关于「实证闭环:联营台账、试算工具与风控表」,《异业联营拆账与频次保护》给出的定量结论是什么?
结论:收入瀑布用分数核对,分项必须加总相等;契约检查器逐条判红黄绿,永远不输出新费率;风控表给出台账字段、判据、动作与日周月节奏;预注册检验需要每组约 2,193 名成员(按渠道聚类放大)与 25 场快闪。三条推翻条件事先写明。
数据口径:纳什议价解先还双方的破裂点,再平分合作红利,五五分只是破裂点对称时的特例;示意参数下,渠道方单干 2,000、操盘方单干 6,000 时,渠道方份额是 0.433。 频次方面,示意名单推 8 次比推 3 次成交多不到一倍,流失成员却是 7.453 倍;一次性操盘方的最优推送会让十二个月联合价值损失 0.1468。
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: 2193, 25
| H2: members per arm, no clustering: 2,193 | ||||
| m per channel | icc | design effect | members/arm | channels |
| 2,000 | 0.001 | 2.999 | 6,577 | 4 |
| 2,000 | 0.005 | 10.995 | 24,113 | 13 |
| 6,000 | 0.001 | 6.999 | 15,349 | 3 |
| H1: launches n = 25, reject if covered <= 19 | ||||
| size at 0.90: 0.0334; power at 0.70: 0.8065 |
查看完整验算程序
# Chapter 8. p83: sample sizes for the pre-registered tests.
# H2 (frequency): members of channels on the capped schedule (<= 3
# pushes / 72 h) have a lower 30-day unsubscribe share than members on
# a heavier schedule. Assumed shares 0.025 versus 0.040, two-sided
# alpha 0.05, power 0.80. Members cluster inside channels, so the
# member count is inflated by the design effect 1 + (m - 1) * icc.
# H1 (reach bounds): the p05..p95 band of p41 should cover the realised
# 72-hour openers in 90 % of launches. Exact binomial: smallest number
# of launches n and cut-off c such that P(X <= c | 0.90) <= 0.05 and
# P(X <= c | 0.70) >= 0.80. All inputs are planning assumptions.
import math
from statistics import NormalDist
ND = NormalDist()
def n_two_prop(p1, p2, alpha=0.05, power=0.80):
za, zb = ND.inv_cdf(1 - alpha / 2), ND.inv_cdf(power)
pbar = (p1 + p2) / 2.0
num = (za * math.sqrt(2 * pbar * (1 - pbar))
+ zb * math.sqrt(p1 * (1 - p1) + p2 * (1 - p2))) ** 2
return math.ceil(num / (p1 - p2) ** 2)
def cdf(c, n, p):
return math.fsum(math.comb(n, k) * p ** k * (1 - p) ** (n - k)
for k in range(c + 1))
def coverage_plan():
for n in range(10, 200):
for c in range(n + 1):
if cdf(c, n, 0.90) > 0.05:
c -= 1
break
if c >= 0 and cdf(c, n, 0.70) >= 0.80:
return n, c, cdf(c, n, 0.90), cdf(c, n, 0.70)
return None
if __name__ == "__main__":
n = n_two_prop(0.025, 0.040)
print("H2: members per arm, no clustering: {:,}".format(n))
print(" m per channel icc design effect members/arm channels")
for m, icc in ((2000, 0.001), (2000, 0.005), (6000, 0.001)):
de = 1 + (m - 1) * icc
tot = math.ceil(n * de)
print(" {:>13,} {:.3f} {:>13.3f} {:>11,} {:>8}".format(
m, icc, de, tot, math.ceil(tot / m)))
n1, c1, a1, b1 = coverage_plan()
print("H1: launches n = {}, reject if covered <= {}".format(n1, c1))
print(" size at 0.90: {:.4f}; power at 0.70: {:.4f}".format(a1, b1))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《异业私域联营拆账、技术换流量零本金借船与联合发售脉冲母典》第 8 章「实证闭环:联营台账、试算工具与风控表」(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-371 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026371,
author = {Fan, Yafu},
title = {Joint-Venture Audience Monetisation, Tech-for-Traffic Barter Structures and Co-Launch Revenue Waterfalls},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}