What does Human-AI Mutual Feeding and the Falsification Ratchet conclude, in numbers, about “A Robust Ratchet: Random Walks with Misclassification, Rollback Audits and the Condition for Compounding”?
Answer: Chapter 5 treats the ratchet as a random walk with misclassification, adds rollback audits and states the condition for compounding: with illustrative parameters the feedback strength must exceed 0.0556.
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 onlyprinted values that match the answer: 0.0556
| Half-life of an unfed library: 124.4 days | ||||||
| rho | growth/day | K90 | K180 | K365 | K730 | limit |
| 0.00 | 0.99444 | 35.5 | 57.0 | 78.2 | 88.5 | 90.0 |
| 0.02 | 0.99644 | 38.6 | 66.6 | 102.3 | 130.2 | 140.6 |
| 0.05 | 0.99944 | 43.9 | 85.7 | 165.2 | 300.1 | 900.0 |
| 0.08 | 1.00244 | 50.3 | 112.9 | 294.1 | 1,011.1 | none |
| Doubling time at rho = 0.08: 284 days | ||||||
| Compounding needs p d u rho > f, i.e. rho > 0.0556 |
Show the full program
# Chapter 5. p53: when does the daily human-agent loop compound?
# Capability index K (count of live, verified rules, weighted by use):
# K_{t+1} = (1 - f) K_t + p d u (h + rho K_t)
# h = 5 human field hypotheses a day; rho = extra hypotheses a day per
# unit of K (the rule library sharpens the human: the feedback arm);
# p = 0.20 gate pass rate; d = 0.5 share of passed rules distilled into
# reusable form; u = 1 unit per distilled rule; f = 1/180 forgetting
# (staleness) per day. Growth factor per day: 1 - f + p d u rho.
# If p d u rho < f, K tends to K* = p d u h / (f - p d u rho);
# otherwise K grows geometrically. Illustrative parameters.
import math
H, P, D, U, F = 5.0, 0.20, 0.5, 1.0, 1.0 / 180.0
def path(rho, days):
k, out = 0.0, {}
for t in range(1, days + 1):
k = (1.0 - F) * k + P * D * U * (H + rho * k)
out[t] = k
return out
if __name__ == "__main__":
print("Half-life of an unfed library: {:.1f} days".format(
math.log(2.0) / -math.log(1.0 - F)))
print(" rho growth/day K90 K180 K365 K730 limit")
for rho in (0.0, 0.02, 0.05, 0.08):
g = 1.0 - F + P * D * U * rho
pth = path(rho, 730)
if P * D * U * rho < F:
lim = "{:.1f}".format(P * D * U * H / (F - P * D * U * rho))
else:
lim = "none"
print(" {:.2f} {:>10.5f} {:>6.1f} {:>6.1f} {:>6.1f} {:>7,.1f}"
" {:>5}".format(rho, g, pth[90], pth[180], pth[365],
pth[730], lim))
g = 1.0 - F + P * D * U * 0.08
print("Doubling time at rho = 0.08: {:.0f} days".format(
math.log(2.0) / math.log(g)))
print("Compounding needs p d u rho > f, i.e. rho > {:.4f}".format(
F / (P * D * U)))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 5: A Robust Ratchet: Random Walks with Misclassification, Rollback Audits and the Condition for Compounding. 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 1 · Variance Collapse under Self-Feeding: Why the Tails Go First
- 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 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