What was preached

All churches › About the study

About the study

Collect what churches actually preach on Sunday, and hold it up against the Bible.

Two questions, answered with evidence rather than opinion:

  1. What does a given church actually preach on a Sunday?
  2. How does that sermon relate to the Bible text it claims to be preaching?

The tool measures and describes. It does not rule on orthodoxy.

What it reports instead

It reports what the sermon claimed, what the verse says in context, and where they diverge, and leaves the reader to judge. Five columns of evidence sit side by side on a church page — scripture, the church's own statement of faith, the tradition it names, what was preached, and how that stands to propositions the historic creeds condemned. Nothing here composes them into a score, and no figure on any page of this app is a verdict about a congregation.

The three measurements

citation, coverage, allusion

Citation, coverage and allusion — what each one counts — and the translation everything is compared against.

  • Citation — the references a preacher names out loud. Spoken forms are parsed as spoken (Acts 5 12-16, Romans chapter 8 verse 28, Ephesians two), context is carried forward for a bare verse 29, and any reference that does not exist in the text is re-anchored or dropped rather than reported.
  • Coverage — when a sermon names its passage, usually in the title, how much of that passage it actually works through.
  • Allusion — scripture echoed without being named, found by searching every sermon window against all 31,102 verses of the translation with a distinctive-word overlap filter. An allusion is a candidate echo, not a claim that the preacher had the verse in mind.

The Berean Standard Bible is the translation everything is compared against, because it is modern English and CC0, so its text can be quoted here without a licence problem.

What the preaching rests on

the nine kinds

The nine kinds of reason a sermon can point to, how steady the counts are between two readings, and which kinds are not shown.

Wherever a preacher gives a reason for what they say, the reading notes what the reason points to, in one of nine kinds, and keeps the words. No page orders churches by these counts or sets them side by side.

  • The Bible
  • A creed or the church’s teaching — not shown: checked against a second, blind reading by another model (Fable), which put too few under the same kind to show it for any church (19 of 30 matched)
  • A Christian teacher or writer
  • A writing presented as prophetic — not shown: checked against a second, blind reading by another model (Fable), which put too few under the same kind to show it for any church (14 of 30 matched)
  • Something God said to the preacher
  • Something that happened
  • Outside knowledge: a study, a figure, history
  • Everyday wisdom and culture — not shown: two readings of the same sermons differed too much to compare one church with another
  • Something else — not shown: two readings of the same sermons differed too much to compare one church with another

676 sermons were read a second time, from 132 sources (78 churches, 23 reference figures and 31 historical preachers): three from each of 102, and from 1 to 29 from each of the other 30. The readings found the same reason 70% of the time, and then put it under the same kind 98% of the time. For stories they found the same passage 47% of the time, so a story count is given and no story is quoted.

Steadiness by kind
same kindchangespread
The Bible99%0.431.57shown
A creed or the church’s teaching88%0.090.16not shown (bar 3, 4)
A Christian teacher or writer91%0.110.33shown
A writing presented as prophetic90%0.000.00not shown (bar 2, 3, 4)
Something God said to the preacher97%0.040.07shown (bar 3)
Something that happened93%0.260.36shown
Outside knowledge: a study, a figure, history87%0.140.27shown
Everyday wisdom and culture91%0.320.35not shown (bar 3)
Something else56%0.090.04not shown (bar 1, 2, 3)

Same kind: of the reasons both readings found. Change: how far a church’s rate per 1,000 words moved between readings of the same three sermons (the 90th percentile over churches). Spread: the middle half of the churches’ rates.

A kind is shown when four things hold: (1) the two readings put the same reason under it at least 80% of the time; (2) its change between readings is below the spread between churches; (3) its estimated change over a church’s 24 sermons is at most a quarter of that spread; (4) it was checked blind on real instances, without the church, the title or the tradition: six kinds in the first study, and the other three by a second model (Fable), whose reading had to match on at least 80% of them.

What this corpus holds

churches, uploads, transcripts, window

How many churches, uploads, transcripts and classified recordings this corpus holds, and the window they were published in.

churches and reference figures in this corpus
81 churches · 24 reference figures
uploads synced from their channels
14,589
of those 14,589, transcripts on disk
6,652
of those 6,652, recordings classified
5,969
window of the congregations’ 4,199 classified recordings
published 2025-08-22 to 2026-08-21; each reference figure is read over dates of its own, named on its page
of those 6,652, indexed for scripture citations
6,577
scripture references parsed from those 6,577
105,294
central publication held beside them, in none of the figures above
1 — 52 written study articles and 2,483 citations read from the verse codes their publisher attaches to each reference. A central publication is not a church and none of its articles is a recording; it is listed with the traditions outside the Nicene consensus.
historical preachers held beside them, in none of the figures above
31 historical preachers and 724 printed sermons, read on their own page

75 of the 6,652 transcripts have not yet been indexed for scripture citations, so the references figure says nothing about them: not yet indexed is not “cited nothing”.

The corpus is what could be transcribed, not what was preached. Of the 6,048 sermon uploads published inside the dates each church and preacher is read over, 5,973 have a transcript. 7,818 of the 7,937 uploads without one were published outside those dates. Every figure on a church page is over that church’s own classified recordings and over nothing else.

How a recording becomes a figure

every stage

Every stage from a channel to a figure on a page, the command that runs it, and the count it ran over.

Every figure on a church page is the end of one chain, and each link is a command you can run.

How the counts are read

A recording can stop at any of them, which is why a church page says which stage it stopped at rather than printing a blank. The counts beside each stage are read out of this corpus now, each over the population of the stage above it.

  1. A registry entry. A church is configuration, never code: a row in the project’s church registry naming its channel, and saying how to read what it uploads.

    14,589 uploads are held, 14,404 of them classified as sermons; 6,891 of the 14,589 are whole services rather than sermons. A further 52 written study articles of a central publication are held beside these and counted in no stage on this list: none of them is a recording, their citations come from the publisher's verse codes, and whatever interpretive passes have run over them are reported only in the separate section outside the Nicene consensus.

    Why and how

    Two of those settings decide what may be measured at all. kind separates sermons from everything else a channel publishes — one church's Ep NN uploads are a companion podcast discussing the sermon afterwards, and counting those as preaching would measure upload conventions. scope says whether an upload is a sermon or a whole service.

  2. sync lists the channel.

    14,589 uploads listed across every channel in this corpus.

    Why and how

    Both tabs, because a congregation that livestreams its service publishes to /streams and the two tabs often do not intersect at all: a channel whose /videos ends years ago may be streaming full worship every Sunday. Each entry is tagged with the tab it came from, and a stream is normally the whole service, so a church whose material arrives that way needs a service scope.

