What does Human-AI Mutual Feeding and the Falsification Ratchet conclude, in numbers, about “Field Tools and Checker Specifications: the Daily Ratchet Ledger, Red-Yellow-Green Criteria and a Pre-registered Refutation”?
Answer: Chapter 8 gives the field tools and checker specifications: a daily ratchet ledger, red-yellow-green criteria, the distilled Ratchet Gate and Collapse Alarm Calculator and a pre-registered refutation plan. Marked boxes separate published evidence, runnable Python calculations (20 programs), classical sources and the author's hypotheses.
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: 20
| z_alpha | 1.6449, z_beta = 0.8416 |
| H1: back-filled success, gate vs confirm (rules per arm) | |
| 0.60 vs 0.45: n | 136 |
| 0.60 vs 0.50: n | 305 |
| 0.55 vs 0.50: n | 1,233 |
| H2: library diversity, low vs adequate real share (weeks) | |
| d | 0.8: n = 20 weeks per group |
| d | 0.5: n = 50 weeks per group |
| d | 0.3: n = 138 weeks per group |
| H3: panel error of 5 agents, independent prediction 0.0579 | |
| true 0.08: n | 769 decided candidates |
| true 0.10: n | 229 decided candidates |
| true 0.14: n | 68 decided candidates |
Show the full program
# Chapter 8. p82: sample sizes for the pre-registered tests.
# H1 (gate): rules admitted by the SPRT gate succeed in back-filled
# field use more often than rules admitted by "confirm"; one-sided
# two-proportion test, alpha 0.05, power 0.80, normal approximation:
# n = (z_a sqrt(2 pbar qbar) + z_b sqrt(p1 q1 + p2 q2))^2 / (p1-p2)^2
# H2 (collapse): weekly diversity of the rule library is lower in weeks
# whose real-data share sat below the alarm line; standardised
# difference d, n per group = 2 (z_a + z_b)^2 / d^2 (one-sided).
# H3 (panel): the empirical panel error of k = 5 correlated agents
# exceeds the independent-vote prediction; one-sample binomial test
# of observed error against p0 (normal approximation).
# All effect sizes are illustrative and must be fixed before data.
import math
from statistics import NormalDist
Z = NormalDist()
ZA, ZB = Z.inv_cdf(0.95), Z.inv_cdf(0.80)
def n_two_prop(p1, p2):
pb = (p1 + p2) / 2.0
a = ZA * math.sqrt(2 * pb * (1 - pb))
b = ZB * math.sqrt(p1 * (1 - p1) + p2 * (1 - p2))
return math.ceil((a + b) ** 2 / (p1 - p2) ** 2)
def n_d(d):
return math.ceil(2 * (ZA + ZB) ** 2 / d ** 2)
def n_one_prop(p0, p1):
a = ZA * math.sqrt(p0 * (1 - p0))
b = ZB * math.sqrt(p1 * (1 - p1))
return math.ceil((a + b) ** 2 / (p1 - p0) ** 2)
if __name__ == "__main__":
print("z_alpha = {:.4f}, z_beta = {:.4f}".format(ZA, ZB))
print("H1: back-filled success, gate vs confirm (rules per arm)")
for p1, p2 in ((0.60, 0.45), (0.60, 0.50), (0.55, 0.50)):
print(" {:.2f} vs {:.2f}: n = {:>5,}".format(
p1, p2, n_two_prop(p1, p2)))
print("H2: library diversity, low vs adequate real share (weeks)")
for d in (0.8, 0.5, 0.3):
print(" d = {:.1f}: n = {:>4,} weeks per group".format(d, n_d(d)))
print("H3: panel error of 5 agents, independent prediction 0.0579")
for p1 in (0.08, 0.10, 0.14):
print(" true {:.2f}: n = {:>5,} decided candidates".format(
p1, n_one_prop(0.0579, p1)))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 8: Field Tools and Checker Specifications: the Daily Ratchet Ledger, Red-Yellow-Green Criteria and a Pre-registered Refutation. 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 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