What does Short-Video Traffic Pools conclude, in numbers, about “Why the Step Becomes a Ramp: Finite Batches, Logistic Promotion and the Bottleneck Gate”?
Answer: Chapter 3 shows that a hard threshold is a ramp in the data, whose width narrows from 0.124 to 0.023 as the batch grows from 100 to 3,000 viewers; the illustrative base video breaks out with probability 0.03585, with the bottleneck at the fourth gate.
Scope of the numbers: Chapter 3 shows that a hard threshold is a ramp in the data, whose width narrows from 0.124 to 0.023 as the batch grows from 100 to 3,000 viewers; the illustrative base video breaks out with probability 0.03585, with the bottleneck at the fourth gate. Marked boxes separate published evidence, runnable Python calculations (18 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: 0.023, 0.124, 100, 3000
| true c | n=100 exact/logit | n=300 exact/logit | n=3000 exact | ||
| 0.34 | 0.123 | 0.098 | 0.017 | 0.021 | 0.000 |
| 0.36 | 0.232 | 0.185 | 0.084 | 0.071 | 0.000 |
| 0.38 | 0.376 | 0.323 | 0.256 | 0.217 | 0.013 |
| 0.40 | 0.538 | 0.500 | 0.522 | 0.500 | 0.507 |
| 0.42 | 0.692 | 0.677 | 0.776 | 0.783 | 0.988 |
| 0.44 | 0.817 | 0.815 | 0.927 | 0.929 | 1.000 |
| 0.46 | 0.904 | 0.902 | 0.984 | 0.979 | 1.000 |
| n = | 100: pass 0.1 -> 0.9 between c = 0.335 and 0.459, width 0.124 | ||||
| n = | 300: pass 0.1 -> 0.9 between c = 0.363 and 0.435, width 0.072 | ||||
| n = 3000: pass 0.1 -> 0.9 between c = 0.389 and 0.412, width 0.023 |
Key formula
Show the full program
# Chapter 3. p31: a hard threshold seen through a small batch.
# Gate 1 shows a new video to n viewers and promotes it if the
# measured completion rate X / n reaches the threshold 0.40, where X
# is binomial(n, c) and c is the video's true completion rate. All
# numbers are illustrative; no platform publishes its gate rules.
# The exact pass probability is the binomial upper tail. A logistic
# curve with the same centre and the same spread, slope
# s = sqrt(3) / pi * sqrt(t (1 - t) / n), t = 0.40 the threshold,
# approximates it; the step becomes a ramp whose width shrinks
# like 1 / sqrt(n).
import math
THR = 0.40
def tail(n, c, k):
# P(X >= k), X ~ binomial(n, c), terms added with math.fsum
lp, lq = math.log(c), math.log1p(-c)
terms = [math.exp(math.lgamma(n + 1) - math.lgamma(j + 1)
- math.lgamma(n - j + 1) + j * lp + (n - j) * lq)
for j in range(k, n + 1)]
return math.fsum(terms)
def logistic(n, c):
s = math.sqrt(3.0) / math.pi * math.sqrt(THR * (1.0 - THR) / n)
return 1.0 / (1.0 + math.exp(-(c - THR) / s))
print("true c n=100 exact/logit n=300 exact/logit n=3000 exact")
for c in (0.34, 0.36, 0.38, 0.40, 0.42, 0.44, 0.46):
row = []
for n in (100, 300):
k = math.ceil(THR * n - 1e-9)
row.append("%7.3f %6.3f" % (tail(n, c, k), logistic(n, c)))
k3 = math.ceil(THR * 3000 - 1e-9)
print("%6.2f %s %s %10.3f" % (c, row[0], row[1],
tail(3000, c, k3)))
for n in (100, 300, 3000):
k = math.ceil(THR * n - 1e-9)
lo = [c / 1000.0 for c in range(250, 551)
if tail(n, c / 1000.0, k) >= 0.1][0]
hi = [c / 1000.0 for c in range(250, 551)
if tail(n, c / 1000.0, k) >= 0.9][0]
print("n = %4d: pass 0.1 -> 0.9 between c = %.3f and %.3f, width %.3f"
% (n, lo, hi, hi - lo))Single source and citation
Fan, Y. (2026). Cold-Start Dynamics of Short-Video Traffic Pools: Step Thresholds in Completion, Engagement and Replay and a Zero-Ad-Spend Breakout Test Matrix (Version 1.0.0), Chapter 3: Why the Step Becomes a Ramp: Finite Batches, Logistic Promotion and the Bottleneck Gate. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-273 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026273,
author = {Fan, Yafu},
title = {Cold-Start Dynamics of Short-Video Traffic Pools: Step Thresholds in Completion, Engagement and Replay and a Zero-Ad-Spend Breakout Test Matrix},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 1 · Promotion as a Product: Four Threshold Gates and the Nonlinear Return to Quality
- Chapter 2 · Branching Processes: Galton–Watson Cascades, Criticality and Overdispersion
- Chapter 4 · Completion, Engagement, Replay: the Statistics of Three Rulers
- Chapter 5 · More Posts or Better Posts: the Linear Ledger and the Multiplicative Ledger
- Chapter 6 · The Zero-Ad-Spend Test Matrix: a Twelve-Run Plackett–Burman Design and Its Sample Size
- Chapter 7 · Classical Sources East and West: Sunzi's "Disposition" and Galton–Watson
- Chapter 8 · Closing the Loop: the Weekly Sheet, the Promotion Calculator and a Pre-registered Refutation Plan