  3. fetch downloads captions, four seconds apart, backing off when it is refused and leaving the rest pending so it can resume.

    of those 14,589 uploads, 6,652 have a transcript on disk, 186 of them from captions.

    Why and how

    Transcripts on disk are the source of truth; the database can be rebuilt from them without touching YouTube again.

  4. asr transcribes the audio locally when captions are off.

    of the 6,652 transcripts on disk, 6,351 were transcribed from audio here and 115 came from a church's own published text. 41 more are held and marked unusable rather than read.

    Why and how

    Roughly a third of some channels' uploads have captions disabled, so a caption-only corpus would be shaped by what YouTube happens to caption rather than by what churches preach — and that bias would land in every comparison invisibly. A church that publishes its sermon as prose is read from its own page instead, which is the one source carrying no transcription error at all.

  5. analyze finds the references a preacher names out loud: the citation measurement above.

    6,536 of the 6,536 transcribed sermons have been read for citations.

    Why and how

    A transcript it has read is marked read even when it found no reference at all, so a sermon that cites nothing is never mistaken for one that was not looked at.

  6. segment and segment-llm find the teaching span inside a recording.

    of the 6,652 transcripts on disk, 6,051 carry a labelled span of any kind, and 6,048 teaching spans were resolved.

    Why and how

    A Sunday upload is not a sermon, it is a service, and the readings, music, announcements and altar call in it are not preaching. A figure computed over the whole recording counts them in both its numerator and its denominator, and is recorded under the basis unsegmented so it can never be read as a measurement of the preaching. A phrase list resolves the churches whose boundaries recur word for word; a model reads the rest, which is every church a phrase list cannot reach.

  7. Quarantine decides whether a boundary is good enough to use.

    of the 6,652 transcripts on disk, 529 are quarantined and 8 are queued for a human to set the boundary.

    Why and how

    Below a stated confidence the recording is never analysed on a guess: it is quarantined where the church keeps enough confidently segmented recordings to spare it, and queued for a human to watch where it does not. Excluding every unresolved recording outright was rejected, because that would bias the corpus toward churches with tidy upload conventions — structurally the same error as collecting only what YouTube happens to caption.

  8. interpret reads each recording against eleven components of the apostolic gospel and records, for each, which of five treatments the sermon gives it, with verbatim evidence that is verified back into the transcript before it is stored.

    5,969 recordings classified of the 6,652 with a transcript on disk, 114,239 rows in all (and the rows of 1 more recording the corpus no longer holds): basis teaching 65,006, unsegmented 28,929, altar_call 20,304.

    A second Claude model (Fable) read 40 of these sermons blind and gave the same treatment as this pass on 381 of the 440 parts, α = 0.82 beyond chance (95% interval 0.78–0.86, resampling recordings), where 1 is full agreement and 0 is chance, lowest on “New life, the Spirit, discipleship” (0.54), “Witnesses to it” (0.59) and “Who God is — creator, holy” (0.62); both readers are Claude models, so this is not a check by a person.

    Why and how

    Every row carries the basis it was read on, so a figure can never be averaged across two different texts. The eleven are anchored on the apostles’ own preaching — nine of them in the speeches of Acts, the rest in 1 Corinthians 15, Romans, John and 1 Peter — and the speeches in Acts are shown whole and read with the same instruments. That is the point of view the measurement is taken from, declared rather than hidden, and giving it is not a verdict on any church.

  9. substitution asks what a reframe put in place of the thing.

    20,910 labelled pairs across 5,765 recordings, out of 35,258 reframes there were to label; 6,060 of the 6,072 pairs labelled within or narrowed have their second reading.

    Why and how

    A reframed label says a component is present in other vocabulary and then stops, and a reframe count on its own measures nothing: reframing is what preaching from an assigned text IS, and the churches that reframe most in this corpus are not the ones a reader would expect. So each reframe gets one bounded label and the two objects the label rests on, judged against the same anchor verse. Within and narrowed are the two labels nearest the verse, and a page shows either only where a second, independent reading gave the same one.

  10. reframe-flags adds two facts to each reframe pair, with no model: whether the recording names the anchor verse aloud, and whether the sermon’s own words there are seven or more words in a row of a Bible verse. Both are descriptions a reader can check, not marks.

    20,874 of the 20,910 current reframe pairs carry the two facts.

  11. scripture-share counts the teaching words that are Scripture read aloud or quoted: words inside seven or more in a row that also stand, in that order, in a verse of a translation on file. No model.

    5,946 of the 6,536 transcribed sermons have a teaching span long enough to count; every other one is held with the reason it is not counted.

    Why and how

    It is a floor, since a reading from a translation not on file is partly missed, and it says nothing about whether a sermon is about its text.

  12. condemned reads a recording against propositions named councils and confessions rejected, one proposition at a time, and records affirms, denies or not addressed with the sermon's own words.

    1,585 recordings read against 115 propositions, 277 of those readings claiming an affirms. The reframe-only track has read 5,786 recordings and claimed 419 affirms; the two tracks rest on different amounts of text and are never pooled.

    Why and how

    Each proposition carries the body that condemned it, the canon and the date, and no count ever crosses an axis: a finding about how a claim is grounded says nothing about the Trinity. “Rejected” means a named council or confession rejected the position; “matched” means a reading found it in the sermon’s own words, shown in the quote.

  13. condemned-review is the gate every affirms has to clear. Three independent passes are asked to refute the claim — not to confirm it — and one withdrawal sinks it.

    2,112 refutation passes recorded: 100 upheld the claim they read and 2,012 withdrew it. A claim is published on three upholding passes and no withdrawal at all.

    Why and how

    It exists because of what was found without it: of the 43 affirms standing across the panel when the gate was built, one survived it. A refuter has to find the reading on which the sentence is ordinary orthodox teaching, and failing to find one is the evidence.

  14. own-doctrine sets the preaching of a church outside the Nicene consensus beside its own tradition’s doctrine as well as beside the historic councils.

    3 congregations in this corpus are outside the consensus.

    Why and how

    Its page lays each measured proposition out in four fields: what its tradition teaches, what its own statement says, what its preaching was measured to say, and what the councils said. Every other church’s page carries the same rows cut to its own statement and its preaching, so none is singled out. A further pass, reading these churches’ recordings against their tradition’s own sentences, is built and has not been run.

