 / 2)
ZB = Z.inv_cdf(0.80)
P0 = 0.20
DEFF = 1 + (5 - 1) * 0.10


def n_per_arm(p0, p1):
    pb = (p0 + p1) / 2
    num = (ZA * math.sqrt(2 * pb * (1 - pb))
           + ZB * math.sqrt(p0 * (1 - p0) + p1 * (1 - p1))) ** 2
    return math.ceil(num / (p1 - p0) ** 2)


if __name__ == "__main__":
    print("z for alpha {:.4f}, z for power {:.4f}, design effect {:.1f}"
          .format(ZA, ZB, DEFF))
    print("  p1    lift   runs/arm  with clustering  queries/arm")
    for p1 in (0.22, 0.24, 0.26, 0.30):
        n = n_per_arm(P0, p1)
        nc = math.ceil(n * DEFF)
        print("{:5.2f}  {:5.2f}  {:9,d}  {:15,d}  {:11,d}".format(
            p1, p1 - P0, n, nc, math.ceil(nc / 5)))
    n = 1500
    se = math.sqrt(2 * P0 * (1 - P0) / (n / DEFF))
    print("with {:,} runs per arm the minimum detectable lift is {:.3f}"
          .format(n, (ZA + ZB) * se))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《机器可读知识库超链接权威传播、生成式问答检索截断博弈与高信噪比摘要提取》第 8 章「实践与证伪方案:文案检查器、出版清单与预注册检验」(版本 1.0.0). 5D Library. DOI 待发布

DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-176 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026176,
  author    = {Fan, Yafu},
  title     = {Hyperlink Authority Propagation in Machine-Readable Knowledge Bases, Retrieval-Truncation Games in Generative Answering and High-SNR Snippet Extraction},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论