What does Deferred-Promise Options and Reciprocal Stopping conclude, in numbers, about “Why the Secretary Problem's Thirty-Seven Per Cent Rule Fails Here”?
Answer: The secretary problem's thirty-seven per cent rule depends on six assumptions, and the situations of this book violate almost all of them. Once the payoff is made continuous, observation is given a cost, recall is allowed or a trend is introduced, the optimal share to pass over ranges from 0.05 to 0.60. A better-suited replacement is a reservation threshold set by the cost of observation: the higher the cost, the lower the threshold and the earlier one stops.
Scope of the numbers: Parameters are illustrative; all calculations are generated by the book's programs; the book is not counselling, relationship advice or clinical judgement, and makes no divinatory inference. Marked boxes separate evidence, runnable 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: 0.05, 0.60
| case | best k | share | payoff | payoff at k=7 | loss |
| classic-like | 4 | 0.20 | 0.801 | 0.767 | 0.034 |
| cost 0.01 | 2 | 0.10 | 0.715 | 0.620 | 0.095 |
| cost 0.03 | 1 | 0.05 | 0.583 | 0.327 | 0.256 |
| recall q=0.5 | 5 | 0.25 | 0.851 | 0.847 | 0.004 |
| declining pool | 1 | 0.05 | 0.575 | 0.213 | 0.362 |
| rising pool | 12 | 0.60 | 1.134 | 1.073 | 0.061 |
Show the full program
# Chapter 5. p52: the 37 percent cutoff once its assumptions break
# n = 20 options seen one by one, value U(0, 1) plus a trend. Payoff =
# value of the option kept minus c per option inspected (cardinal, not
# best-only). After passing the first k, take the first best-so-far;
# if none comes, recall the best passed option with probability q
# (it may have moved on), else keep the last one. Common random draws
# for every k; the best k is compared with the 1/e cutoff k = 7.
import math
import random
N, SIMS = 20, 4000
rng = random.Random(42352)
base = [[rng.random() for _ in range(N)] for _ in range(SIMS)]
gate = [rng.random() for _ in range(SIMS)]
def run(k, c, q, trend):
tot = []
for s in range(SIMS):
row = [base[s][i] + trend * i / (N - 1) for i in range(N)]
best = max(row[:k]) if k > 0 else -1.0e9
pick, seen = None, N
for i in range(k, N):
if row[i] > best:
pick, seen = row[i], i + 1
break
if pick is None:
pick = best if (k > 0 and gate[s] < q) else row[N - 1]
tot.append(pick - c * seen)
return math.fsum(tot) / SIMS
CASES = [("classic-like", 0.00, 0.0, 0.0),
("cost 0.01", 0.01, 0.0, 0.0),
("cost 0.03", 0.03, 0.0, 0.0),
("recall q=0.5", 0.00, 0.5, 0.0),
("declining pool", 0.00, 0.0, -0.4),
("rising pool", 0.00, 0.0, 0.4)]
print("case best k share payoff payoff at k=7 loss")
for name, c, q, tr in CASES:
vals = [run(k, c, q, tr) for k in range(N)]
k_best = max(range(N), key=lambda k: vals[k])
print("{:16s} {:7d} {:6.2f} {:8.3f} {:14.3f} {:6.3f}".format(
name, k_best, k_best / N, vals[k_best], vals[7],
vals[k_best] - vals[7]))Single source and citation
Fan, Y. (2026). Unbundling Paternal Selection from Romantic Commitment: Deferred-Promise Options, Avoidant Autonomy and Reciprocal Stopping Rules (Version 1.0.0), Chapter 5: Why the Secretary Problem's Thirty-Seven Per Cent Rule Fails Here. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-423 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026423,
author = {Fan, Yafu},
title = {Unbundling Paternal Selection from Romantic Commitment: Deferred-Promise Options, Avoidant Autonomy and Reciprocal Stopping Rules},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 1 · A Deferred Promise Is an Option: Who Holds It, Who Writes It, Who Was Paid the Premium
- Chapter 2 · Costly Signals and Cheap Talk: How Much Information Is Left in a Costless Title
- Chapter 3 · Base Rates and Dimensions: The Error Rate of Labelling from a Single Behaviour
- Chapter 4 · The Sequential Probability Ratio Test: How Many Weeks of Records Are Enough to Stop
- Chapter 6 · Sunk Costs and the Investment Model: Why Stopping Gets Harder the More One Has Invested
- Chapter 7 · Classical Sources: the Distant Oath of "Mang", You Ruo on Words That Can Be Made Good, and Thales's Olive Presses
- Chapter 8 · Reciprocal Floor Stopping: The Empirical Loop, Field Tools and Checker Specifications