关于「原典考辨:段玉裁「定其是非之难」与伊拉斯谟 1516 年的回到原文」,《抗模型崩溃双锚工程》给出的定量结论是什么?

结论:第七章考辨段玉裁「定其是非之难」与伊拉斯谟 1516 年的《新约全书》。

数据口径:书中参数为示意;全部计算由书中程序生成,不调用任何模型接口,不使用任何读者或客户的数据;本书是非个人化的经营与工程参考。 书中以标注框区分已发表的证据、可运行的 Python 计算(22 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

gone_witnesswrong_charsk=3k=5k=7
10.99802.000.010.000.00
50.99009.960.300.010.00
100.980219.821.160.080.01
200.960839.254.500.570.08
500.904795.2525.497.452.27
(k columns: expected wrong characters after majority collation)
claim errors per 1000 claims after collation: 30.0 (a = 0.03)
text errors per 1000 chars, g=20, k=5: 0.57
查看完整验算程序
# Chapter 7. p71: Duan Yucai's two kinds of right and wrong, in numbers.
# Model: a text of 1000 characters is copied along independent lines
# of descent; each copy alters a character with probability u = 0.002
# (illustrative). After g copies a character survives in one witness
# with probability (1 - u)^g. Collating k independent witnesses by
# strict majority restores a character when more than k/2 of them
# keep it (errors assumed not to coincide). That repairs the base
# text. It does nothing for a claim that the author got wrong in the
# first place (rate a = 0.03, illustrative): only recomputation or a
# fresh test of the claim reaches that kind of error.
import math

U, L, A = 0.002, 1000, 0.03


def keep(g):
    return (1.0 - U) ** g


def majority(k, q):
    need = k // 2 + 1
    return math.fsum(math.comb(k, j) * q ** j * (1.0 - q) ** (k - j)
                     for j in range(need, k + 1))


print("g    one_witness  wrong_chars  k=3      k=5      k=7")
for g in (1, 5, 10, 20, 50):
    q = keep(g)
    cells = "".join("%-9.2f" % (L * (1.0 - majority(k, q)))
                    for k in (3, 5, 7))
    print("%-4d %-12.4f %-12.2f %s" % (g, q, L * (1.0 - q),
                                         cells.rstrip()))
print("(k columns: expected wrong characters after majority collation)")
q = keep(20)
print("claim errors per 1000 claims after collation: %.1f (a = %.2f)"
      % (L * A, A))
print("text errors per 1000 chars, g=20, k=5: %.2f"
      % (L * (1.0 - majority(5, q))))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《抗模型崩溃双锚真源工程、自产语料回喂方差坍缩速率与标准库验算校勘免疫母典》第 7 章「原典考辨:段玉裁「定其是非之难」与伊拉斯谟 1516 年的回到原文」(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-265 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026265,
  author    = {Fan, Yafu},
  title     = {Anti-Collapse Engineering with Two Anchors of Truth: the Variance-Collapse Rate of Self-Generated Training Data and Immunity through Standard-Library Recomputation and Textual Collation},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论