What does Citation Weight in Generative Search conclude, in numbers, about “The Paragraph Topology of a Bilingual Direct-Answer Page: Self-Contained Paragraphs, One Set of Numbers on Four Surfaces, and Compliant Structured Data”?
Answer: The narrative page has five paragraphs; only 0.2000 of them stand alone, none names a source, and a single lifted paragraph has an expected value of 0.4400. The inverted-pyramid page has four paragraphs; half stand alone, for an expected value of 0.6500. The page of question-and-answer pairs has three paragraphs; every one stands alone, carries a figure, names a source and opens with the answer, for an expected value of 1.0000. The lower table varies the illustrative value kept by a paragraph that cannot stand alone: even if it keeps 0.6, the narrative page reaches only 0.6800 and the inverted pyramid 0.8000, while the pairs stay at 1.0000.
Scope of the numbers: Its parameters are illustrative; all calculations are generated by its programs. It is not individual business advice.
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: 0.2000, 0.4400, 0.6, 0.6500, 0.6800, 0.8000
| topology n alone number source ans-1st value | |
| narrative | 5 0.2000 0.4000 0.0000 0.2000 0.4400 |
| inverted | 4 0.5000 0.7500 0.2500 0.2500 0.6500 |
| qa-pairs | 3 1.0000 1.0000 1.0000 1.0000 1.0000 |
| value of a lifted chunk if a dependent one keeps share d: | |
| topology d | 0.0 d=0.3 d=0.6 |
| narrative | 0.2000 0.4400 0.6800 |
| inverted | 0.5000 0.6500 0.8000 |
| qa-pairs | 1.0000 1.0000 1.0000 |
Show the full program
# Chapter 6. p61: three page topologies seen one chunk at a time.
# A retriever lifts single paragraphs out of a page. A paragraph that
# opens with a pointer back ("This", "It", "As noted above") loses its
# subject when lifted alone. We tag each paragraph by regex: does it
# stand alone, carry a number, name a source, and does its opening
# sentence state the answer? A lifted paragraph is worth 1 if it stands
# alone and 0.3 if not (illustrative). Texts are invented examples.
import math
import re
BACK = re.compile(r"^(this|that|these|those|it|they|such|as noted"
r"|as mentioned|the above)\b", re.I)
NUM = re.compile(r"\d")
SRC = re.compile(r"\b(according to|source:|reported by|per the)\b",
re.I)
ANS = re.compile(r"^[^.;]*\b(is|are|takes|lasts|costs)\b[^.;]*\d")
PAGES = {
"narrative": [
"Many shoppers ask us about returns after the holidays.",
"It depends on how the order was placed and paid.",
"As noted above, online orders follow their own rules.",
"These rules were revised in March 2026 by the store.",
"That is why the window for online orders is now 30 days.",
],
"inverted": [
"The return window for online orders is 30 days from delivery.",
"This applies to orders paid by card or wallet.",
"Refunds take 5 to 7 days, according to the policy page.",
"It was last revised in March 2026.",
],
"qa-pairs": [
"The return window for online orders is 30 days from delivery,"
" according to the store policy page revised in March 2026.",
"A card refund takes 5 to 7 days after the parcel arrives,"
" according to the same policy page.",
"Store pickup orders are 14 days, a shorter window,"
" per the pickup terms.",
],
}
print("%-10s %4s %8s %8s %8s %8s %8s" % (
"topology", "n", "alone", "number", "source", "ans-1st", "value"))
for name in ("narrative", "inverted", "qa-pairs"):
paras = PAGES[name]
alone = [not BACK.search(p) for p in paras]
vals = [1.0 if a else 0.3 for a in alone]
n = len(paras)
def share(flags):
return len([f for f in flags if f]) / n
print("%-10s %4d %8.4f %8.4f %8.4f %8.4f %8.4f" % (
name, n, share(alone), share([NUM.search(p) for p in paras]),
share([SRC.search(p) for p in paras]),
share([ANS.search(p) for p in paras]),
math.fsum(vals) / n))
print("value of a lifted chunk if a dependent one keeps share d:")
print("%-10s%9s%9s%9s" % ("topology", "d=0.0", "d=0.3", "d=0.6"))
for name in ("narrative", "inverted", "qa-pairs"):
paras = PAGES[name]
k = len([p for p in paras if not BACK.search(p)])
print("%-10s" % name + "".join(
"%9.4f" % ((k + d * (len(paras) - k)) / len(paras))
for d in (0.0, 0.3, 0.6)))Single source and citation
Fan, Y. (2026). The Geometry of Citation Weight in Generative Search: Retrieval and Reranking Features and the Paragraph Topology of Highly Cited Bilingual Answer Pages (Version 1.0.0), Chapter 6: The Paragraph Topology of a Bilingual Direct-Answer Page: Self-Contained Paragraphs, One Set of Numbers on Four Surfaces, and Compliant Structured Data. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-260 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026260,
author = {Fan, Yafu},
title = {The Geometry of Citation Weight in Generative Search: Retrieval and Reranking Features and the Paragraph Topology of Highly Cited Bilingual Answer Pages},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 1 · Measuring Visibility: Why Keyword Stuffing Fails and Why Sources, Figures and Quotations Help
- Chapter 2 · The Geometry of Hybrid Scoring: BM25 Saturation, Cosine Direction and Rank Fusion
- Chapter 3 · Paragraph Length and the Point of the Cut: Chunking, Truncation and an Interior Optimum
- Chapter 4 · The Discount by Citation Position: Cumulated Gain, Position-Adjusted Word Count and the Sag in the Middle
- Chapter 5 · Verifiability: Why Sources Raise the Chance of Citation, and the Trap of Inferring Features from Outside
- Chapter 7 · Classical Sources East and West: the Three Gauges of the Mozi and Garfield's Citation Index
- Chapter 8 · Closing the Loop: an A/B Test of Answer-Paragraph Templates, the Citability Scorer and a Pre-Registered Test