What a topic search can and cannot find

The question

The map of meaning compares chapters with one another. This companion compares topic queries with passages and asks what the resulting scores can tell us about the search.

The experiment uses Wikipedia's level-3 list of vital articles: 1,001 titles in eleven sections. Each title is compared with every indexed passage from the BSB and KJV, using the model behind this site's search box. The highest score is recorded for each title. A high score identifies a passage to examine; a low score says only that this model found no closer match for that particular query.

The experiment does not count how often a topic appears, assess its importance, or establish its absence. Even the highest-scoring passage may use a word in a different sense. The examples below show why reading the passage remains essential.

What was measured

• The list. Wikipedia's level-3 list of vital articles, 1,001 article titles in eleven sections: People; History; Geography; Arts; Everyday life; Philosophy and religion; Society and social sciences; Health, medicine and disease; Science; Technology; Mathematics. Where a title carries a qualifier in parentheses, as Mercury (planet) does, the qualifier went into the search with it.
• The search. Each title was turned into a search query by the model behind the site's search and scored against every passage the search scans: 41,444 entries of the Berean Standard Bible, four-verse windows and single verses, and the 31,102 verses of the King James Version, 72,546 entries in all. The score is the similarity of the query to the passage, from 0.315 to 0.670 on this list. The best entry is the subject's best passage, and its score is the subject's score.
• The threshold. The search box treats 0.55 as its floor: a query whose best passage scores below it gets no results at all, rather than a poor match. The page reports every subject against that line because it is the line the reader's own search will draw.
• The nearest chapter. For each subject the page also records the chapter whose language as a whole comes nearest, by comparing the query with the average of each of the 1,189 chapters.
• The word test. Whether the title itself, as a whole word or phrase, singular or plural, occurs anywhere in the text of either edition. It is a check on the threshold, not a measure of subject: Israel is named, and so are water, money and the sea, and the false friends below show that a name in the text is no promise that the best passage is about the subject.

Lower-scoring searches

The scores run from 0.315, for carbon dioxide, to 0.670, for fear. The mean is 0.447 and the median 0.438, and the fullest bin of the figure, 74 subjects, runs from 0.41 to 0.42. Against the threshold, 921 of the 1,001 subjects, 92%, fall below 0.55, and 80 reach it.

The threshold is not a test of whether a topic appears in the Bible. Even familiar terms can fall below it: Israel scores 0.536, water 0.473, money 0.496 and marriage 0.546, in the recorded experiment. So 921 is not the count of subjects the Bible does not talk about. It is the count of titles below this experiment's search threshold. The lists below rank the model's scores, not the Bible's coverage of these subjects.

The fifty lowest-scoring

The fifty lowest-scoring searches all score 0.353 or lower. They are modern science and its people (carbon dioxide, RNA, organic chemistry, the theory of relativity, Charles Darwin, Alan Turing, Carl Friedrich Gauss), modern place names (Jakarta, Indonesia, Colombia, Argentina, Mumbai), empires that rose after its last book was written (the Mughal, Mongol, Inca and Ottoman), and the machines of the last two centuries (the electric motor, the internal combustion engine, nuclear power). 12 of the fifty come from the list's Science section, 12 from People and 9 from History. The top matches can be difficult to relate to the intended topic: for carbon dioxide the model's best guess is Habakkuk 3:5, for Jakarta it is Jacob waking at Bethel in Genesis 28:16-19. The lowest of all is carbon dioxide, at 0.315.

Higher-scoring searches

The fifty highest-scoring searches all score 0.568 or higher, and they are of a piece: feelings and conditions of life (fear, sadness, anger, peace, love, happiness, poverty, famine, old age, sleep), the people and places of the text itself (Abraham, Moses, Jesus, Cyrus the Great, Egypt, the Nile), the things of a household and a field (bread, drinking water, sheep, the pig, birds, iron, the wheel), and the weather and the body (fire, wind, lightning, earthquakes, blood, bone, the eye). Where they come from is telling: 17 of the fifty are from the list's Science section, which keeps its animals, its weather and its anatomy, and 14 from Society and social sciences, which keeps its feelings and its social ills. The highest score is for fear, at 0.670, matched by Psalm 55:5, then sadness at 0.654 and Abraham at 0.652; theft, at 0.641, is matched by the commandment itself.

By section

Every section of the list has most of its subjects below the threshold, from 81% of Philosophy and religion (48 of 59) to all 40 of the Arts (40 of 40). Everyday life has the highest mean best score, 0.493, and Mathematics the lowest, 0.414, with People at 0.414 and History at 0.415 next above it. The figure shows the spread inside each section rather than one number for it, because the spread is the finding: ten of the eleven sections reach across the threshold, and the Arts section sits wholly below it, its highest subject being Fashion at 0.531.

The groups inside the sections tell it more sharply. Psychology (mean 0.524, 14 of 19 below the threshold), Food and drink (0.511) and Earth science (0.498) sit near the top; Scientists and inventors (0.402, 18 of 18 below), Modern history (0.408, 23 of 23) and Continents and regions (0.404, 56 of 57) near the bottom. The highest group of all is Family, kinship and friendship, at 0.534 across its 4 subjects, and the lowest is Explorers, at 0.358, a group of 1: Christopher Columbus. The Science section includes both high- and low-scoring queries: queries about animals, weather and anatomy are among the higher-scoring examples, while its Physics group (0.442, 44 of 46 below) is among the least.

