What does Anti-Collapse Engineering with Two Anchors conclude, in numbers, about “The Empirical Loop: a Two-Anchor Procedure for Every Volume, the Feedback Collapse Rate Calculator and a Pre-Registered Test”?
Answer: Chapter 8 gives an eight-step two-anchor procedure for every volume, the distilled Feedback Collapse Rate Calculator and a pre-registered test. Marked boxes separate published evidence, runnable Python calculations (22 programs), classical sources and the author's hypotheses.
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.
Data table
The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python only
| N=100 n=100 share=0.0 rule=mle p=0.001 | |
| expected_ratio | 0.3660 |
| fresh_coverage_of_p | 0.0000 |
| fresh_items | 0 |
| fresh_needed_95 | 2995 |
| half_life_expected_mle | 68.9676 |
| half_life_typical | 34.3391 |
| mean_drift_accumulate | 1.6351 |
| mean_drift_replace | 101.0000 |
| p_missing_one_gen | 0.9048 |
| typical_ratio_no_fresh | 0.1328 |
| N=100 n=100 share=0.1 rule=mle p=0.001 | |
| expected_ratio | 0.9497 |
| fresh_coverage_of_p | 0.0100 |
| fresh_items | 10 |
| fresh_needed_95 | 2995 |
| half_life_expected_mle | 68.9676 |
| half_life_typical | 34.3391 |
| mean_drift_accumulate | 1.6351 |
| mean_drift_replace | 101.0000 |
| p_missing_one_gen | 0.9048 |
| typical_ratio_no_fresh | 0.1328 |
Show the full program
# Chapter 8. p81: the core of the feedback collapse rate calculator.
# Model: the formulas of p11, p12, p31, p41 and p43 in one place, as
# used by distill/tool.py. Inputs: N points per generation, n
# generations, fresh share a, refit rule (mle or unbiased), the
# probability p of one rare category. Outputs: the expected variance
# ratio, the typical (geometric-mean) ratio without fresh data, the two
# half-lives, the replace and accumulate mean drift N E[e_n^2], the
# chance p is missing after one generation, and fresh coverage of p.
import math
EULER = 0.5772156649015329
def psi_half(k):
if k % 2 == 0:
return -EULER + math.fsum(1.0 / j for j in range(1, k // 2))
m = (k - 1) // 2
return (-EULER - 2.0 * math.log(2.0)
+ math.fsum(2.0 / (2 * j - 1) for j in range(1, m + 1)))
def expected_ratio(n_samp, gens, share, rule):
c = 1.0 if rule == "mle" else n_samp / (n_samp - 1.0)
v, m = 1.0, 0.0
for _ in range(gens):
mix = share + (1.0 - share) * v
v, m = (c * ((1.0 - 1.0 / n_samp) * mix
+ share * (1.0 - share) * m),
(1.0 - share) ** 2 * m + mix / n_samp)
return v
def drift(n_samp, rule):
k = n_samp - 1
div = n_samp if rule == "mle" else k
return psi_half(k) + math.log(2.0) - math.log(div)
def report(n_samp, gens, share, rule, p):
d = drift(n_samp, rule)
fresh = round(share * n_samp)
out = {
"expected_ratio": expected_ratio(n_samp, gens, share, rule),
"typical_ratio_no_fresh": math.exp(gens * d),
"half_life_expected_mle": math.log(0.5)
/ math.log(1.0 - 1.0 / n_samp),
"half_life_typical": math.log(0.5) / d,
"mean_drift_replace": float(gens + 1),
"mean_drift_accumulate": math.fsum(
1.0 / (k * k) for k in range(1, gens + 2)),
"p_missing_one_gen": (1.0 - p) ** n_samp,
"fresh_items": fresh,
"fresh_coverage_of_p": 1.0 - (1.0 - p) ** fresh,
"fresh_needed_95": math.ceil(math.log(0.05)
/ math.log(1.0 - p)),
}
return out
for share in (0.0, 0.1):
r = report(100, 100, share, "mle", 0.001)
print("N=100 n=100 share=%.1f rule=mle p=0.001" % share)
for key in sorted(r):
val = r[key]
txt = "%d" % val if isinstance(val, int) else "%.4f" % val
print(" %-24s %s" % (key, txt))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 8: The Empirical Loop: a Two-Anchor Procedure for Every Volume, the Feedback Collapse Rate Calculator and a Pre-Registered Test. 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
- Chapter 1 · Variance Collapsing Generation by Generation: the Expectation Formula of a Gaussian Feedback Chain and Almost-Sure Extinction
- Chapter 2 · The Tails Die First: Generational Extinction in a Discrete Law and the Absorbing States of Wright–Fisher
- Chapter 3 · Replace or Accumulate: the Counter-Evidence of Gerstgrasser and the Bound of pi^2/6
- Chapter 4 · The Threshold of Mixing in Real Sources: How Much Variance a Fixed Share Preserves, and How Much of the Tail It Misses
- Chapter 5 · Examining for Collapse: Why Inward-Looking Metrics Go Blind, and How Much Data an Outward-Looking Anchor Needs
- Chapter 6 · The First Anchor: Standard-Library Recomputation, Deterministic Fingerprints and the Correlated Failure of Two Anchors
- Chapter 7 · The Classics Examined: Duan Yucai on "the Difficulty of Settling Right and Wrong" and Erasmus's Return to the Original in 1516