What does One Commander and the Serial Bottleneck conclude, in numbers, about “Send Up Only the Exceptions: Automatic Gates, Random Audits and Coordination Overhead”?
Answer: Chapter 5 covers automatic gates: reviewing about a fifth lifts the ceiling to 21.75, and coherency overhead between windows makes capacity peak at 21.21 windows.
Scope of the numbers: With illustrative parameters, cutting the serial share from a fifth to a twentieth raises the speedup by 3.325, while raising the agents from ten to a hundred adds only 1.236. Its parameters are illustrative; all calculations are generated by its programs.
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: 21.75
| c | TPR | FPR | review | s | 1/s | slip/1k |
| -inf | 1.000 | 1.000 | 1.000 | 0.2000 | 5.00 | 0.00 |
| 0.0 | 0.977 | 0.500 | 0.524 | 0.1158 | 8.64 | 1.14 |
| 0.5 | 0.933 | 0.309 | 0.340 | 0.0783 | 12.77 | 3.34 |
| 1.0 | 0.841 | 0.159 | 0.193 | 0.0460 | 21.75 | 7.93 |
| 1.5 | 0.691 | 0.067 | 0.098 | 0.0239 | 41.80 | 15.43 |
| 2.0 | 0.500 | 0.023 | 0.047 | 0.0115 | 86.81 | 25.00 |
| 2.5 | 0.309 | 0.006 | 0.021 | 0.0053 | 188.56 | 34.57 |
| no gate, no review: s = 0, slip = 50.0 per 1000 |
Show the full program
# Chapter 5. p51: An automatic gate sends only exceptions upstairs.
# Model: a share PI = 0.05 of agent outputs is bad. An automatic gate
# scores each output; good scores ~ Normal(0, 1), bad ~ Normal(2, 1)
# (a binormal ROC with d' = 2). Outputs scoring above the threshold c
# go to the commander (r = 12 min each); the rest ship unreviewed.
# Agent work per item a = 48 min. Reviewed share f, serial share
# s = f r / (a + f r), ceiling 1/s, and bad outputs that slip through
# per 1000 shipped. c = -inf reviews everything (s = 0.20).
# All parameters are illustrative.
from statistics import NormalDist
PI, DPRIME, A, R = 0.05, 2.0, 48.0, 12.0
Z = NormalDist()
def row(c):
if c is None:
tpr, fpr = 1.0, 1.0
else:
tpr = 1.0 - Z.cdf(c - DPRIME)
fpr = 1.0 - Z.cdf(c)
f = PI * tpr + (1.0 - PI) * fpr
s = f * R / (A + f * R)
slip = PI * (1.0 - tpr) * 1000.0
return tpr, fpr, f, s, slip
print("%6s %7s %7s %8s %7s %8s %9s" % ("c", "TPR", "FPR", "review",
"s", "1/s", "slip/1k"))
for c in [None, 0.0, 0.5, 1.0, 1.5, 2.0, 2.5]:
tpr, fpr, f, s, slip = row(c)
lab = " -inf" if c is None else "%6.1f" % c
print("%6s %7.3f %7.3f %8.3f %7.4f %8.2f %9.2f" % (
lab, tpr, fpr, f, s, 1.0 / s, slip))
print("no gate, no review: s = 0, slip = %.1f per 1000" % (
PI * 1000.0))Single source and citation
Fan, Y. (2026). Constitutional Engineering for a One-Commander Silicon Empire: Serial Decision Bandwidth and Breaking Amdahl's Bottleneck (Version 1.0.0), Chapter 5: Send Up Only the Exceptions: Automatic Gates, Random Audits and Coordination Overhead. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-253 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026253,
author = {Fan, Yafu},
title = {Constitutional Engineering for a One-Commander Silicon Empire: Serial Decision Bandwidth and Breaking Amdahl's Bottleneck},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 1 · The Ceiling of the Serial Share: Amdahl's Law and the Exact Account of Twenty Percent against Five
- Chapter 2 · Does the Problem Grow? Gustafson's Law, Per-Item Review and the Measured Serial Share
- Chapter 3 · The Commander Is a Single Server: Queueing Theory, the Finite-Source Model and Little's Law
- Chapter 4 · Batch Approval Windows: Harris's Square-Root Formula and the Price of Waiting
- Chapter 6 · Decision Bandwidth: a Written Constitution as a Cache, a Faster Serial Part and the Span of Control
- Chapter 7 · Eastern and Western Sources Examined: Han Fei's Tally of Name and Form and Smith's Pin Factory
- Chapter 8 · Closing the Loop: a Weekly Decision Log, Batch Approval Windows and a Pre-Registered Test