关于「推演深度的最优值:后果树、正反合证据与尾部压力」,《非最短路径与推演深度》给出的定量结论是什么?
结论:第六章搭后果树,求推演深度的最优值,漏记反方证据会把 0.643 的判断看成 0.900。
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: 0.643, 0.900
A. Evidence ledger (prior odds 1:1) pro only sum log10 LR +0.9542 P(detour better) 0.900 pro + con sum log10 LR +0.2553 P(detour better) 0.643 pro + con + synthesis sum log10 LR +0.3345 P(detour better) 0.684 B. Tail stress, 50,000 draws each option mean q05 q01 ES05 shortcut 8.54 -34.06 -56.42 -48.75 detour 8.38 3.36 -20.30 -6.92
查看完整验算程序
# Chapter 6. p63: thesis, antithesis, synthesis - then the tail
# Part A: evidence for "the detour beats the shortcut" is logged as
# likelihood ratios (LR > 1 supports, LR < 1 opposes). Log10 LRs add
# (math.fsum); posterior odds = prior odds x 10^total. We show what
# dropping the opposing column does to the verdict. LRs illustrative.
# Part B: tail stress. Option payoffs (hours-equivalent) are mixtures:
# normal case with prob 1 - t, tail case with prob t. 50,000 fixed-seed
# draws; we report the mean, the 5% and 1% quantiles and the mean of
# the worst 5% (expected shortfall). Illustrative parameters.
import math
import random
PRO = [("pilot ran 18% faster", 2.0), ("tooling reused twice", 1.5),
("rework fell in 3 of 4 trials", 3.0)]
CON = [("one trial slower", 0.5), ("learning curve not counted", 0.4)]
SYN = [("gain holds only for long jobs", 1.2)]
def post(items, prior=1.0):
tot = math.fsum(math.log10(lr) for _, lr in items)
odds = prior * 10 ** tot
return tot, odds / (1 + odds)
print("A. Evidence ledger (prior odds 1:1)")
for label, items in (("pro only", PRO), ("pro + con", PRO + CON),
("pro + con + synthesis", PRO + CON + SYN)):
tot, p = post(items)
print(" {:<22} sum log10 LR {:+.4f} P(detour better) {:.3f}"
.format(label, tot, p))
rng = random.Random(38463)
OPTS = {"shortcut": (12.0, 4.0, 0.06, -45.0, 12.0),
"detour": (9.0, 3.0, 0.02, -20.0, 6.0)}
print("B. Tail stress, 50,000 draws each")
print(" option mean q05 q01 ES05")
for name, (mu, sd, t, tmu, tsd) in OPTS.items():
xs = sorted(rng.gauss(tmu, tsd) if rng.random() < t
else rng.gauss(mu, sd) for _ in range(50000))
n5 = len(xs) // 20
print(" {:<8} {:6.2f} {:6.2f} {:6.2f} {:6.2f}".format(
name, math.fsum(xs) / len(xs), xs[n5], xs[len(xs) // 100],
math.fsum(xs[:n5]) / n5))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 6 章「推演深度的最优值:后果树、正反合证据与尾部压力」(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}
}