What does The Nightly Funnel conclude, in numbers, about “Elimination Rules: Minimum Exposure, Posterior Win Probability and Time-uniform Confidence Sequences”?
Answer: Take the four lines at test level 0.05 first. The false-deletion rate lies between 0.003 and 0.006, and the deletion rate of the good version is always 0.000. The share of weak versions deleted by night 28 lies between 0.139 and 0.143, rising to between 0.532 and 0.577 by night 56. Raising the minimum exposure from 0 to 8,000 makes almost no difference.
Scope of the numbers: Its parameters are illustrative; all calculations are generated by its 23 programs; it is general business analysis, not business, legal or financial advice, and it promises no rise in conversion.
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: 0.000, 0.003, 0.006, 0.05, 0.139, 0.143
| min n alpha false kill good killed bad caught d28 d56 | ||||
| 0 | 0.05 | 0.003 | 0.000 | 0.142 0.559 |
| 2,000 | 0.05 | 0.006 | 0.000 | 0.143 0.577 |
| 4,000 | 0.05 | 0.004 | 0.000 | 0.139 0.534 |
| 8,000 | 0.05 | 0.005 | 0.000 | 0.141 0.532 |
| 0 | 0.10 | 0.010 | 0.000 | 0.258 0.677 |
| 2,000 | 0.10 | 0.012 | 0.000 | 0.239 0.671 |
| 4,000 | 0.10 | 0.012 | 0.000 | 0.251 0.682 |
| 8,000 | 0.10 | 0.011 | 0.000 | 0.252 0.663 |
Show the full program
# Chapter 5. p53: minimum exposure and alpha for the p52 rule, over
# 56 nights. For each setting: false kill rate when B equals A (0.050),
# kill rate of a 10 percent better B (0.055), catch rate of a 10
# percent worse B (0.045) by night 28 and by night 56. Same boundary
# and traffic as p52; 2,000 runs per cell.
import math
import random
from statistics import NormalDist
PHI = NormalDist().cdf
DAYS, PER, REPS, BAR = 56, 600, 2000, 0.05
RHO = 10000 * 2 * 0.05 * 0.95
rng = random.Random(40553)
def binom(n, p):
x = round(n * p + math.sqrt(n * p * (1 - p)) * rng.gauss(0, 1))
return min(max(x, 0), n)
def p_better(sa, sb, n):
a1, b1, a2, b2 = 1 + sa, 1 + n - sa, 1 + sb, 1 + n - sb
m1, m2 = a1 / (a1 + b1), a2 / (a2 + b2)
v1 = a1 * b1 / ((a1 + b1) ** 2 * (a1 + b1 + 1))
v2 = a2 * b2 / ((a2 + b2) ** 2 * (a2 + b2 + 1))
return PHI((m2 - m1) / math.sqrt(v1 + v2))
def upper_gap(sa, sb, n, alpha):
pa, pb = (sa + 1) / (n + 2), (sb + 1) / (n + 2)
v = n * (pa * (1 - pa) + pb * (1 - pb))
half = math.sqrt((v + RHO) * (2 * math.log(1 / alpha)
+ math.log((v + RHO) / RHO)))
return (sb - sa + half) / n
def kill_day(pb, min_n, alpha):
sa = sb = n = 0
for d in range(1, DAYS + 1):
sa += binom(PER, 0.050)
sb += binom(PER, pb)
n += PER
if (n >= min_n and p_better(sa, sb, n) < BAR
and upper_gap(sa, sb, n, alpha) < 0):
return d
return None
print("min n alpha false kill good killed bad caught d28 d56")
for alpha in [0.05, 0.10]:
for min_n in [0, 2000, 4000, 8000]:
out = []
for pb in [0.050, 0.055, 0.045]:
ds = [kill_day(pb, min_n, alpha) for _ in range(REPS)]
k28 = len([d for d in ds if d is not None and d <= 28])
k56 = len([d for d in ds if d is not None])
out.append((k28 / REPS, k56 / REPS))
print("{:>5,} {:>6.2f} {:>11.3f} {:>12.3f} {:>15.3f} {:>5.3f}"
.format(min_n, alpha, out[0][1], out[1][1], out[2][0],
out[2][1]))Single source and citation
Fan, Y. (2026). Autonomous Nightly Funnel Micro-Surgery, Bandit Traffic Allocation and Zero-Touch Conversion Self-Improvement (Version 1.0.0), Chapter 5: Elimination Rules: Minimum Exposure, Posterior Win Probability and Time-uniform Confidence Sequences. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-405 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026405,
author = {Fan, Yafu},
title = {Autonomous Nightly Funnel Micro-Surgery, Bandit Traffic Allocation and Zero-Touch Conversion Self-Improvement},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 1 · The Funnel Check-up: Six-Step Beta Posteriors, Nightly Change Priorities and a Compliant Candidate Pool
- Chapter 2 · Thompson Sampling and Regret: from Two Small Samples in 1933 to the Even Split
- Chapter 3 · The Reward Function: Net Contribution per Visitor, Not the Conversion Rate
- Chapter 4 · Non-stationary Traffic: Weekday Effects, Discounted Memory and Sliding Windows
- Chapter 6 · Trustworthy Data: Negative Bias in Adaptive Data, Sample Ratio Mismatch and Guardrail Rollback
- Chapter 7 · Classical Sources East and West: "At the Limit, Change" in the Great Commentary, the Sunzi on Weak and Strong Points, and Random Allocation in Peirce and Jastrow
- Chapter 8 · The Nightly Funnel Ledger: Field Tools and Checker Specifications, a Pre-registered Test and the Conditions of Refutation