关于「回归审计的抽样理论:零失败时的三法则与有限总体修正」,《书库冲突检测、可复现与纠删码》给出的定量结论是什么?

结论:第三章给出零失败审计的三法则,要以 95 % 的置信度证明冲突率低于 0.01,需要 299 条全部合格的抽查,且几乎与书库大小无关。

数据口径:书中的书库规模、错误率、风险率与故障率均为示意参数;全部计算由书中程序生成;本书不是法律、财务、采购或存储服务建议。 第一章把书库画成知识图谱,说明卷数越多错误越难藏身:示意书库到 200 卷时,0.9821 的偏差定义处在可以被另一卷反驳的位置。

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: 0.01, 299, 95

(a) upper 95 % bound on the conflict rate
n x0 x = 1 x = 2 x = 5
1000.02951 0.04656 0.06162 0.10225
3000.00994 0.01571 0.02084 0.03472
1,0000.00299 0.00473 0.00628 0.01048
(b) n needed to certify a rate below 0.01
x <0: n = 299; x <= 1: n = 473; x <= 2: n = 628; x <= 3: n = 773
(c) N19,200 claims, D = 192 bad (rate 0.01)
n100: P(no bad in sample) hypergeometric 0.36508, binomial 0.36603
n299: P(no bad in sample) hypergeometric 0.04838, binomial 0.04954
n500: P(no bad in sample) hypergeometric 0.00615, binomial 0.00657
查看完整验算程序
# Chapter 3. p32: audits that find some conflicts, and finite libraries.
# (a) Exact (Clopper-Pearson) one-sided 95 % upper bound when x of n
#     audited claims conflict, found by bisection on the binomial tail.
# (b) Smallest n that certifies a rate below 0.01 if at most x conflicts
#     are allowed in the sample (an acceptance-sampling plan).
# (c) Finite population: a library of 19,200 claims (the illustrative
#     size of p11) with 192 bad ones; the hypergeometric chance that a
#     sample of n contains none, against the binomial approximation.
import math

ALPHA = 0.05


def tail_le(x, n, p):
    terms = [math.comb(n, k) * p ** k * (1 - p) ** (n - k)
             for k in range(x + 1)]
    return math.fsum(terms)


def upper(x, n):
    lo, hi = x / n, 1.0
    for _ in range(80):
        mid = (lo + hi) / 2
        if tail_le(x, n, mid) > ALPHA:
            lo = mid
        else:
            hi = mid
    return hi


print("(a) upper 95 % bound on the conflict rate")
print("{:>6} {:>10} {:>10} {:>10} {:>10}".format(
    "n", "x = 0", "x = 1", "x = 2", "x = 5"))
for n in (100, 300, 1000):
    print("{:>6,} {:>10.5f} {:>10.5f} {:>10.5f} {:>10.5f}".format(
        n, upper(0, n), upper(1, n), upper(2, n), upper(5, n)))
print("(b) n needed to certify a rate below 0.01")
plan = []
for x in (0, 1, 2, 3):
    n = x + 1
    while upper(x, n) > 0.01:
        n += 1
    plan.append("x <= {}: n = {}".format(x, n))
print("    " + ";  ".join(plan))
N, D = 19200, 192
print("(c) N = {:,} claims, D = {} bad (rate 0.01)".format(N, D))
for n in (100, 299, 500):
    hyper = math.comb(N - D, n) / math.comb(N, n)
    print("    n = {:>3}: P(no bad in sample) hypergeometric {:.5f},"
          " binomial {:.5f}".format(n, hyper, (1 - D / N) ** n))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《超大规模双语专著库跨卷定理冲突检测、代码可复现性衰减防腐与冷存储纠删码》第 3 章「回归审计的抽样理论:零失败时的三法则与有限总体修正」(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-182 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026182,
  author    = {Fan, Yafu},
  title     = {Cross-Volume Theorem-Conflict Detection in Large Bilingual Monograph Libraries, Anti-Decay of Code Reproducibility and Erasure-Coded Cold Storage},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论