What does Library Conflicts, Reproducibility and Erasure Codes conclude, in numbers, about “The Decay of Code Reproducibility: Hazard Rates of Dependency Drift and Byte-Exact Output Regression”?

Answer: Chapter 4 uses a hazard model to show that standard-library programs have a half-life of 34.7 years against 2.2 years with five dependencies, and that a yearly byte-exact output comparison cuts the unknown failure-years of a decade to 9.21 per hundred programs.

Scope of the numbers: Library sizes, error rates, hazards and failure rates are illustrative; all calculations are generated by the programs in the book; the book is not legal, financial, procurement or storage-service advice. Chapter 1 draws the library as a knowledge graph and shows that the more volumes there are, the fewer places an error can hide: at 200 volumes, 0.9821 of deviant definitions in the illustrative library sit where another volume can contradict them.

中文版

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: 2.2, 34.7

deps half-life S(5) S(10) S(20)
034.7 0.9048 0.8187 0.6703
18.7 0.6703 0.4493 0.2019
33.5 0.3679 0.1353 0.0183
52.2 0.2019 0.0408 0.0017
101.1 0.0450 0.0020 0.0000
200.6 0.0022 0.0000 0.0000
library of 3,400 programs, expected still reproducing at year 10:
standard library only2,784
five dependencies each139
0.6 of programs with five1,197
half-life ratio, none against five dependencies 16.0
Show the full program
# Chapter 4. p41: a constant-hazard model of reproducibility decay. A
# program stops reproducing its printed output at rate H0 per year from
# changes in the language and standard library, plus HD per year for
# each third-party dependency it imports (directly or transitively).
# Survival S(t) = exp(-(H0 + d HD) t); half-life ln 2 / (H0 + d HD).
# H0 = 0.02 and HD = 0.06 are illustrative, not measured rates. The last
# lines count expected survivors in a library of 200 volumes with 17
# programs each.
import math

H0, HD = 0.02, 0.06
PROGS = 200 * 17


def surv(d, t):
    return math.exp(-(H0 + d * HD) * t)


print("{:>5} {:>10} {:>10} {:>10} {:>10}".format(
    "deps", "half-life", "S(5)", "S(10)", "S(20)"))
for d in (0, 1, 3, 5, 10, 20):
    print("{:>5} {:>10.1f} {:>10.4f} {:>10.4f} {:>10.4f}".format(
        d, math.log(2) / (H0 + d * HD), surv(d, 5), surv(d, 10),
        surv(d, 20)))
print("library of {:,} programs, expected still reproducing at year 10:"
      .format(PROGS))
print("  standard library only      {:>8,.0f}".format(PROGS * surv(0, 10)))
print("  five dependencies each     {:>8,.0f}".format(PROGS * surv(5, 10)))
mix = math.fsum([0.4 * surv(0, 10), 0.6 * surv(5, 10)])
print("  0.6 of programs with five  {:>8,.0f}".format(PROGS * mix))
print("half-life ratio, none against five dependencies {:.1f}".format(
    (H0 + 5 * HD) / H0))

Single source and citation

Fan, Y. (2026). Cross-Volume Theorem-Conflict Detection in Large Bilingual Monograph Libraries, Anti-Decay of Code Reproducibility and Erasure-Coded Cold Storage (Version 1.0.0), Chapter 4: The Decay of Code Reproducibility: Hazard Rates of Dependency Drift and Byte-Exact Output Regression. 5D Library. DOI pending

DOI pending · monograph page goes live after publication · 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}
}

Other chapter answers in this book