  15. appeals is the authority measurement: what the preaching offers as a reason. It is being run over a spread sample of each church’s recordings and is not yet published; no page shows a figure from it.
  16. ritual and apostolic-core ask two questions of every sermon, in the preacher’s own words: does it say that Jesus died, that he rose, and call the hearers to repent or believe; and does its last fifth call them to respond to Jesus.
    Why and how

    ritual first sets aside, with no model, the text a church recites week after week — eight words recurring in a fifth of its messages, such as a weekly prayer, a creed or a sign-off — and Scripture read aloud never counts either. Both are built and gated, measure by measure: on a sample of sermons from six churches each question was asked twice and set against a blind second model reading (Fable) — checked against a second model reading, not by a person. The core message, how it ends and the altar call cleared that check. A fourth measure, how many of the eleven parts a sermon’s most focused 250 words state, was tried twice and could not be measured reliably, so it is not shown.

A recording that stopped at any stage is still held, and its church's page names the stage rather than reporting a zero.

Every affirms found in the reframe quotes alone goes through the same gate as one found in a whole transcript: three independent refutation passes, every one of which must uphold it.

How to read what these pages show

glossary

Every word a church page prints, what it does not mean, and a worked example read out of this corpus.

A church page and its four sections print the measurement's own vocabulary, and several of those words mean something narrower than they look. Every one of them is below, with what it does not mean beside what it does, and with a worked example read out of this corpus — the church and the recording named, so any of them can be checked. Where the corpus holds no real case for a word, it says so rather than inventing one.

No example here is a verdict about a church. Each was chosen for how plainly it shows the word, one church per entry and never two side by side; where the clearest case in the corpus would have read as an accusation, a duller one was taken instead. Every quotation is verbatim from the recording named beside it, and every example was read out of this corpus on 2026-09-28.

What a sermon does with scripture

3 terms: citation, coverage, allusion
citation

A reference the preacher names out loud. Spoken forms are parsed as spoken, and a reference that does not exist in the text is re-anchored or dropped rather than reported.

Scholars who index how early Christian writers used Scripture tell quotation from allusion by reading. Here three rules do it: named aloud (citation), read word for word (Scripture read aloud or quoted), echoed in other words (allusion, a candidate). The lines are not the same as theirs.

What it does not mean. It is not a measure of how much scripture a sermon used. A preacher can expound a verse for ten minutes after naming it once, and a preacher can name twelve references in a list. Both are counted the same way here.

From this corpus. This one recording has 32 parsed references. The raw spoken forms include “Romans 6, 1 through 4”, “Romans chapter 6 verse 1” and a bare “Verse 3” that only means anything because the chapter above it was carried forward. One of the 32, the reading of Romans 6:1-4 five seconds in, is marked as belonging to the lay reading rather than to the preaching. vccc recording #77, “Romans 6:1-4 | New In Christ” (2025-11-10)

Romans 6, 1 through 4 ▸ 0:04 not inside the text the claim was read from; linked at its first occurrence.

transcript 0:00–0:37 · Whisper

Please remain standing for the reading of the word. My name is [speaker] and I'll be reading from Romans 6, 1 through 4. What shall we say then? Are we to continue in sin that grace may abound? By no means. How can we who died to sin still live in it? Do you not know that all of us who have been baptized into Christ Jesus were baptized into his death? We were buried therefore with him by baptism into death. In order that just as Christ was raised from the dead by the glory of the Father, we too might walk in newness of life.

What do the ▸ links mean?

Receipts: ▸ opens the recording at the moment the words begin; a note follows only when there is no single exact moment. Transcript sources: Whisper = our Whisper transcription of the audio; auto captions = YouTube’s automatic captions; channel captions = captions the channel uploaded to YouTube. How receipts work →

Romans chapter 6 verse 1 ▸ 13:04

transcript 12:31–13:36 · Whisper

But it also doesn't just add heaven onto the end of my life. No, no, no. It's way more full than that. It completely redefines who we are now. We are a new spiritual person. We have a new identity. I'm a new creation in Christ. Not a little bit new. I am new. That's what it says. Because I'm in Christ. Romans 6 explains this really well. And we're actually going to spend the rest of our time in here. So if you have your Bibles open to Romans 6. I'm going to give you one second. Open Romans chapter 6 verse 1. And this was the scripture reading at the beginning of the service that everybody was still talking when Noah started reading. But just kidding. Romans 6 verse 1. You there? Say amen. If you're not there, say hold on. Okay, I got one. Hold on. There we go. I'll give you. Romans 6 verse 1. Here's what he says. What shall we say then? Are we to continue in sin that grace may abound?

Verse 3 ▸ 15:00 Said 2 times in this recording; linked at the first time inside the text the claim was read from. Also at ▸ 15:01.

transcript 14:28–15:31 · Whisper

Again, what does Paul say? No! That is ridiculous, horrible thinking. He says, by no means. He says, we died to it, so we can't keep living in it. He's saying it's either death or it's life. You can't have it both ways. You can't both, you're spiritually, and you can't be alive to both and dead to both. It's either you're dead to one and alive to other, or vice versa. What is he saying? Verse 3. Romans 6, verse 3. Do you not know that all of us who have been baptized into Christ Jesus were baptized into his death? We were buried, we were buried, therefore, with him by baptism into death, in order that just as Christ was raised from the dead by the glory of the Father, we too might walk in newness of life. Death is a real thing. Death is a very final thing. If you haven't been to a funeral recently, you should go to a funeral.

coverage

When a sermon names its passage — usually in the video title — how much of that passage it works through. Counted on the teaching span only: reading a passage aloud in the opening liturgy is not working through it.

What it does not mean. It is not a measure of exposition, because a verse expounded at length without being named again counts as untouched. It is the share of the declared passage's verses that the preaching refers to.

From this corpus. The title declares Romans 6:1-4. On the teaching span, verses 1 and 3 are named and verses 2 and 4 are not, so coverage is 50.0% of four verses. Verse 4 WAS read aloud — in the lay reading, before the sermon began — and does not count, which is exactly what the teaching-only rule means. vccc recording #77, “Romans 6:1-4 | New In Christ” (2025-11-10)

allusion

Scripture echoed without being named, found by searching every window of the sermon against all of the translation's verses and keeping only matches that share distinctive words.

What it does not mean. An allusion is a candidate echo and never a claim about what the preacher had in mind. The search matches words, and two verses can share them.

