关于「东西方原典考辨:《系辞下》「穷则变」、《孙子·虚实》与皮尔士和贾斯特罗的随机分配」,《夜间自治漏斗》给出的定量结论是什么?

结论:第七章考辨《周易·系辞下》「穷则变,变则通,通则久」、《孙子·虚实》与皮尔士和贾斯特罗 1885 年的随机分配,并给出四件套蒸馏规格;

数据口径:书中参数为示意;全部计算由书中 23 段程序生成;本书是非个人化的经营分析,不构成经营、法律或财务建议,不宣称任何转化率提升的保证。

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

policychanges triedkeptharmful keptshare harmfuluplift
one263.00.010.002+0.093
uplift P5 +0.000P50 +0.082P95 +0.217
two529.20.110.012+0.238
uplift P5 +0.040P50 +0.217P95 +0.480
2x2526.00.010.002+0.195
uplift P5 +0.040P50 +0.170P95 +0.369
查看完整验算程序
# Chapter 7. p72: one change per cycle versus two changes bundled.
# Each candidate change has a true relative effect on net contribution
# per visitor: 0 (prob 0.60), +4 percent (0.25) or -4 percent (0.15).
# A cycle is one week of testing; the estimate of a variant's effect
# has standard error 3.2 percent (about 42,000 visitors per arm, see
# chapter 8). Keep the variant if estimate / SE > 1.645. Policy "one"
# tests one change per cycle; policy "two" ships two changes as one
# variant, kept or dropped together. Policy "2x2" runs the two changes
# as a factorial design: each main effect is read on all traffic with
# the same SE (no interaction assumed) and kept on its own merit.
# 26 cycles, 4,000 runs.
import math
import random

SE, ZBAR, CYCLES, REPS = 0.032, 1.645, 26, 4000
rng = random.Random(40572)


def effect():
    u = rng.random()
    return 0.0 if u < 0.60 else (0.04 if u < 0.85 else -0.04)


def run(policy):
    kept, harm, total = 0, 0, []
    for _ in range(CYCLES):
        if policy == "one":
            groups = [[effect()]]
        elif policy == "two":
            groups = [[effect(), effect()]]
        else:
            groups = [[effect()], [effect()]]
        for fx in groups:
            est = math.fsum(fx) + SE * rng.gauss(0, 1)
            if est / SE > ZBAR:
                for f in fx:
                    kept += 1
                    harm += 1 if f < 0 else 0
                    total.append(math.log1p(f))
    return kept, harm, math.expm1(math.fsum(total))


print("policy  changes tried  kept  harmful kept  share harmful  uplift")
for policy in ("one", "two", "2x2"):
    per = 1 if policy == "one" else 2
    k, h, u = 0, 0, []
    for _ in range(REPS):
        a, b, c = run(policy)
        k += a
        h += b
        u.append(c)
    u.sort()
    print("{:<7} {:>13} {:>5.1f} {:>13.2f} {:>14.3f} {:>+7.3f}".format(
        policy, CYCLES * per, k / REPS, h / REPS,
        h / k, math.fsum(u) / REPS))
    print("        uplift P5 {:+.3f}  P50 {:+.3f}  P95 {:+.3f}".format(
        u[REPS // 20], u[REPS // 2], u[19 * REPS // 20]))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《夜间自治漏斗微创调优、多臂赌博机流量自动倾斜与零人工转化率自举母典》第 7 章「东西方原典考辨:《系辞下》「穷则变」、《孙子·虚实》与皮尔士和贾斯特罗的随机分配」(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-405 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026405,
  author    = {Fan, Yafu},
  title     = {Autonomous Nightly Funnel Micro-Surgery, Bandit Traffic Allocation and Zero-Touch Conversion Self-Improvement},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论