What does Human-AI Mutual Feeding and the Falsification Ratchet conclude, in numbers, about “Variance Collapse under Self-Feeding: Why the Tails Go First”?
Answer: Self-feeding multiplies the expected variance by (n-1)/n per generation, typical paths shrink faster still, and the median half-life is about half the expected one. The tails leave before the variance does: a fifth less variance costs almost four fifths of the one-in-a-thousand tail. Mixing in a real source gives the variance a steady state, and keeping old real data gives the error an upper bound. Collapse is a conditional statement, not a fate.
Scope of the numbers: Parameters are illustrative and all calculations are generated by the book's programs. Chapter 2 treats the jagged frontier: with illustrative parameters, clicking confirm outside the frontier is right 0.604 of the time against 0.800 for answering alone, while falsifying first and then comparing leads on both sides.
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
| Refit-own-samples recursion, v0 | 1 |
| n shrink c mean half-life log drift median half-life | |
| 10 | 0.9000 6.6 -0.22057 3.1 |
| 20 | 0.9500 13.5 -0.10485 6.6 |
| 50 | 0.9800 34.3 -0.04075 17.0 |
| 100 | 0.9900 69.0 -0.02019 34.3 |
| 1000 | 0.9990 692.8 -0.00200 346.3 |
| Simulation: n | 10, 30 generations, 2,000 paths |
| exact mean variance | 0.0424 |
| simulated mean | 0.0336 |
| simulated median | 0.0013 |
| share of paths v < 0.1 | 0.952 |
Show the full program
# Chapter 1. p11: variance collapse when a model refits its own samples.
# Each generation draws n points from the previous fitted normal and
# refits mean and variance by maximum likelihood (divide by n).
# Exact: E[v_t] = c^t v_0 with c = (n - 1) / n.
# Typical path: ln v_t is a random walk with drift
# d = E ln(chi2_{n-1} / n) = psi((n - 1) / 2) + ln 2 - ln n < 0,
# so the median variance halves after ln 2 / |d| generations, far
# sooner than the mean. A fixed-seed simulation checks n = 10.
# All parameters are illustrative.
import math
import random
import statistics
def digamma(x):
acc = []
while x < 8.0:
acc.append(-1.0 / x)
x += 1.0
inv = 1.0 / x
inv2 = inv * inv
tail = (math.log(x) - 0.5 * inv - inv2 / 12.0 + inv2 * inv2 / 120.0
- inv2 ** 3 / 252.0)
acc.append(tail)
return math.fsum(acc)
def drift(n):
return digamma((n - 1) / 2.0) + math.log(2.0) - math.log(n)
def simulate(rng, n, gens, reps):
finals = []
for _ in range(reps):
mu, sd = 0.0, 1.0
for _ in range(gens):
xs = [rng.gauss(mu, sd) for _ in range(n)]
mu = statistics.fmean(xs)
sd = math.sqrt(statistics.pvariance(xs, mu))
finals.append(sd * sd)
return finals
if __name__ == "__main__":
print("Refit-own-samples recursion, v0 = 1")
print(" n shrink c mean half-life log drift median half-life")
for n in (10, 20, 50, 100, 1000):
c = (n - 1) / n
mean_half = math.log(2.0) / -math.log(c)
d = drift(n)
med_half = math.log(2.0) / -d
print("{:>4} {:>8.4f} {:>14.1f} {:>9.5f} {:>16.1f}".format(
n, c, mean_half, d, med_half))
rng = random.Random(37411)
gens = 30
finals = simulate(rng, 10, gens, 2000)
print("Simulation: n = 10, {} generations, 2,000 paths".format(gens))
print(" exact mean variance {:.4f}".format(0.9 ** gens))
print(" simulated mean {:.4f}".format(statistics.fmean(finals)))
print(" simulated median {:.4f}".format(
statistics.median(finals)))
low = 0
for v in finals:
if v < 0.1:
low += 1
print(" share of paths v < 0.1 {:.3f}".format(low / len(finals)))Single source and citation
Fan, Y. (2026). Co-Evolutionary Human-AI Writing Spirals, Falsification-Gated Ratchets and Symbiotic Capability Compounding (Version 1.0.0), Chapter 1: Variance Collapse under Self-Feeding: Why the Tails Go First. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-374 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026374,
author = {Fan, Yafu},
title = {Co-Evolutionary Human-AI Writing Spirals, Falsification-Gated Ratchets and Symbiotic Capability Compounding},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 2 · The Jagged Frontier: Why Clicking Confirm Loses Accuracy, and Why Better Individuals Make a Narrower Crowd
- Chapter 3 · The Gain from Injected Outliers: How Worse-on-Average Candidates Raise the Best
- Chapter 4 · The Falsification Gate: Sequential Boundaries, Sample Sizes and the Leak Rate
- Chapter 5 · A Robust Ratchet: Random Walks with Misclassification, Rollback Audits and the Condition for Compounding
- Chapter 6 · Multi-Agent Falsification: Majority Error and the Effective Number of Votes under Correlation
- Chapter 7 · Classical Sources: Teaching and Learning in the Xueji, Blue from Indigo in the Xunzi, Bacon's Idols and Popper's Conjectures
- Chapter 8 · Field Tools and Checker Specifications: the Daily Ratchet Ledger, Red-Yellow-Green Criteria and a Pre-registered Refutation