From this corpus. 30 candidate echoes, five of them never named aloud. One of the five is 1 Corinthians 15:57, “But thanks be to God, who gives us the victory through our Lord Jesus Christ!” — matched against the sermon's “Thanks be to God who delivers me through Jesus Christ”. The words being read aloud at that moment were Romans 7:25. The instrument found a real verbal echo and the wrong verse, which is why an allusion is offered as a candidate. vccc recording #77, “Romans 6:1-4 | New In Christ” (2025-11-10)

Thanks be to God who delivers me through Jesus Christ ▸ 10:12

transcript 9:39–10:42 · Whisper

But I see another law at work in me, waging war against the law of my mind and making me a prisoner of the law of sin at work within me. Oh, wretched man that I am, who will rescue me from this body that is subject to death? What's Paul saying? There is a battle going on, for sure, right? I'm not diminishing or saying there's not a battle. But we still wonder who's going to win. But for the person who's in Christ, do we have to wonder? We don't have to wonder why. Because verse 25 says, Thanks be to God who delivers me through Jesus Christ our Lord. Jesus delivers us, right? He saves us. He frees us. How does he do this? Look at Galatians 2.20. If you don't have this memorized, you should. It says, It is no longer I who lives, but Christ who lives in me, right? And this life I now live in the flesh, I live by faith in the Son of God who loved me and gave himself for me.

What one recording did with a gospel component

6 terms: taught, in_passing, reframed, formulaic, absent, unverifiable
taught

The component is defined, explained or applied in the body of the message — worked out, not merely named. It is the taxonomy's own word, and the narrowest of the five.

From this corpus. On one recording, death_for_sins is taught: “He dealt with the consequences of my sin, the penalty for my sin, but he also dealt with the root cause of our sin that got us into the problem to begin with.” vccc recording #77, “Romans 6:1-4 | New In Christ” (2025-11-10)

He dealt with the consequences of my sin, the penalty for my sin, but he also dealt with the root cause of our sin that got us into the problem to begin with. ▸ 17:46

transcript 17:15–18:26 · Whisper

is exactly what needed to happen to our sin. Our old self wasn't just a little sick and it needed to be fixed, needed some Tylenol, needed to be cleaned up. It just didn't need a slap on the wrist or a timeout. It had to be killed. It had to be crucified. How? With him. With Christ. Now there's something I want to point out here. Jesus didn't just die so we don't have to die. There's more to his death than that. Okay? He dealt with the consequences of my sin, the penalty for my sin, but he also dealt with the root cause of our sin that got us into the problem to begin with. Okay? He says, Psalm 139 verse 23, Search me, O God, and know my what? Heart. Try me and know my thoughts. See if there be any grievous way in me and lead me in the way everlasting. What was our problem? Our hearts. Right? Jeremiah 17, 9. The heart is deceitful above all things. Desperately sick. Who can understand it? See, the root of our problem was our hearts. And so what did God do? We have a,

in_passing

Named but not developed. One mention, a clause inside a list, or a single summary sentence. What separates it from taught is whether the sermon develops it — not who said it, and not whether anything was being quoted.

From this corpus. On the same recording, burial is in passing: “Our old self died in a tomb three days, dead with Christ.” The other quote behind that row is the sermon reading Romans 6:4 aloud. Named twice, developed neither time. vccc recording #77, “Romans 6:1-4 | New In Christ” (2025-11-10)

Our old self died in a tomb three days, dead with Christ. ▸ 17:01

transcript 16:29–17:35 · Whisper

Not like zombies at Halloween, kind of half dead, half alive. No, dead, dead. Jesus didn't like negotiate with our old self and he's still like trying to work out the terms of the agreement and stuff. It's like dead, dead with Christ. Jesus didn't just scream at our old self and say, if you don't stop it right now, right? Kind of like Marty McFly in the first back. You take your hands off her, right? No. Like he didn't just count to three at our old self. He crucified it. Our old self died in a tomb three days, dead with Christ. Dead, dead. That's our old self. That's our old sinful desires, our flesh. Okay? This, and precisely this, is exactly what needed to happen to our sin. Our old self wasn't just a little sick and it needed to be fixed, needed some Tylenol, needed to be cleaned up. It just didn't need a slap on the wrist or a timeout. It had to be killed. It had to be crucified. How? With him. With Christ.

reframed

The single most misreadable word on this site. It means one thing: the component is present, and carried by the sermon's own vocabulary and register rather than the anchor verse's. Because the label can be unfair to a preacher, the classifier must also record, in words, what that vocabulary was — so a reader can set it beside the verse and judge.

What it does not mean. It does not mean wrong, watered down, evasive or unfaithful. It is not a milder absent, and it is not a failure to preach the component. Whether a reframe counts as preaching a component is the reader's call, and this tool does not make it — which is why all five treatments are printed side by side for every component and none of them is summed away. It is also why a reframe count on its own is not a finding: taking an assigned text and applying it to the people in front of you is what preaching IS, so reframing is ordinary, and the churches that do it most in this corpus are not the ones the word would lead a reader to expect.

From this corpus. burial against 1 Corinthians 15:4. The verse's object is “that He was buried”; the sermon's is “the grave couldn't hold Him”, and the recorded vocabulary is that the sermon speaks of Christ rising from the grave rather than stating that He was buried. The substitution pass then labelled that pair within — the sermon's object stays inside what the verse supports. A reframe, and nothing missing. gracechurch recording #710, “Presenting Every Man Perfect, Part 1” (2026-01-12)

the grave couldn't hold Him ▸ 58:42

transcript 58:11–59:16 · Whisper

Him, had given Him from eternity past. Everyone who would place their faith in Him, their sins were laid upon the shoulders of the substitute. And He was forsaken that we might be received. He was abandoned that we might be welcomed. He was punished that we might be justly forgiven. And then He was raised from the grave. He Himself had the authority to take up His life again and rise from the grave on the third day in victory and conquest over sin and death, testifying that the grave couldn't hold Him, that all stood accomplished. And He ascended into heaven and now calls everyone within the sound of His voice through this word, through His preachers, to come to Him and find forgiveness and righteousness and rest. By laying down all of the quote-unquote good works you would seek to offer Him, to please Him, to earn your way to heaven, He says, put all of those away. Come to Me, receive My righteousness, receive My forgiveness through

formulaic

The component occurs only inside text that recurs near-verbatim across this church's other sermons — a fixed altar-call script or a weekly recited confession. Decided mechanically, by comparing this recording's passages against the church's own others, before any model sees it.

What it does not mean. It does not mean insincere or empty. It means this sermon carried the component in words the church says every week, so this recording is not evidence that this sermon developed it. A church may recite something weekly and also teach it; the other four treatments on the same row are where that would show.

