What does Anti-Collapse Engineering with Two Anchors conclude, in numbers, about “The First Anchor: Standard-Library Recomputation, Deterministic Fingerprints and the Correlated Failure of Two Anchors”?

Answer: Chapter 6 covers the first anchor: standard-library recomputation, the order independence of math.fsum and the correlated failure of two anchors, which together raise the probability of an error-free book of 300 numbers to 0.6769.

Scope of the numbers: Parameters are illustrative; all calculations are generated by the book's programs, which call no model interface and use no reader's or customer's data. Marked boxes separate published evidence, runnable Python calculations (22 programs), classical sources and the author's hypotheses.

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: 0.6769, 300

anchors wrong/number per_300 P(book clean)
none0.05000 15.00 0.0000
recompute0.02030 6.09 0.0021
collate0.03100 9.30 0.0001
both0.00130 0.39 0.6769
shared-kind error, each anchor misses with m0.1
rho0.0: both miss 0.0100, gain over one anchor 10.0 times
rho0.5: both miss 0.0550, gain over one anchor 1.8 times
rho0.9: both miss 0.0910, gain over one anchor 1.1 times
rho1.0: both miss 0.1000, gain over one anchor 1.0 times
check: P(clean) with both0.6769, exp(-n r) = 0.6771
Show the full program
# Chapter 6. p62: one anchor or two, and what correlation costs.
# Model: each number in a book is wrong with probability e0 = 0.05
# (illustrative). A share c = 0.6 of the errors are computational
# (a formula, a sum, a seed); the rest are source errors (a misread
# year, a wrong author, a quote altered). Anchor 1, recomputation with
# the standard library, catches a computational error with probability
# d1 = 0.99 and a source error never. Anchor 2, collation with the
# original, catches a source error with d2 = 0.95 and a computational
# error never. A third kind, an error both anchors could see, is
# missed by each with m = 0.1; the misses have correlation rho.
import math

E0, C, D1, D2 = 0.05, 0.6, 0.99, 0.95
NUMS = 300


def residual(a1, a2):
    comp = E0 * C * ((1.0 - D1) if a1 else 1.0)
    src = E0 * (1.0 - C) * ((1.0 - D2) if a2 else 1.0)
    return comp + src


print("anchors      wrong/number  per_%d  P(book clean)" % NUMS)
for name, a1, a2 in (("none", 0, 0), ("recompute", 1, 0),
                     ("collate", 0, 1), ("both", 1, 1)):
    r = residual(a1, a2)
    print("%-12s %-13.5f %-8.2f %.4f" % (name, r, r * NUMS,
                                         (1.0 - r) ** NUMS))
m = 0.1
print("shared-kind error, each anchor misses with m = %.1f" % m)
for rho in (0.0, 0.5, 0.9, 1.0):
    both = m * m + rho * m * (1.0 - m)
    print("rho = %.1f: both miss %.4f, gain over one anchor %.1f times"
          % (rho, both, m / both))
print("check: P(clean) with both = %.4f, exp(-n r) = %.4f" % (
    (1.0 - residual(1, 1)) ** NUMS, math.exp(-NUMS * residual(1, 1))))

Single source and citation

Fan, Y. (2026). 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 (Version 1.0.0), Chapter 6: The First Anchor: Standard-Library Recomputation, Deterministic Fingerprints and the Correlated Failure of Two Anchors. 5D Library. DOI pending

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

Other chapter answers in this book