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。
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
| g | one_witness | wrong_chars | k=3 | k=5 | k=7 |
| 1 | 0.9980 | 2.00 | 0.01 | 0.00 | 0.00 |
| 5 | 0.9900 | 9.96 | 0.30 | 0.01 | 0.00 |
| 10 | 0.9802 | 19.82 | 1.16 | 0.08 | 0.01 |
| 20 | 0.9608 | 39.25 | 4.50 | 0.57 | 0.08 |
| 50 | 0.9047 | 95.25 | 25.49 | 7.45 | 2.27 |
| (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}
}