From this corpus. burial rests only on the recited line “I believe he died that I would be forgiven, was buried, and rose again to give me life.” The same sentence appears, near word for word, across this church's other recordings. elevation recording #590, “It's Worth The Dirt” (2026-01-25)

I believe he died that I would be forgiven, was buried, and rose again to give me life. ▸ 58:21 not inside the text the claim was read from; linked at its first occurrence.

transcript 57:51–59:02 · Whisper

I'm going to pray a prayer right now. If you want to give your life to Jesus, repeat this prayer after me. We're going to pray it out loud in the room for the benefit of those all over the world who are coming to Christ in this moment. Repeat after me. Heavenly Father, I am a sinner in need of a Savior, and I believe that Jesus Christ is the Son of God and the Savior of the world. Today, I make Jesus the Lord of my life. I believe he died that I would be forgiven, was buried, and rose again to give me life. I receive your life. Here's mine. I am a child of God. I am a new creation. On the count of three, if you prayed that prayer, shoot your hand in the air. One, two, three. Or if you're watching online, just put it in the chat. Say, I receive Jesus. Come on, let's thank God all over the world, all over the globe. Everybody who

absent

The component's subject does not appear anywhere in the recording, directly or conceptually. This is a claim about a negative, so the classifier must publish the terms and concepts it searched for, and those terms are matched back against the transcript before the claim is allowed to stand.

What it does not mean. It is not a claim about the church. It is a claim about one recording, over the text named in that row's basis. A component absent from a twelve-minute homily on one Sunday is an ordinary fact about a sermon, and the recurrence figure beside it — present in how many of how many — is the only place a pattern could appear.

From this corpus. burial is recorded absent on the teaching span, with searched_for published as “buried”, “burial”, “tomb”. A reader can run those three words against the transcript and check the claim. barron recording #250, “Fear No One” (2026-06-20)

unverifiable

Not a treatment, and deliberately not one: it is what an absent claim becomes when one of its own search terms turns up in the transcript. Such a row counts under none of the five treatments, which is why the five sometimes do not add up to the window — the shortfall is published rather than absorbed.

From this corpus. death_for_sins was claimed absent with “cross” among its search terms, and the transcript contains “The great John of the cross knew what it meant to walk through the dark night”. The word occurs, the claim cannot be confirmed, and the row is stored as unverifiable rather than as a finding in either direction. wncathedral recording #343, “Sunday Sermon by The Most Rev. Jeremy Greaves” (2025-08-24)

The great John of the cross knew what it meant to walk through the dark night ▸ 9:15

transcript 8:45–9:50 · Whisper

any other place. One is a story about fear and one is a story about hope. hope. A few years ago Anne Lamott wrote hope begins in the dark. The stubborn hope that if you just show up and try to do the right thing the dawn will come. You watch and wait and work you don't give up. hope that will done. The great John of the cross knew what it meant to walk through the dark night of the soul bent low by the darkness. But for John the darkness was not an absence of God it was the place where God was most present teaching us to trust beyond what we can see. Or as Barbara Brown Taylor says at the heart of the faith is learning to walk in the dark. And it seems to me that that is

Which text the reading was made against

4 terms: teaching, unsegmented, altar_call, withheld
teaching

The row was read against a resolved teaching span: the preaching, with the readings, music, announcements and altar call outside it. This is the basis a per-thousand-word figure about preaching can honestly be computed on.

unsegmented

The row was read against the whole recording, because no teaching boundary could be resolved for it. The figure is real and it answers a different question: it counts everything in the upload.

What it does not mean. It is not a worse version of the same number. Numbers on the two bases are over different populations of words and may not be averaged or compared, which is why the basis rides with every figure instead of being assumed.

altar_call

The row was read against an altar call span in its own right, not pooled with the preaching. A recited appeal is a real part of a service and a fixed script; measuring it apart is what keeps it from either vanishing or standing in for the sermon.

withheld

Two different instruments print this word on one church page, and they mean different things. In the provenance line and on a component, it means the per-thousand-word rates were not published: thin_window when the church has fewer classified recordings than the floor of twenty, stale_basis when classifications were made against text this database no longer produces, and unknown_basis when rows do not record which text they read at all, so there is no denominator to divide by. In the condemned block it is a placement: the church is below the same floor, so counts are reported and no stance is stated.

What it does not mean. Neither is a finding about the church, and neither is a missing number. Both say the population is too small or too incoherent for a rate to mean anything, and both publish the underlying counts anyway.

From this corpus. A church with four classified recordings has every rate withheld with the reason thin_window, against a floor of twenty. Its counts are all still on the page. apologia

What a reframe put in place of the verse's object

5 terms: within, narrowed, redirected, inverted, unrelated
within

The sermon's object stays inside what the anchor verse supports. The nearest of the five, and the common case in a faithful paraphrase.

From this corpus. “that He was buried” against “the grave couldn't hold Him”, on 1 Corinthians 15:4. gracechurch recording #710, “Presenting Every Man Perfect, Part 1” (2026-01-12)

the grave couldn't hold Him ▸ 58:42

transcript 58:11–59:16 · Whisper

Him, had given Him from eternity past. Everyone who would place their faith in Him, their sins were laid upon the shoulders of the substitute. And He was forsaken that we might be received. He was abandoned that we might be welcomed. He was punished that we might be justly forgiven. And then He was raised from the grave. He Himself had the authority to take up His life again and rise from the grave on the third day in victory and conquest over sin and death, testifying that the grave couldn't hold Him, that all stood accomplished. And He ascended into heaven and now calls everyone within the sound of His voice through this word, through His preachers, to come to Him and find forgiveness and righteousness and rest. By laying down all of the quote-unquote good works you would seek to offer Him, to please Him, to earn your way to heaven, He says, put all of those away. Come to Me, receive My righteousness, receive My forgiveness through

narrowed

A narrower object than the verse names. Part of what the verse says, and not something else.

From this corpus. Acts 2:36 names Jesus “both Lord and Christ”; this sermon's object is “Lord”, the title Mary and Martha use in the Lazarus narrative the sermon works through. One of the two, which is what narrowed means. christlutheran recording #933, “Mourning & Wonder” (2026-03-22)

Lord ▸ 14:42 Said 24 times in this recording; linked at the first time inside the text the claim was read from. Also at ▸ 1:03, ▸ 1:25, ▸ 1:33, ▸ 1:43, ▸ 1:58, ▸ 2:07, ▸ 2:14, ▸ 2:28, ▸ 2:59, ▸ 3:26, ▸ 3:41, ▸ 3:57, ▸ 4:11, ▸ 4:25, ▸ 4:33, ▸ 5:31, ▸ 6:22, ▸ 6:55, ▸ 7:37, ▸ 7:51, ▸ 8:21, ▸ 9:16.

transcript 14:10–15:14 · Whisper

for someone he loved. Does that make much sense? Why would he do that? I feel frustrated just reading about it. I want Jesus to come sooner. So can you imagine how Mary and Martha felt? I'll tell you what. They let Jesus know how they feel. On this fourth day, Mary runs to Jesus. Martha kneels before Jesus and both at different times exclaim, Lord, if you had been here, my brother would not have died. Such a tender declaration. So raw. And so born out of real, frustrating grief. As our Matisse devotional, which we're reading as a community, shares, this kind of declaration points to an ancient question. Where is God in times of suffering and death?

redirected

A different object from the verse's. Not a smaller version of it and not its opposite — another thing.

