关于「实证闭环:月度公域合规体检、蒸馏小工具与预注册证伪方案」,《公域反垃圾检测与合规桥接页》给出的定量结论是什么?

结论:第八章给出月度公域合规体检、蒸馏小工具「公域合规体检表」与预注册证伪方案,每组需 112 个话题。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;本书是非个人化的经营分析,不构成经营、法律或理财建议,法规与平台规则以上架前人工复核为准。 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。

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: 112, 20

topics per arm; alpha 0.05 two-sided, power 0.80
mu0 per 90 dthetaRR 1.3RR 1.5RR 2.0
5126911238
531174816
20123910034
203873613
simulated power, mu0 5, RR 1.5, theta 1, n 112: 0.798
查看完整验算程序
# Chapter 8. p82: sample size for the pre-registered commission test.
# Design: product topics are randomised to two compliant arms, both
# with disclosed links and both within the platform's published rules:
#   arm 0  short link posts, several a week per topic
#   arm 1  one bridge page per topic plus one or two posts a week
# Outcome: credited orders per topic over 90 days, negative binomial
# with mean mu and dispersion theta (variance mu + mu^2 / theta). By the
# delta method Var(log rate ratio) = (1/mu0 + 1/mu1 + 2/theta) / n, so
#   n per arm = (z_0.975 + z_0.80)^2 (1/mu0 + 1/mu1 + 2/theta) / ln(RR)^2.
# A seeded simulation checks the power for one cell (2,000 trials).
import math
import random
import statistics

ND = statistics.NormalDist()
ZA, ZB = ND.inv_cdf(0.975), ND.inv_cdf(0.80)


def n_arm(mu0, rr, theta):
    mu1 = mu0 * rr
    return math.ceil((ZA + ZB) ** 2 * (1 / mu0 + 1 / mu1 + 2 / theta)
                     / math.log(rr) ** 2)


def nb(rng, mu, theta):
    lam = rng.gammavariate(theta, mu / theta)
    k, p, L = 0, 1.0, math.exp(-lam)
    while True:
        p *= rng.random()
        if p < L:
            return k
        k += 1


def power(rng, mu0, rr, theta, n, trials=2000):
    hit = 0
    for _ in range(trials):
        a = [nb(rng, mu0, theta) for _ in range(n)]
        b = [nb(rng, mu0 * rr, theta) for _ in range(n)]
        m0, m1 = statistics.fmean(a), statistics.fmean(b)
        if m0 == 0 or m1 == 0:
            continue
        v0, v1 = statistics.variance(a), statistics.variance(b)
        se = math.sqrt(v0 / (n * m0 * m0) + v1 / (n * m1 * m1))
        hit += abs(math.log(m1 / m0)) / se > ZA
    return hit / trials


if __name__ == "__main__":
    print("topics per arm; alpha 0.05 two-sided, power 0.80")
    print("mu0 per 90 d   theta   RR 1.3   RR 1.5   RR 2.0")
    for mu0 in (5.0, 20.0):
        for theta in (1.0, 3.0):
            print("{:>13.0f} {:>7.0f} {:>8} {:>8} {:>8}".format(
                mu0, theta, n_arm(mu0, 1.3, theta),
                n_arm(mu0, 1.5, theta), n_arm(mu0, 2.0, theta)))
    rng = random.Random(20261129)
    n = n_arm(5.0, 1.5, 1.0)
    print("simulated power, mu0 5, RR 1.5, theta 1, n {}: {:.3f}".format(
        n, power(rng, 5.0, 1.5, 1.0, n)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《公域社交图谱反垃圾检测边界、合规联盟桥接页转化与限流自愈动力学母典》第 8 章「实证闭环:月度公域合规体检、蒸馏小工具与预注册证伪方案」(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-342 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026342,
  author    = {Fan, Yafu},
  title     = {Social-Graph Anti-Spam Detection Bounds, Compliant Affiliate Bridge-Page Funnels and the Kinetics of Recovery after Reach Throttling},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论