Literal matches below the threshold

A literal word check provides a useful comparison. After removing parenthetical qualifiers and allowing the specified plural forms, 271 of the 1,001 topic labels match words or phrases in at least one edition: Israel, water, money, law, the sea. Their mean best score is 0.512, against 0.422 for the titles that never occur, though a literal match does not confirm that the intended subject is being discussed. But 200 of the 271, 74%, still fall below the threshold. Wheat, at 0.549, misses it by a hair; Israel scores 0.536 with Hosea 8:8 as its best passage; law scores 0.527. Only 71 titles with literal matches reach the threshold, while 9 without such matches do, among them Cyrus the Great, whom the text calls Cyrus, at 0.602, and skeleton, whose passages say bones, at 0.570.

This result shows that the threshold does not reliably identify even topics with literal word matches. A score below it means "no passage resembles this word closely enough for the search to answer", which is a fact about the search and the query, not about the text. The table lists the titles with literal word matches whose scores fall below the threshold.

Ambiguous matches

Fifteen examples below illustrate ambiguous or potentially misleading matches, selected by reading the returned passages. Six of the 15 score at or above the threshold. The notes compare the query's intended sense with the passage; they do not prove which word caused the model's score.

• Compass (0.588, Deuteronomy 2:3, KJV). Compassed is an older verb for going around Mount Seir in this passage, not a reference to a navigation instrument.
• Strong interaction (0.586, Isaiah 1:31, KJV). The verse and the force that binds atomic nuclei share one word, strong; the strong of Isaiah 1 are the powerful of Judah, who will burn like tinder.
• Season (0.579, Ecclesiastes 3:1). Season here means a fitting time, as the rest of the verse says; the spring, summer, autumn and winter the article describes are not its subject.
• Free will (0.565, John 8:36, KJV). The passage speaks of freedom through the Son. Whether it addresses the philosophical question intended by the search requires interpretation in context.
• Waste (0.557, Job 13:28). The article is about refuse; Job's wastes away is a body decaying like a moth-eaten garment, and the passages after it are cities laid waste.
• Liberalism (0.550, Isaiah 32:8, KJV). The liberal of the King James Version is the generous person, who devises liberal things; the political tradition the article describes is centuries away from the verse.
• Sea (0.540, 2 Chronicles 4:15, KJV). The sea of 2 Chronicles 4 is the great bronze basin of Solomon's temple, standing on twelve oxen; for the ocean, the model's best passage is a basin that borrowed its name.
• Matter (0.532, Job 32:18, KJV). Elihu is full of matter in the old sense, of things he must say; the stuff the world is made of, the article's matter, is the other sense of the word.
• Fashion (0.531, Exodus 28:13). Fashion in Exodus 28 is the verb, to make: the gold settings of the priest's vestments are being made, and the changing styles of dress the article describes are not in view.
• Smoking (0.530, Job 41:20, KJV). The passage describes smoke from Leviathan's nostrils, not a person smoking tobacco.
• Dam (0.492, Isaiah 17:1, KJV). The best result mentions Damascus; a lower-scoring result uses dam for the mother bird in Deuteronomy 22:7. Neither verse describes a structure holding back water. The precise cause of the model's ranking has not been isolated.
• Stroke (0.438, Proverbs 18:6, KJV). The strokes a fool's mouth calls for are blows; the article's stroke is a blocked or burst vessel in the brain.
• Homer (0.435, Hosea 3:2, KJV). Homer in Hosea is a measure of barley, not a reference to the Greek poet. This illustrates an ambiguous name, not proof of what caused the score.
• Mars (0.416, Luke 10:40, KJV). The best result concerns Martha of Bethany; another mentions Mars' hill, the Areopagus, in Acts 17:22. Neither discusses the planet. The role of similar names in the ranking has not been isolated.
• Mercury (planet) (0.347, Acts 14:12, KJV). Mercurius is the deity the people of Lystra associate with Paul. The shared name does not make this a passage about the planet.

The list is a judgment, not a measurement: it names the cases found by reading the best passages of every subject whose title has more than one sense or shares its spelling with a name in the text. There are surely others.

Try it yourself

What it cannot tell you

• It reflects two translations and one model. The passages come from the BSB and KJV. EmbeddingGemma was trained on multilingual data; different editions, query wording or models may produce different scores. This experiment does not isolate those effects.
• The threshold is not a test of absence. 200 topic labels with literal word matches in the text fall under it. A score below 0.55 says that the search would return nothing; it does not say the Bible is silent.
• A title is not a subject. The search saw a word or a phrase, not the article behind it. A title with more than one sense, like season or matter, is matched on whichever sense a passage happens to carry, and the false friends above show the cost.
• The ranking has not been validated as a measure of coverage. Scores range from 0.315 to 0.670, and ambiguous matches occur throughout the results. Neither a small gap nor the ordering alone establishes which topics the Bible discusses more fully.
• The list is one list. The counts belong to Wikipedia's choice of a thousand vital subjects on the day it was fetched. A different list would give different counts, and the same method.
• The scores do not rank importance. The experiment makes no judgment about a topic's religious value or a tradition's interpretation. The ambiguous-match notes are editorial readings of examples, not a consensus assessment by religious communities.

How this was made

To cite a figure elsewhere, say which measure it is (a subject's best score, its nearest chapter, or the word test) and carry the list, the two editions and the model with it. Corrections are welcome through the contact page.