From this corpus. Against Acts 2:36's “both Lord and Christ”, the sermon's object is “Teacher”: Jesus questioned by an expert in the law and answering in parables. A title scripture also gives him, and not the one the anchor verse is about. chulavistacoc recording #3324, “The Good Samaritan: the neighbor who celebrates Jubilee” (2025-11-16)

Teacher ▸ 2:47

transcript 2:16–3:20 · Whisper

And this is another one that came to me. If we could just boil it all down and say, what does it mean to be a child of God? What does it mean to be a Christian? What would that look like? What would be inside of that jar that would say, you can make a Christian out of this? This is what it means. In Luke chapter 10, Jesus gets asked such a question. An expert in the law stood up to test Jesus. Teacher, he said, what must I do to inherit eternal life? Now, if you know the Old Testament, the Old Testament is a very large set of books. Two thirds of your Bible is Old Testament. And he comes to Jesus and he says, what do I have to do? Can you boil down what we would call two thirds of our Bible? For them, they called it 100% of their Bible. Can you boil it down and just tell me, what are the basic things I need to do to inherit eternal life?

inverted

The verse's object, reversed. The rarest label in this corpus by a wide margin.

What it does not mean. It is not a finding that the sermon contradicted scripture. It is a coordinate against one anchor verse, and the sermon's own words are very often another part of scripture. A label on this scale says how far the object travelled from the anchor's, and nothing about whether the journey was legitimate.

From this corpus. The anchor is Acts 17:31, “judge the world with justice”. The sermon's object comes from a sung line: “For God did not send his Son into the world to condemn the world” — which is John 3:17. Reversed with respect to this anchor, and a quotation of the Bible. bethlehemlutheran recording #3119, “BLC Good Friday Worship April 3, 2026” (2026-04-04)

For God did not send his Son into the world to condemn the world ▸ 17:50

transcript 17:20–18:37 · Whisper

the For God did not send his Son into the world To condemn the world, not to condemn the world But through the Son the world might be saved The world might be saved God so loved, so loved the world

unrelated

An object the anchor verse does not bear on at all. The furthest of the five.

Unexemplified. Not one of the labelled pairs in this corpus carries it. The value is in the vocabulary and a page can still print it, so it is glossed here and left unexemplified rather than illustrated with a case nobody measured.

Where preaching stands to a condemned proposition

11 terms: affirms, denies, not_addressed, claimed_not_published, mixed, insufficient_recurrence, preacher_asserts, quoted_to_refute, attributed_to_another, not_quoted, reads_scripture
affirms

One pass read one recording as asserting something a named council or confession condemned, and quoted the sentence it rests on. At church level the same word is a placement, and needs the pattern to recur across the body of work before it is stated at all.

What it does not mean. It does not mean the church is heretical, and it is not a finding about a congregation. It is one reading of one sermon against one clause, and most such readings do not survive the gate: three independent passes are asked to refute every one, and one withdrawal sinks it. A claim that has not cleared that is shown as provisional or withdrawn, never as a finding. Nor does an affirms on one axis say anything about another: the position condemned as Pelagian affirms the Nicene Creed, and the position condemned as Arian denies it and says nothing about grace.

From this corpus. A sermon was read as affirming a proposition the presbyters of the church at Rome condemned in Marcion around 207-212 — that the God of the Hebrew scriptures is the creator of this world and the giver of the law. The sentence behind the reading is “God created the heavens and the earth… He is the sovereign over this universe.” All three refutation passes withdrew the claim on the ground an_orthodox_reading_is_available, one of them writing that the sentence is “the Nicene confession of God as maker of heaven and earth, taught by every mainstream tradition”. The proposition is also marked as one an orthodox church affirms too, so it is withdrawn structurally as well. The claim does not stand; it is on the page because a withdrawn claim is shown rather than deleted, so the gate can be judged rather than trusted. gracechurch recording #615, “How Should We Think About Israel?” (2026-08-11)

God created the heavens and the earth… He is the sovereign over this universe. ▸ 9:34 Not word-for-word — assembled from fragments.

transcript 9:04–10:10 · Whisper

Why doesn't God start, if God is all about Israel, for the sake of argument, if God is focused upon them, if they're the chosen people, for the sake of argument, why doesn't God just start with the story of Abraham? Why doesn't he start in Genesis 12? Why does he have 11 chapters not about Israel, if that's all that God is about? And the answer is, it's a reminder to Israel and a reminder to us, God's plan is bigger than any nation. God's plan is bigger than you and me. God's plan is in the beginning. God created the heavens and the earth. His plan extends to all things. He is the sovereign over this universe. This is his created order. This is our father's world and he will have glory in it all. Israel needed to remember that they were not about them, that God was not just about them, that God had chosen them not just for their sake, but for the sake of his plan for the entire universe. That is the glory of God. And the way it plays out is that God has made all things. This is his world. He will get glory from every facet and every aspect of it.

denies

The recording was read as asserting the opposite of the condemned proposition. In this corpus it is far the commoner of the two stances wherever anything was addressed at all, and like affirms it is a reading of particular sentences rather than a statement about a congregation.

not_addressed

Nothing in the text bore on the proposition either way. On the statement side the same word means the document was read and is silent on a component — which is not the same as no statement having been collected. Silence is not a position, and nothing is asserted about a proposition that reads this.

claimed_not_published

At church level: every claim on the proposition was an affirms that the refutation gate has not published — withdrawn by a pass, or not yet through all three. The church page says which, and how many of each, and points to the passes themselves in the refutation layer.

What it does not mean. It is not not_addressed. The preaching was claimed to affirm the proposition and the gate refused or has not yet decided that claim; reporting it as silence would say nothing was looked at. Nor is it an affirmation: nothing is placed.

mixed

At church level, both stances recur across the body of work. A real state, and not a midpoint between them: it is never averaged into either.

insufficient_recurrence

The proposition was addressed, but by fewer recordings than the floor a stance needs. One sermon is never a finding about a church.

