What does Hard Gates and Soft-Prompt Decay conclude, in numbers, about “The Product Account: Why Stacked Soft Layers Never Leave a Zero Residual”?
Answer: Soft reminders work in proportion; stacked, they leave a product, and with illustrative parameters four layers leave 0.004800, not zero. A hard gate sorts violations into an in-class part, which escapes only when the gate breaks and passes, and an out-of-class part, which it does not touch. Seeing zero yields only an upper bound; the guarantee of zero comes from structure. The product account, however, rests on a strong assumption: that the layers are independent. If every soft layer reads the same context and suffers the same lapse, will the holes open together, and where will the gain from stacking stop? That is the question of the next chapter.
Scope of the numbers: Soft reminders work in proportion; when layers are independent the residual violation rate is the product of their miss rates, and with illustrative parameters four layers still leave 0.0048, not zero. 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: 0.004800
base violation share v = 0.10 per save layer added miss residual per 1e6 saves prompt 0.30 0.030000 30000.0 self-check 0.50 0.015000 15000.0 peer 0.40 0.006000 6000.0 human 0.80 0.004800 4800.0 residual is exactly 3/625, never 0 layers of miss 0.30 needed to reach a target residual: target 1e-04 layers 6 target 1e-06 layers 10 target 1e-09 layers 16 a deterministic layer with miss 0 inside its class: residual 0
Key formula
Show the full program
# Chapter 1. p11: residual violation rate of stacked soft layers, exact.
# Model: an agent would commit one class of violation in a share v of
# its saves. Soft layers (a system prompt, a self-check reminder, a
# peer review, a sampled human look) each let a violation through with
# miss rate m_i. If the layers fail independently, the residual rate
# is v times the product of all m_i: it shrinks but never reaches
# zero while every m_i > 0. All rates are illustrative.
from fractions import Fraction as F
import math
V = F(1, 10)
LAYERS = [("prompt", F(3, 10)), ("self-check", F(1, 2)),
("peer", F(2, 5)), ("human", F(4, 5))]
print("base violation share v = %.2f per save" % float(V))
print("%-12s %6s %12s %14s" % ("layer added", "miss", "residual",
"per 1e6 saves"))
r = V
for name, m in LAYERS:
r = r * m
print("%-12s %6.2f %12.6f %14.1f" % (name, float(m), float(r),
float(r) * 1e6))
print("residual is exactly %s, never 0" % r)
print("layers of miss 0.30 needed to reach a target residual:")
for target in (F(1, 10 ** 4), F(1, 10 ** 6), F(1, 10 ** 9)):
k = math.ceil(math.log(float(target / V)) / math.log(0.3))
print(" target %.0e layers %2d" % (float(target), k))
print("a deterministic layer with miss 0 inside its class: residual 0")Single source and citation
Fan, Y. (2026). Hard Gates across the Client Lifecycle: the Decay of Soft-Prompt Compliance over Turns and State Machines for Interception before Publication (Version 1.0.0), Chapter 1: The Product Account: Why Stacked Soft Layers Never Leave a Zero Residual. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-256 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026256,
author = {Fan, Yafu},
title = {Hard Gates across the Client Lifecycle: the Decay of Soft-Prompt Compliance over Turns and State Machines for Interception before Publication},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 2 · Holes That Line Up: the Floor That Common-Cause Failure Puts under Stacked Soft Layers
- Chapter 3 · Decay over Turns: an Exponential Model of Soft-Prompt Compliance, Re-Sending Periods and Reading the Log
- Chapter 4 · Where to Intercept: before the Write, after It or before Publication, and Whether a Broken Gate Passes or Refuses
- Chapter 5 · Exhausting the State Machine: Proving a Dirty Publication Unreachable and Finding the Shortest Counterexamples of Three Faulty Designs
- Chapter 6 · The Price of Hardness: Pattern Width, Wrong Blocks and Gates That Get Switched Off
- Chapter 7 · Classical Sources East and West: Law and the Inked Line in the Han Feizi, and Reason's Classification of Error
- Chapter 8 · Closing the Loop: the Weekly Interception Report, the Residual Calculator and a Pre-Registered Test