What does Computational Bibliometrics and Self-Verifying Books conclude, in numbers, about “Proofreading Floors and Whitelist Gates”?
Answer: Chapter 5 shows that when errors differ in visibility three human passes still leave 0.14950 of them, and that a whitelist cannot catch swaps between true numbers.
Scope of the numbers: Its readers, books, error rates and costs are illustrative models; all calculations are generated by its programs; the book is not legal, financial or publishing advice. Marked boxes separate published evidence, runnable Python calculations (16 programs), classical sources and the author's hypotheses.
Data table
The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python onlyprinted values that match the answer: 0.14950
| passes | equal p=0.70 | Beta(1.4,0.6) | Beta(0.7,0.3) | MC(0.7,0.3) |
| 1 | 0.30000 | 0.30000 | 0.30000 | 0.29907 |
| 2 | 0.09000 | 0.16000 | 0.19500 | 0.19368 |
| 3 | 0.02700 | 0.10400 | 0.14950 | 0.14975 |
| 5 | 0.00243 | 0.05741 | 0.10607 | 0.10545 |
| 10 | 0.00001 | 0.02401 | 0.06600 | 0.06582 |
| errors left of 81 after | 3 passes: equal 2.19, Beta(0.7,0.3) 12.11 | |||
| errors left of 81 after 10 passes: equal 0.00, Beta(0.7,0.3) 5.35 |
Show the full program
# Chapter 5. p09: Why human proofreading has a floor. Each error has its
# own detectability p; a pass finds it with chance p. If every error had
# p = 0.70, k passes would leave 0.30^k. But errors differ: with p drawn
# from Beta(a, b) of the same mean 0.70, the share missed after k passes
# is B(a, b + k) / B(a, b), which falls only like k^-a. Draft: 81 wrong
# numbers (450 at 0.18). We also print a Monte Carlo of the Beta case
# (Johnk's sampler, built on random() alone), fixed seed 101.
import random
from math import lgamma, exp
rng = random.Random(101)
def beta_miss(a, b, k):
lb = lambda x, y: lgamma(x) + lgamma(y) - lgamma(x + y)
return exp(lb(a, b + k) - lb(a, b))
def johnk(a, b):
while True:
x = rng.random() ** (1 / a)
y = rng.random() ** (1 / b)
if 0 < x + y <= 1:
return x / (x + y)
def mc_miss(a, b, k, n=200000):
left = 0
for _ in range(n):
p = johnk(a, b)
if all(rng.random() > p for _ in range(k)):
left += 1
return left / n
if __name__ == "__main__":
print(" passes equal p=0.70 Beta(1.4,0.6) Beta(0.7,0.3)"
" MC(0.7,0.3)")
for k in (1, 2, 3, 5, 10):
print(" {:6d} {:12.5f} {:13.5f} {:13.5f} {:11.5f}".format(
k, 0.3 ** k, beta_miss(1.4, 0.6, k), beta_miss(0.7, 0.3, k),
mc_miss(0.7, 0.3, k)))
for k in (3, 10):
print(" errors left of 81 after {:2d} passes: equal {:.2f},"
" Beta(0.7,0.3) {:.2f}".format(k, 81 * 0.3 ** k,
81 * beta_miss(0.7, 0.3, k)))Single source and citation
Fan, Y. (2026). Computational Bibliometrics, Falsification-First Rhetoric and Self-Verifying Monograph Compilers (Version 1.0.0), Chapter 5: Proofreading Floors and Whitelist Gates. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-101 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026101,
author = {Fan, Yafu},
title = {Computational Bibliometrics, Falsification-First Rhetoric and Self-Verifying Monograph Compilers},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 1 · Two Survival Laws: Why Most Books Fade and a Few Keep Being Read
- Chapter 2 · Surprise, Anchors and Completion
- Chapter 3 · Falsify, Then Bound: The Bayesian Case and Its Limits
- Chapter 4 · The Trust Ratchet Turns Both Ways
- Chapter 6 · The Self-Verifying Compiler: Code Points and the Whole Pipeline
- Chapter 7 · Liu Xie, Dai Zhen and Knuth: Old Rules for Checkable Writing
- Chapter 8 · Practice: Costs, the Long Tail and a Checklist