What it does not mean. It is not not_addressed. Reporting a proposition one sermon affirmed as unaddressed would be false, and reporting it as affirmed would be a finding drawn from a single sentence.

preacher_asserts

How the sentence stands in the mouth that said it. An affirms may rest on this attribution and on no other, and that is enforced where the row is stored rather than asked of the model.

quoted_to_refute

The preacher quoted the position in order to reject it. Cannot support an affirms.

attributed_to_another

The preacher reported someone else's position. Cannot support an affirms.

not_quoted

No quotation was produced for the sentence. Cannot support an affirms.

reads_scripture

The sentence is a biblical text being read aloud, not a claim being made. It has its own word because a vocabulary without one forces a verse read from the lectern into preacher_asserts.

What it does not mean. It is not a judgement about the reading. It is the reason a verse cannot, on its own, make a preacher the author of what it says.

What the refutation gate did with a claim

13 terms: provisional, published, withdrawn, upheld, evidence_is_scripture_read_aloud, preacher_is_not_asserting_in_his_own_voice, evidence_does_not_support_the_proposition, proposition_does_not_say_what_the_classification_claims, an_orthodox_reading_is_available, proposition_is_not_discriminating, verbatim_or_near, echoes_scripture, no_scripture_match
provisional

An affirms that has been stored and not yet cleared by three passes. Never published, and never read as a finding.

published

Three independent refutation passes upheld the claim and none withdrew it.

What it does not mean. It is still a claim about sentences in one recording against one clause. It is not a label on a church, a preacher or a tradition, and the axis it sits on bounds it: a published claim on one axis says nothing about any other.

withdrawn

A pass found a reading on which the sentence is ordinary teaching, or the claim misread the quote or the proposition. One withdrawal is enough. Withdrawn rows are kept and shown, because a pattern of withdrawals is itself a finding about the classifier.

upheld

One pass could not refute the claim. Three of these and no withdrawal is what publishing requires.

evidence_is_scripture_read_aloud

Withdrawal ground: the quoted sentence is a verse being read, not a claim being made.

preacher_is_not_asserting_in_his_own_voice

Withdrawal ground: the preacher was quoting, reporting or voicing someone else.

evidence_does_not_support_the_proposition

Withdrawal ground: the quote does not carry the claim the classification put on it.

proposition_does_not_say_what_the_classification_claims

Withdrawal ground: the proposition says something narrower, or other, than the classification took it to say.

an_orthodox_reading_is_available

Withdrawal ground: there is a reading on which the sentence is ordinary orthodox teaching. The most common ground in this corpus.

proposition_is_not_discriminating

Withdrawal ground, and the only one no model may return. The proposition itself is one an orthodox church affirms too, so no reading of any sermon could make an affirms on it mean anything. It is decided from the proposition data without a model call.

What it does not mean. It is not a judgement about the sermon at all. It says the question could not have distinguished anything, which is a fact about the instrument.

verbatim_or_near

A scripture check on the quoted sentence, carried with every pass as information and never as the answer: the words overlap a verse closely. A mechanical rule that can be wrong in both directions may not publish or withdraw anything, so it is handed to the reviewer instead.

echoes_scripture

The quoted sentence overlaps a verse loosely.

no_scripture_match

No verse was matched to the quoted sentence. Not evidence that it is unbiblical — a loose paraphrase, or a translation none of the indexed five matches, barely overlaps anything here.

What the church's own statement of faith says

4 terms: addresses, affirmed, unsupported, not_reached_by_this_instrument
addresses

The church's own statement of faith speaks to this component, and the clause that does is quoted verbatim from the document.

What it does not mean. It is not agreement with the sermon, and it is not a claim about sermon content at all. A statement of faith is an act of assent: a church may genuinely hold that Christ rose and not preach the resurrection every Sunday. What any gap between this column and the preaching column reports is a distance between two documents over a stated window — never insincerity, hypocrisy or failure.

From this corpus. For god_holy the verdict is addresses, and the clause it rests on is quoted out of the document: “We believe in one God, revealed as Father, Son, and Holy Spirit, the loving creator, redeemer and transformer, and Lord of all.” The same statement is silent on several other components, and those read not_addressed — read and silent, which is a different fact from no statement having been collected. crossroadscrc statement of faith — Who We Are, Our Core Values in nine articles

affirmed

What this verdict was called before 2026-09-15, and what records written earlier still store. It was renamed because the condemned layer uses affirms for something sharper and nearly opposite in valence — preaching that ASSERTS a condemned proposition — and one root word doing both jobs is how a reader ends up thinking a statement of faith affirmed a heresy.

unsupported

Assigned by this project and never by the model: addresses was claimed, and no clause survived verification against the source document. It counts as neither an affirmation nor a silence.

Unexemplified. Every component verdict in the statements classified for this corpus is addresses or not_addressed; none is currently stored as unsupported. A page can still print it, so it is glossed and left unexemplified.

not_reached_by_this_instrument

The cell is empty because the instrument does not reach here, not because a church was silent. Some axes have no gospel component that lands on them at all — where a claim is grounded is one of them — so the preaching, scripture and statement columns have nothing to say about those rows for any church in this corpus.

What it does not mean. It is not an absence in the preaching, and it is not a zero. It is this tool reporting the shape of its own reach, which is why those rows are printed instead of left out.

The two gaps, which are not two tallies

2 terms: addressed_absent_in_window, addressed_never_taught_in_window
addressed_absent_in_window

The statement addresses this component and no recording in the window contains it at all — present in zero of them.

addressed_never_taught_in_window

The statement addresses this component and no recording in the window develops it, whether or not it appears.

What it does not mean. These two are not two tallies to add. The second is a superset of the first: a component present in every recording and developed in none is counted here and not there, so they must never be added together. Both are published, neither is chosen for the reader, and both can be recomputed from the per-component treatment counts on the same page.

Measured here, shown on no page of this app

4 terms: confirms, contradicts, silent, addressed_without_settling
confirms

A statement of faith, read against a condemned proposition, says what the proposition says.

contradicts

The statement says the opposite of the condemned proposition.

silent

The statement does not speak to the proposition either way.

addressed_without_settling

The statement speaks about the subject and does not settle the question the clause turns on. There is no page in this app that shows any of these four verdicts — they are measured by the command line and served nowhere here — and they are glossed because the word is easy to meet elsewhere in this project and to read as a verdict about a church.

What it does not mean. None of the four is a ruling on orthodoxy, and the middle two are not a spectrum with a good end. This verdict in particular is the instrument declining to answer, and it was worth building for exactly that: reading ten of these pairs by hand, knowing each church's tradition, produced confirms four times where the instrument — held to judge only what is on the page — said the page does not settle it. The instrument was right. A verdict about a document is a claim about its sentences, never about its denomination.

