What does Collapsing Idea-to-Delivery Latency conclude, in numbers, about “Decomposing End-to-End Latency: Four Drags and the Slowest Tenth”?
Answer: Chapter 2 splits latency into four drags, decision, start-up, feedback and hand-off, with an illustrative median of 99.9 hours and P90 of 199.8 hours, the tail driven mainly by feedback and decision.
Scope of the numbers: Parameters are illustrative and all calculations are generated by the book's programs; it offers general, not individual, business guidance, promises no launch date or income, and every production deployment is carried out only after a person confirms it. Chapter 2 splits latency into four drags, decision, start-up, feedback and hand-off, with an illustrative median of 99.9 hours and P90 of 199.8 hours, the tail driven mainly by feedback and decision.
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: 199.8, 90, 99.9
| drag | P50 | P90 | mean share | tail share |
| D decide | 18.1 | 64.2 | 0.249 | 0.255 |
| S start | 10.0 | 28.0 | 0.116 | 0.063 |
| F feedback | 35.8 | 116.7 | 0.461 | 0.602 |
| H handoff | 17.8 | 34.7 | 0.173 | 0.081 |
| end to end | 99.9 | 199.8 | (mean 118.2 hours) | |
| sum of the four medians 80.0 h; P50 of sum is 1.25 times | ||||
| P90 / P50 end to end = 2.00; ideas over 168 h: 0.162 |
Show the full program
# Chapter 2. p21: idea-to-delivery latency as the sum of four drags.
# D decision delay, S start-up and build, F wait for first outside
# feedback, H hand-off delay (two hand-offs, each its own draw). Each
# is lognormal with an illustrative median (hours) and log-sd; 20,000
# Monte Carlo draws with a fixed seed. We report P50 / P90 per drag and
# end to end, each drag's share of the mean, and its share of the time
# inside the slowest tenth of ideas (the tail that customers notice).
import math
import random
import statistics
DRAGS = [("D decide", 18.0, 1.0, 1), ("S start", 10.0, 0.8, 1),
("F feedback", 36.0, 0.9, 1), ("H handoff", 8.0, 0.7, 2)]
N = 20000
def draw(rng, drags):
parts = []
for _, med, sd, k in drags:
parts.append(math.fsum(rng.lognormvariate(math.log(med), sd)
for _ in range(k)))
return parts
def quant(xs, q):
s = sorted(xs)
return s[min(len(s) - 1, int(q * len(s)))]
if __name__ == "__main__":
rng = random.Random(42001)
rows = [draw(rng, DRAGS) for _ in range(N)]
tot = [math.fsum(r) for r in rows]
p90 = quant(tot, 0.9)
tail = [r for r, t in zip(rows, tot) if t >= p90]
mean_tot = statistics.fmean(tot)
print(" drag P50 P90 mean share tail share")
for j, (name, _, _, _) in enumerate(DRAGS):
col = [r[j] for r in rows]
tcol = math.fsum(r[j] for r in tail)
tall = math.fsum(math.fsum(r) for r in tail)
print(" {:<11} {:>6.1f} {:>6.1f} {:.3f} {:.3f}".format(
name, quant(col, 0.5), quant(col, 0.9),
statistics.fmean(col) / mean_tot, tcol / tall))
print(" end to end {:>6.1f} {:>6.1f} (mean {:.1f} hours)".format(
quant(tot, 0.5), p90, mean_tot))
sum_med = math.fsum(m * k for _, m, _, k in DRAGS)
print(" sum of the four medians {:.1f} h; P50 of sum is {:.2f} times"
.format(sum_med, quant(tot, 0.5) / sum_med))
print(" P90 / P50 end to end = {:.2f}; ideas over 168 h: {:.3f}"
.format(p90 / quant(tot, 0.5),
len([t for t in tot if t > 168.0]) / N))Single source and citation
Fan, Y. (2026). Collapsing Idea-to-Reality Latency: Feedback-Bandwidth Limits, Zero-Friction Toolchains and High-Velocity Execution Physics (Version 1.0.0), Chapter 2: Decomposing End-to-End Latency: Four Drags and the Slowest Tenth. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-420 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026420,
author = {Fan, Yafu},
title = {Collapsing Idea-to-Reality Latency: Feedback-Bandwidth Limits, Zero-Friction Toolchains and High-Velocity Execution Physics},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 1 · Loop Bandwidth: With the Delay Unchanged, More Gain Only Buys Oscillation
- Chapter 3 · Queues: Why Full Utilisation Is Not Efficiency
- Chapter 4 · Batches and Rehearsal: the U-Shaped Cost and the Smith Predictor
- Chapter 5 · Hand-offs, Switching and Hesitation: Three Losses Along the Chain
- Chapter 6 · Toolchain Friction and Gates: Why Speed Never Skips a Gate
- Chapter 7 · Classical Sources East and West: Clumsy Speed, the Ant Hole and Maxwell on Governors
- Chapter 8 · Closing the Loop: the Latency Ledger, Field Tools and Checker Specifications