关于「止血红绿灯:回本概率、转化滞后与预设止损额度」,《现金纪律与止血红绿灯》给出的定量结论是什么?

结论:止血红绿灯由三道闸组成:序贯边界判决,七十二小时上限收尾,预设止损额度兜底。回本概率是给人看的读数,判决仍由边界做。转化滞后会让不修正的检验把好广告的 0.967 判成红灯,按暴露修正后绿灯比例回到 0.890。在同一批合成测试上,红绿灯与一刀切的总账几乎打平,测试期间的风险资金却只有后者的五分之一左右;追亏是最糟的策略。

数据口径:北极星指标换成每日落袋净现金,即当日到账减当日真实付出再减退款准备计提;固定种子合成账本里,营收涨了 4.05 倍,落袋净现金却一路走低。 书中参数为示意,全部计算由书中程序生成。

English version

对照数据表

下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 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
查看完整验算程序
# 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))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《翻身战现金流铁血纪律、七十二小时止血红绿灯与日进五千复利重投母典》第 4 章「止血红绿灯:回本概率、转化滞后与预设止损额度」(版本 1.0.0). 5D Library. DOI 待发布

DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 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}
}

本书其他章节结论