关于「锯齿调度:钩子间距与追更留存的马尔可夫模型」,《长篇叙事的因果图、悬念熵与伏笔回收》给出的定量结论是什么?

结论:第四章在锯齿形熵、随缺口变化的流失与钩子习惯化之下求最优间距:示意参数下每三章一个钩子第 100 章留住 0.234,章章留钩只留住 0.098,换一组读者特征,一刀切的间距最多少留住 0.140。

数据口径:书中的连载、读者与参数均为示意模型;全部计算由书中程序生成;本书不是法律、财务或投资建议,并坚持一条红线:不得以虚假悬念骗取点击。 第一、二章用卡恩拓扑排序与按键分组的版本检查找出时间悖论与同一属性的两个值,并说明跨卷吃书藏在推得出的先后里:示意连载第六卷的一条边与写在第 2 卷到第 6 卷之间的 10 条边相冲突。

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: 0.098, 0.234, 100

khook bitsmean H endmean churnkept@30kept@100
10.2000.8060.02290.5040.098
20.6151.2190.01500.6250.220
31.0001.3110.01440.6290.234
41.2801.2540.01600.5940.200
51.4711.1490.01840.5440.155
61.6001.0360.02120.4950.116
71.6900.9430.02400.4580.087
81.7530.8530.02680.4100.065
91.8000.7780.02960.3890.049
101.8350.7060.03230.3460.037
best spacing over 100 chapters: every 3 chapters, kept 0.234
H at chapter ends 1 to 9: 0.75 0.56 1.42 1.07 0.80 1.60 1.20 0.90 1.67
查看完整验算程序
# Chapter 4. p41: sawtooth suspense and reader retention, hook every k.
# Open entropy H (bits) is released mid-chapter by a fraction RHO; every
# k-th chapter ends on a hook that adds h(k) = H0K k^2 / (k^2 + KAP^2)
# bits, capped at HMAX: hooks fired too close together are discounted by
# the reader (habituation). After each chapter a reader leaves with
# q = QF + A exp(-B H) + C max(0, H - HC)^2: a small floor, a term that
# falls as the information gap H grows (Loewenstein 1994) and an overload
# term when too much is left open. Retention is the product of (1 - q)
# over the chapters, a Markov chain with two states (reading, gone).
# All parameters are illustrative, chosen to show the shape.
import math

HMAX, RHO, H0K, KAP = 3.0, 0.25, 2.0, 3.0
QF, A, B, C, HC = 0.005, 0.06, 1.5, 0.03, 2.0


def hook(k):
    return H0K * k * k / (k * k + KAP * KAP)


def run(k, n=100, h=1.0, rho=RHO, kap=KAP):
    s, trace, qs = 1.0, [], []
    for t in range(1, n + 1):
        mid = h * (1 - rho)
        h = mid
        if t % k == 0:
            h = min(HMAX, mid + H0K * k * k / (k * k + kap * kap))
        q = QF + A * math.exp(-B * h) + C * max(0.0, h - HC) ** 2
        s *= 1 - q
        trace.append(h)
        qs.append(q)
        if t == 30:
            s30 = s
    return s30, s, trace, qs


print(" k  hook bits  mean H end  mean churn  kept@30  kept@100")
best = None
for k in range(1, 11):
    s30, s100, trace, qs = run(k)
    print("{:>2}  {:>9.3f}  {:>10.3f}  {:>10.4f}  {:>7.3f}  {:>8.3f}".format(
        k, hook(k), math.fsum(trace) / len(trace),
        math.fsum(qs) / len(qs), s30, s100))
    if best is None or s100 > best[1]:
        best = (k, s100)
print("best spacing over 100 chapters: every {} chapters, kept {:.3f}"
      .format(*best))
tr = run(best[0])[2]
print("H at chapter ends 1 to 9: " +
      " ".join("{:.2f}".format(x) for x in tr[:9]))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《长篇叙事因果有向无环图校验、章节悬念信息熵锯齿调度与伏笔遗忘衰减回收》第 4 章「锯齿调度:钩子间距与追更留存的马尔可夫模型」(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-178 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026178,
  author    = {Fan, Yafu},
  title     = {Causal-DAG Validation of Long-Form Narrative, Sawtooth Scheduling of Chapter Suspense Entropy and Forgetting-Aware Foreshadowing Recovery},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论