关于「实证闭环:全息推演决策台账、实战工具与检查器规格」,《非最短路径与推演深度》给出的定量结论是什么?
结论:第八章给出全息推演决策台账、蒸馏小工具「迂回铺垫盈亏平衡与二阶返工负债推演器」、四件套检查器规格与预注册检验,主检验每组需要 198 项决策。书中「全息」只指多维全覆盖,不含任何玄学含义;深度推演的时间预算包含休息,不以透支换产出。书中以标注框区分已发表的证据、可运行的 Python 计算(23 段程序)、原典与作者的假说。
数据口径:书中参数为示意,全部计算由书中程序生成。 本书不是投资意见,也不是任何具体经营决策意见,不承诺任何收益。
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: 198
| z_alpha/2 | 1.9600, z_beta = 0.8416 |
| H1 rework 0.30 vs 0.18 | 198 decisions per arm, 396 in all |
| power at 99 per arm | 0.507 |
| H2 log time, delta 0.1625 | 149 per arm |
| power at 74 per arm | 0.507 |
| H3 regret gap 6 h, sd 12 h | 63 per arm |
| power at 31 per arm | 0.503 |
| at 10 logged decisions a week, H1 needs 39.6 weeks |
查看完整验算程序
# Chapter 8. p82: sample sizes for the pre-registered tests
# H1 (rework): share of decisions with any rework inside 90 days,
# ledger-and-detour arm 0.18 against shortcut-default arm 0.30.
# Two-sided alpha 0.05, power 0.80, normal approximation for two
# proportions with pooled variance under the null.
# H2 (time): log completion time, sd 0.5; detect a 15% cut in the
# geometric mean, delta = ln(1 / 0.85).
# H3 (depth): regret in hours, sd 12, detect a 6-hour gap between
# decisions stopped at k* and decisions pushed past it.
# Also: power of each test at half the planned size. Illustrative.
import math
from statistics import NormalDist
Z = NormalDist()
ALPHA, POWER = 0.05, 0.80
za, zb = Z.inv_cdf(1 - ALPHA / 2), Z.inv_cdf(POWER)
def n_prop(p1, p2):
pb = 0.5 * (p1 + p2)
num = (za * math.sqrt(2 * pb * (1 - pb)) +
zb * math.sqrt(p1 * (1 - p1) + p2 * (1 - p2))) ** 2
return math.ceil(num / (p1 - p2) ** 2)
def n_mean(sd, delta):
return math.ceil(2 * ((za + zb) * sd / delta) ** 2)
def power_mean(sd, delta, n):
return Z.cdf(delta / (sd * math.sqrt(2 / n)) - za)
def power_prop(p1, p2, n):
pb = 0.5 * (p1 + p2)
se0 = math.sqrt(2 * pb * (1 - pb) / n)
se1 = math.sqrt((p1 * (1 - p1) + p2 * (1 - p2)) / n)
return Z.cdf((abs(p1 - p2) - za * se0) / se1)
print("z_alpha/2 = {:.4f}, z_beta = {:.4f}".format(za, zb))
n1 = n_prop(0.30, 0.18)
print("H1 rework 0.30 vs 0.18: {} decisions per arm, {} in all".format(
n1, 2 * n1))
print(" power at {} per arm: {:.3f}".format(
n1 // 2, power_prop(0.30, 0.18, n1 // 2)))
d2 = math.log(1 / 0.85)
n2 = n_mean(0.5, d2)
print("H2 log time, delta {:.4f}: {} per arm".format(d2, n2))
print(" power at {} per arm: {:.3f}".format(
n2 // 2, power_mean(0.5, d2, n2 // 2)))
n3 = n_mean(12.0, 6.0)
print("H3 regret gap 6 h, sd 12 h: {} per arm".format(n3))
print(" power at {} per arm: {:.3f}".format(
n3 // 2, power_mean(12.0, 6.0, n3 // 2)))
print("at 10 logged decisions a week, H1 needs {:.1f} weeks".format(
2 * n1 / 10))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《非最短路径最优解、二阶连锁效应全息推演与周密思考复利母典》第 8 章「实证闭环:全息推演决策台账、实战工具与检查器规格」(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-384 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026384,
author = {Fan, Yafu},
title = {When the Shortest Path Fails: Brachistochrone Detours, Second-Order Consequence Mapping and Holographic Deep Thinking},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}