What does Bootstrap Cashflow and Kill-Switch Lights conclude, in numbers, about “The Kill-Switch Lights: Payback Probability, Conversion Lag and a Preset Stop-Loss”?

Answer: The kill-switch lights consist of three gates: the sequential boundary decides, the seventy-two-hour cap closes the grey zone, and the preset stop-loss amount catches whatever is left. The payback probability is a reading for people; the decision is still made by the boundary. Conversion lag makes an uncorrected test give red to 0.967 of good ads; corrected by exposure, the share of green lights returns to 0.890. On the same batch of synthetic tests, the lights and the blind deadline tie on the bottom line, while the money at risk during testing under the lights is only about a fifth of the other; chasing losses is the worst policy of the four.

Scope of the numbers: The north star becomes daily net cash banked (cash received minus real outlays minus the refund reserve); in a fixed-seed synthetic ledger, revenue grows by a factor of 4.05 while net cash banked keeps sinking. Parameters are illustrative; all figures come from the book's programs.

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: 0.890, 0.967

A. share of eventual orders already seen at hour t
tau h t24 t=48 t=72
60.755 0.875 0.917
180.448 0.651 0.755
360.270 0.448 0.568
B. hourly SPRT with 72 h cap, tau18 h
method true p GREEN RED cap mean hours
naive0.015 0.000 1.000 0.000 7.8
naive0.030 0.029 0.967 0.004 10.6
corrected0.015 0.030 0.947 0.024 33.4
corrected0.030 0.890 0.079 0.030 33.2
Show the full program
# Chapter 4. p42: conversion lag makes a 72-hour test kill winners.
# An order follows its click after an exponential delay with mean tau
# hours (illustrative: a buyer asks questions before paying). With
# clicks flowing evenly since time 0, the share of eventual orders seen
# by hour t is F(t) = 1 - (tau / t)(1 - exp(-t / tau)). Part A prints
# F. Part B runs the hourly SPRT of p31 in its Poisson form, LLR =
# k ln(p1 / p0) - (p1 - p0) n_eff, either naively (n_eff = clicks) or
# lag-corrected (n_eff = sum over clicks of P(delay <= age)), with the
# 72-hour cap, tau = 18 hours, 10,000 tests per row, fixed seed.
import math
import random

P0, P1, RATE, HOURS, TAU = 0.015, 0.030, 25, 72, 18.0
LA = math.log(0.90 / 0.05)
LB = math.log(0.10 / 0.95)
TRIALS = 10000


def seen(t, tau):
    return 1.0 - tau / t * (1.0 - math.exp(-t / tau))


def exposure(h, tau, corrected):
    if not corrected:
        return float(h * RATE)
    parts = []
    for j in range(h):            # cohort of clicks in hour j
        age = h - j - 0.5
        parts.append(RATE * (1.0 - math.exp(-age / tau)))
    return math.fsum(parts)


def run(p, corrected, rng, expo):
    arrivals = [0] * HOURS
    lq = math.log(1.0 - p)
    pos = 0
    while True:
        pos += int(math.log(1.0 - rng.random()) / lq) + 1
        t = (pos - 1 + rng.random()) / RATE
        if t >= HOURS:
            break
        t += rng.expovariate(1.0 / TAU)
        if t < HOURS:
            arrivals[int(t)] += 1
    k = 0
    for h in range(1, HOURS + 1):
        k += arrivals[h - 1]
        llr = k * math.log(P1 / P0) - (P1 - P0) * expo[h]
        if llr >= LA:
            return "green", h
        if llr <= LB:
            return "red", h
    return "cap", HOURS


if __name__ == "__main__":
    print("A. share of eventual orders already seen at hour t")
    print("  tau h    t=24    t=48    t=72")
    for tau in (6.0, 18.0, 36.0):
        print("  {:>5.0f}  {:>6.3f}  {:>6.3f}  {:>6.3f}".format(
            tau, seen(24, tau), seen(48, tau), seen(72, tau)))
    print("B. hourly SPRT with 72 h cap, tau = 18 h")
    print("  method     true p  GREEN   RED    cap   mean hours")
    rng = random.Random(20261120)
    for corrected in (False, True):
        expo = [0.0] + [exposure(h, TAU, corrected)
                        for h in range(1, HOURS + 1)]
        for p in (P0, P1):
            tally = {"green": 0, "red": 0, "cap": 0}
            hrs = []
            for _ in range(TRIALS):
                how, h = run(p, corrected, rng, expo)
                tally[how] += 1
                hrs.append(h)
            print("  {:<10} {:.3f}  {:.3f}  {:.3f}  {:.3f}  {:>8.1f}".format(
                "corrected" if corrected else "naive", p,
                tally["green"] / TRIALS, tally["red"] / TRIALS,
                tally["cap"] / TRIALS, math.fsum(hrs) / TRIALS))

Single source and citation

Fan, Y. (2026). Bootstrap Cashflow Discipline, Seventy-Two-Hour Kill-Switch Dashboards and Reinvestment Compounding to Five Thousand Daily (Version 1.0.0), Chapter 4: The Kill-Switch Lights: Payback Probability, Conversion Lag and a Preset Stop-Loss. 5D Library. DOI pending

DOI pending · monograph page goes live after publication · 5DL-B-372 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026372,
  author    = {Fan, Yafu},
  title     = {Bootstrap Cashflow Discipline, Seventy-Two-Hour Kill-Switch Dashboards and Reinvestment Compounding to Five Thousand Daily},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

Other chapter answers in this book