What does Flywheel Ignition and Replication Losses conclude, in numbers, about “Loop Gain: the Product, the Logarithmic Friction Account and the Weakest Ring”?

Answer: With 100 buyers per turn, the delta interval runs from 0.606 to 2.089 and the bootstrap interval from 0.480 to 1.840, and the estimate falls below one with probability 0.405. A flywheel whose true gain is 1.125 therefore looks as if it were shrinking in four quarters out of ten when the sample is this small. With 400 buyers per turn the interval narrows to 0.826 to 1.533 (0.797 to 1.478 by the bootstrap), but it still contains one, and the chance of an estimate below one is 0.234. Only at 1,600 buyers per turn does the lower bound approach one (0.964, or 0.961 by the bootstrap), and only at 6,400 buyers does it reach 1.041 and clear one with room to spare.

Scope of the numbers: Its parameters are illustrative and all calculations come from its programs. It is not individual business advice, promises no revenue, growth or payback, and discusses only single-tier referral rewards paid to the direct referrer.

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: 0.234, 0.405, 0.480, 0.606, 0.797, 0.826

true G1.1250; 95 percent interval for G-hat
buyers/cycle delta lo delta hi boot lo boot hi P(Ghat<1)
1000.606 2.089 0.480 1.840 0.405
4000.826 1.533 0.797 1.478 0.234
1,6000.964 1.313 0.961 1.295 0.067
6,4001.041 1.215 1.037 1.211 0.002
n400: se of ln G = 0.1579; share of variance by ring:
buy->fb0.301 fb->accept=0.401 accept->pages=0.201
pages->visits0.010 visits->view=0.004 view->buy=0.084
median boot G-hat at n400: 1.1150
Show the full program
# Chapter 1. p13: how wide is a one-quarter estimate of the loop gain?
# The ring ratios telescope: the product of observed ratios equals
# loop-attributed buys next cycle over buys this cycle. Its error comes
# from every ring: var(ln p) ~ (1-p)/(n p) for a conversion, and
# var(ln a) ~ phi/(n a) for an amplification with dispersion phi.
# A parametric bootstrap (fixed seed) checks the delta method.
# Counts are synthetic; true G = 9/8 as in p11.
import math
import random
import statistics

rng = random.Random(41301)
TRUE = [("conv", 0.25), ("conv", 0.5), ("amp", 4.0), ("amp", 120.0),
        ("conv", 0.30), ("conv", 0.0625)]
PHI_VISITS = 6.0  # visits per page are over-dispersed


def delta_ci(n0):
    n, terms = float(n0), []
    for idx, (kind, v) in enumerate(TRUE):
        if kind == "conv":
            terms.append((1.0 - v) / (n * v))
        else:
            phi = PHI_VISITS if idx == 3 else 1.0
            terms.append(phi / (n * v))
        n *= v
    se = math.sqrt(math.fsum(terms))
    g = math.prod(v for _, v in TRUE)
    return g * math.exp(-1.96 * se), g * math.exp(1.96 * se), se


def draw(n, kind, v, idx):
    mean = n * v
    if kind == "conv":
        sd = math.sqrt(n * v * (1.0 - v))
    else:
        phi = PHI_VISITS if idx == 3 else 1.0
        sd = math.sqrt(phi * mean)
    return max(0, int(round(rng.gauss(mean, sd))))


def boot(n0, reps):
    out = []
    for _ in range(reps):
        n = n0
        for idx, (kind, v) in enumerate(TRUE):
            n = draw(n, kind, v, idx)
        out.append(n / n0)
    out.sort()
    return out[int(0.025 * reps)], out[int(0.975 * reps) - 1], out


print("true G = {:.4f}; 95 percent interval for G-hat".format(
    math.prod(v for _, v in TRUE)))
print("buyers/cycle  delta lo  delta hi   boot lo  boot hi  P(Ghat<1)")
for n0 in (100, 400, 1600, 6400):
    lo, hi, se = delta_ci(n0)
    blo, bhi, xs = boot(n0, 2000)
    below = len([x for x in xs if x < 1.0]) / len(xs)
    print("{:>12,}  {:>8.3f}  {:>8.3f}  {:>8.3f}  {:>7.3f}  {:>9.3f}".format(
        n0, lo, hi, blo, bhi, below))
print()
lo, hi, se = delta_ci(400)
print("n=400: se of ln G = {:.4f}; share of variance by ring:".format(se))
n, parts = 400.0, []
for idx, (kind, v) in enumerate(TRUE):
    if kind == "conv":
        parts.append((1.0 - v) / (n * v))
    else:
        parts.append((PHI_VISITS if idx == 3 else 1.0) / (n * v))
    n *= v
tot = math.fsum(parts)
names = ["buy->fb", "fb->accept", "accept->pages", "pages->visits",
         "visits->view", "view->buy"]
print("  " + "  ".join("{}={:.3f}".format(a, p / tot)
                       for a, p in zip(names[:3], parts[:3])))
print("  " + "  ".join("{}={:.3f}".format(a, p / tot)
                       for a, p in zip(names[3:], parts[3:])))
print("median boot G-hat at n=400: {:.4f}".format(
    statistics.median(boot(400, 2000)[2])))

Single source and citation

Fan, Y. (2026). Critical Ignition Conditions for Business Flywheels, Loop-Gain Friction Accounting and Cross-Vertical Fractal Replication (Version 1.0.0), Chapter 1: Loop Gain: the Product, the Logarithmic Friction Account and the Weakest Ring. 5D Library. DOI pending

DOI pending · monograph page goes live after publication · 5DL-B-413 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026413,
  author    = {Fan, Yafu},
  title     = {Critical Ignition Conditions for Business Flywheels, Loop-Gain Friction Accounting and Cross-Vertical Fractal Replication},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

Other chapter answers in this book