What does Paid-Ad Scaling for Paywalls conclude, in numbers, about “Closing the Loop: a Three-Site Ad Ledger, the Distilled Tool and a Pre-Registered Refutation”?
Answer: Chapter 8 gives a three-site ad ledger, the distilled Ad Stop-Loss and Scaling Ladder 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: Under illustrative parameters one icebreaker buyer on tarotmalaysia is worth 2.68 times the icebreaker price, and a first-order break-even stop rule captures only 0.222 of the attainable profit across forty synthetic creatives. Parameters are illustrative and all calculations are generated by the book's programs.
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
| row | spend | buy | V | incCPA | llr | state | next/day |
|---|---|---|---|---|---|---|---|
| D5D d-07 | 61 | 0 | 9.80 | n/a | 3.11 | STOP | 0 |
| D5D d-11 | 140 | 22 | 10.19 | 9.09 | -3.80 | KEEP | 86 |
| TMY m-03 | 420 | 31 | 28.22 | 20.84 | -6.53 | KEEP | 157 |
| TMY m-09 | 160 | 2 | 25.85 | 123.08 | 2.19 | TEST | 100 |
| TMY m-12 | 2,600 | 118 | 22.77 | 33.90 | 3.93 | STOP | 0 |
| TTW t-02 | 3,400 | 30 193.13 | 188.89 | -3.67 | KEEP | 900 | |
| TTW t-05 | 520 | 0 186.64 | n/a | 1.39 | TEST | 600 |
Show the full program
# Chapter 8. p81: one day of the ad ledger for the three live paywalls.
# Each row is one creative: spend, icebreaker buyers from the site's own
# order table, upsells, refunds, and (for kept creatives) the marginal
# ROAS from the last dither pair (chapter 6). Steps:
# 1. funnel value V with upsell and refund rates shrunk toward the
# site prior (prior weight 50 buyers; p11 funnel)
# 2. incremental buyers = buyers * kappa (weekly holdout, chapter 5)
# 3. sequential state, H0 CPA = V, H1 CPA = 2 V (p32): STOP, KEEP or
# TEST; zero buyers at 2.996 V also stops (p21)
# 4. next budget: STOP -> 0; TEST -> hold; KEEP -> x exp(clip(0.25
# (mROAS - 1), -0.2, 0.2)) (p62). All rows are synthetic.
import math
SITE = { # P1, rate, fix, tax, cost1, u0, P2, cost2, refund0, rep, kappa
"D5D": (4.90, 0.044, 0.30, 0.00, 0.20, 0.10, 39.0, 0.60, 0.05, 2.5,
0.70),
"TMY": (9.90, 0.034, 0.50, 0.06, 0.40, 0.12, 129.0, 1.50, 0.06, 6.0,
0.65),
"TTW": (99.0, 0.028, 5.0, 0.05, 4.0, 0.08, 990.0, 15.0, 0.07, 45.0,
0.60)}
CB, W = 0.005, 50.0
A_UP, B_LO = math.log(0.90 / 0.05), math.log(0.10 / 0.95)
ROWS = [ # site, id, budget/day, spend to date, buyers, upsells,
# refunds, mROAS or None
("D5D", "d-07", 60.0, 61.0, 0, 0, 0, None),
("D5D", "d-11", 80.0, 140.0, 22, 3, 1, 1.31),
("TMY", "m-03", 150.0, 420.0, 31, 5, 2, 1.18),
("TMY", "m-09", 100.0, 160.0, 2, 0, 1, None),
("TMY", "m-12", 300.0, 2600.0, 118, 9, 14, 0.74),
("TTW", "t-02", 900.0, 3400.0, 30, 3, 2, None),
("TTW", "t-05", 600.0, 520.0, 0, 0, 0, None)]
def funnel(site, buyers, ups, refunds):
p1, rate, fix, tax, c1, u0, p2, c2, rf0, rep, kap = SITE[site]
cbf = 300.0 if site == "TTW" else 15.0
u = (ups + u0 * W) / (buyers + W)
rf = (refunds + rf0 * W) / (buyers + W)
def net(price, refund, cost):
kept = price * (1 - refund - CB) * (1 - tax)
return kept - (price * rate + fix) - CB * cbf - cost
return math.fsum([net(p1, rf, c1), u * net(p2, 0.04, c2), rep]), kap
if __name__ == "__main__":
print(" row spend buy V incCPA llr state next/day")
for site, cid, bud, spend, buy, ups, rfd, mroas in ROWS:
v, kap = funnel(site, buy, ups, rfd)
inc = buy * kap
llr = inc * math.log(0.5) + 0.5 * spend / v
if llr >= A_UP or (inc == 0 and spend >= 2.996 * v):
state, nxt = "STOP", 0.0
elif llr <= B_LO:
state = "KEEP"
step = 0.0 if mroas is None else 0.25 * (mroas - 1.0)
nxt = bud * math.exp(max(-0.2, min(0.2, step)))
else:
state, nxt = "TEST", bud
cpa = spend / inc if inc > 0 else float("inf")
cpa_s = "{:>7.2f}".format(cpa) if inc > 0 else " n/a"
print(" {} {} {:>8,.0f} {:>4d} {:>6.2f} {} {:>6.2f} {:<5} {:>9,.0f}"
.format(site, cid, spend, buy, v, cpa_s, llr, state, nxt))Single source and citation
Fan, Y. (2026). Paid-Ad Scaling for Diagnostic Paywalls: Meta and TikTok Stopping Rules, Server-Side CAPI Calibration and Hundred-Order Daily Unit Economics (Version 1.0.0), Chapter 8: Closing the Loop: a Three-Site Ad Ledger, the Distilled Tool and a Pre-Registered Refutation. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-345 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026345,
author = {Fan, Yafu},
title = {Paid-Ad Scaling for Diagnostic Paywalls: Meta and TikTok Stopping Rules, Server-Side CAPI Calibration and Hundred-Order Daily Unit Economics},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 1 · Funnel Unit Economics: Why First-Order Break-Even Is Not a Stop Line
- Chapter 2 · Stopping at Zero: the Rule of Three, Gamma-Poisson Posteriors and Spend Red Lines
- Chapter 3 · Signal, Noise and Sequential Tests: Small Budgets Can Only Call "Clearly Bad"
- Chapter 4 · Many Creatives on a Small Budget: Thompson Sampling, Even Splits and Screening Speed
- Chapter 5 · Calibrating Server-Side Events: Deduplication, Net-Value Signals and the Report Inflation Factor
- Chapter 6 · Control Laws for Scaling: Saturation, Carry-Over and Marginal Returns
- Chapter 7 · Classical Sources: the Sunzi on Waging War and Hopkins's Scientific Advertising