What does Causal Graphs, Suspense Entropy and Foreshadowing conclude, in numbers, about “The Entropy of Suspense: Bayesian Readers, Contrary Evidence and the Suspense Budget”?
Answer: Once suspense is written as entropy, a hook acquires a precise mathematical identity: it is evidence for an ending the reader did not favour, so it can raise entropy, even though evidence lowers entropy on average. The illustrative story's entropy follows a sawtooth for six chapters and then falls to 0.060 bits at the reveal. The total budget of suspense is conserved, 0.2500 in the illustrative question; saving everything for the finale yields a sum of square roots of 0.500, and giving each chapter equal suspense yields 1.500. The spoiler studies remind us that this is only one part of what drives reading. The next chapter links the sawtooth to retention and solves for the best spacing of hooks.
Scope of the numbers: Its serials, readers and parameters are illustrative models; all calculations are generated by its programs; it is not legal, financial or investment advice, and it keeps one red line: no false suspense to win clicks. Marked boxes separate published evidence, runnable Python calculations (21 programs), classical sources and the author's hypotheses.
Data table
The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python onlyprinted values that match the answer: 0.2500, 0.500, 1.500
| schedule | total | finale | sum sqrt | bits left | chapter 1 |
| hold to finale | 0.2500 | 1.000 | 0.500 | 1.000 | 0.0000 |
| front-loaded | 0.2500 | 0.190 | 0.668 | 0.286 | 0.2025 |
| flat 0.62 | 0.2500 | 0.660 | 1.227 | 0.720 | 0.0144 |
| equalised | 0.2500 | 0.111 | 1.500 | 0.141 | 0.0278 |
| equalised clue accuracy, ch 1 and ch 8: 0.667 and 0.906 |
Show the full program
# Chapter 3. p32: how to spend a fixed budget of suspense.
# Ely, Frankel and Kamenica (2015): the reader's belief that the hero
# survives is a martingale, so the expected squared belief moves over the
# whole book always add up to mu0 (1 - mu0), whatever the author does.
# Suspense in a chapter is the expected squared move it causes; a reader
# whose enjoyment is concave (here the square root) in each chapter's
# suspense prefers the budget spread out. Exact enumeration of a binary
# state, prior 0.5, eight chapters of clues with accuracy a_t and a
# finale that reveals the state. Schedules are illustrative.
import math
T, MU0 = 8, 0.5
def step(dist, a):
out, move = {}, 0.0
parts = []
for mu, w in dist.items():
for s in (1, 0):
ps = mu * a + (1 - mu) * (1 - a) if s else \
mu * (1 - a) + (1 - mu) * a
if ps <= 0:
continue
post = mu * (a if s else 1 - a) / ps
key = round(post, 12)
out[key] = out.get(key, 0.0) + w * ps
parts.append(w * ps * (post - mu) ** 2)
return out, math.fsum(parts)
def H2(m):
if m <= 0 or m >= 1:
return 0.0
return -(m * math.log2(m) + (1 - m) * math.log2(1 - m))
def evaluate(acc):
dist, sus = {MU0: 1.0}, []
for a in acc:
dist, m = step(dist, a)
sus.append(m)
sus.append(math.fsum(w * mu * (1 - mu) for mu, w in dist.items()))
left = math.fsum(w * H2(mu) for mu, w in dist.items())
return sus, left
def equalised():
dist, acc = {MU0: 1.0}, []
for t in range(T):
target = MU0 * (1 - MU0) / (T + 1)
lo, hi = 0.5, 0.999999
for _ in range(60):
mid = (lo + hi) / 2
if step(dist, mid)[1] < target:
lo = mid
else:
hi = mid
acc.append(lo)
dist = step(dist, lo)[0]
return acc
SCHED = [
("hold to finale", [0.5] * T),
("front-loaded", [0.95] + [0.5] * (T - 1)),
("flat 0.62", [0.62] * T),
("equalised", equalised()),
]
print("schedule total finale sum sqrt bits left chapter 1")
for name, acc in SCHED:
sus, left = evaluate(acc)
print("{:<14} {:.4f} {:>6.3f} {:>8.3f} {:>9.3f} {:.4f}".format(
name, math.fsum(sus), sus[-1] / math.fsum(sus),
math.fsum(math.sqrt(x) for x in sus), left, sus[0]))
acc = SCHED[-1][1]
print("equalised clue accuracy, ch 1 and ch 8: {:.3f} and {:.3f}".format(
acc[0], acc[-1]))Single source and citation
Fan, Y. (2026). Causal-DAG Validation of Long-Form Narrative, Sawtooth Scheduling of Chapter Suspense Entropy and Forgetting-Aware Foreshadowing Recovery (Version 1.0.0), Chapter 3: The Entropy of Suspense: Bayesian Readers, Contrary Evidence and the Suspense Budget. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 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}
}Other chapter answers in this book
- Chapter 1 · The Causal Graph: Kahn's Topological Sort and Version Conflicts in a Setting
- Chapter 2 · Cross-Volume Retcons: Incremental Cycle Detection, Transitive Closure and the Author's Own Forgetting
- Chapter 4 · Sawtooth Scheduling: Hook Spacing and a Markov Model of Serial Retention
- Chapter 5 · The Forgetting of Foreshadowing: Best Payoff Time, Mid-Way Reminders and a Ceiling on Open Threads
- Chapter 6 · The Survivors Who Keep Reading: Reader Heterogeneity, Survivor Bias and Censoring
- Chapter 7 · The Classics Examined: Jin Shengtan's "How to Read the Fifth Book of Genius" against Aristotle's *Poetics*
- Chapter 8 · Practice and Falsification: A Publisher's Checklist, a Persuasive-Copy Checker and a Pre-Registered Test