From this corpus. Against the clause condemning those who divide the one Godhead into three separated hypostases as three gods, the statement's own words are “We believe in God, the Eternal Father, and in His Son, Jesus Christ, and in the Holy Ghost.” That names three and settles neither side of what the clause turns on, so the verdict is addressed_without_settling. riverwood statement of faith — the Articles of Faith, article 1

Rows these pages hold apart from the main list

2 terms: central_publication, outside-creedal-consensus
central_publication

A row that is a text written centrally rather than a congregation: the weekly study article of a tradition whose congregations all study the same article in the same week. It is counted in no figure about churches, anywhere here. Its citations are read from the verse codes its publisher attaches to each reference, so none of them was parsed from speech, and its receipts link a paragraph rather than a moment.

What it does not mean. It is not a congregation and not a sermon, and its figures are not comparable with a church's as though it were: authorship, medium, length, format and timing all differ, and the separate section says how. The interpretive passes — the gospel components, the reframes and the condemned propositions — are run on it only as a run of their own, read each article whole (basis article, never teaching), and are reported only in that section; where it says a pass has not run, that is a pass not run and never a measurement of nothing.

outside-creedal-consensus

The family a tradition file in this project records for a tradition whose own published doctrine stands outside the Nicene consensus — one God in three persons, and the Son of one substance with the Father, as Nicaea (325) and Constantinople (381) set it out. Every church naming such a tradition, and every central publication of one, is listed in a separate section of this site and in none of the main list's groups, beside its tradition's own words and what was measured of its preaching.

What it does not mean. It is not a verdict by this project on anyone's faith. The field is set from the tradition's own documents, and the section quotes them; the section is separate so that these churches are not read as peers of a list they do not claim to belong to, and every measurement on their pages is made exactly as it is for any other church.

How to read a receipt

date, excerpt, link

What the date, the excerpt and the ▸ mm:ss link beside a quotation mean, and how early a link can land.

Every quotation a page here takes from a recording carries a receipt: the date and title of the recording, a transcript excerpt that opens in place, and a link (▸ mm:ss) to the recording at the moment the words begin — so the words can be heard rather than taken from our transcript or our labelling. Where no moment can be honestly established the receipt says which reason applies, and no two reasons are worded alike:

at a moment
▸ 12:34 — Said once in this recording, at this moment.
said several times
Said N times in this recording; linked at the first time inside the text the claim was read from. The other moments are listed after it, each its own link.
assembled from fragments
Not word-for-word — assembled from fragments. A quote joined with … was found piece by piece, in order; the excerpt marks each piece and not the words between them.
paraphrased
Paraphrased — there is no exact moment to link. The link goes to where the teaching begins, or to the recording itself where no teaching span is established.
not located
Quote not located in the transcript. The link goes to the whole recording, never to a guessed moment.
audio only
Audio only — no video exists for this recording. The link opens the church's own audio file at the moment, with a #t= media fragment.
published as text
Published as text; no moment exists. The link goes to the page the church published, and the excerpt is cut from that text.
a paragraph of a written article
A written article, not a recording — linked at its paragraph, so there is no moment to give. A central publication — a weekly study article written centrally and studied in every congregation of its tradition — has no clock and no recording. Its receipt links the paragraph on wol.jw.org, the publisher’s own library, by the paragraph id the publisher’s pages share, and a citation printed in the article is linked the same way, at the paragraph it is printed in. An excerpt, where there is one, is cut from inside that one paragraph and is at most 300 characters: that is all of the article’s text a published copy of this corpus keeps.
a paragraph of a printed sermon
A printed sermon, not a recording — linked at the edition it was printed in, with its paragraph named; there is no moment. A historical preacher’s sermon survives only in print. Its receipt links the page of the edition it was read from and names the paragraph, numbered as the stored text numbers them; the excerpt is cut from inside that paragraph.
a spoken citation
Named aloud at this moment. On the search page each citation of a passage links to the moment it was named; named aloud; the citation pass stored no moment for it.
a word search hit
▸ 12:34 — Linked at the start of the transcript line these words begin in, so it can land up to one line early and never late. On the search page each place the words were found shows the words around them, the matched words marked, and links to the moment of the transcript line they begin in. A recording with no clock links its published page or audio file instead, and never a moment of zero.
a word search hit not located in time
Matched by the index, but not located before the time limit, so it is listed without its moment or its words.
not built
Receipts were not built for this corpus, so this quote has no moment or excerpt here — that is a step not run, not a quote that could not be found.
built without this quote
This corpus's receipts were built without this quote, so it has no moment or excerpt yet — the quote is newer than the receipts, not missing from the recording.
not looked up
No receipt was looked up for this quote by the page that shows it.

The excerpt is labelled with what it is a transcript of — YouTube's automatic captions, captions the channel uploaded to YouTube, our Whisper transcription of the audio, the text as the church published it, the article as published on jw.org, the sermon as its printed edition gives it — and the quoted words are marked inside it. It is an excerpt and not the transcript: a few minutes either side of the quote, wider where a refutation pass judged the claim, and the recording itself is the church's.

A link can land early, never late. A moment is the start of the transcript segment the quote begins in. Most segments are a few seconds long, but a Whisper segment can run longer, and then the link lands up to that segment's length before the words. It never lands after them.

The limits that apply to every church

caveats

The limits stated once here because they are worded identically on every church page, and the comparison that put them here.

17 caveats, every one of them returned by the measurement itself rather than written here, and every one of them worded identically on every page that carries a panel — 78 churches and 23 reference figures. They are stated here once instead of on every page, and every one of them applies to every figure on every church page. A caveat whose wording is about a particular church is not here: it stays beside that church's figures. These are the measurement's own words, unedited, so where one of them says this table or below it means the church page the figures are on, not this one.

Derived, not listed: every caveat of every page was rendered and compared, and a caveat reaches this page only if its text is byte-identical on all 101 of them and all 101 carry it. The 3 churches and 1 reference figure with nothing measured yet have no panel, so there is no caveat block there to have agreed or disagreed with; each of those pages says which stage its own work stopped at. The historical preachers are not in this comparison: their printed sermons are read on their own page.

high The preaching, scripture and statement measurements cannot reach these axes for ANY church: no gospel component lands on them. Their cells read "not reached by this instrument", which is a fact about the instrument and never about the church.

About the doctrine axes

high The comparison window is whatever sermons happen to have been classified. It is not a sample of a church's year and was not randomly selected. Every preaching figure here is over that window and nothing else, and a component absent from it is absent from those sermons rather than from the pulpit.

About the preaching column

high Classifier variance is larger than the differences measured between churches so far: the same sermon on a byte-identical prompt scored 9/11 and 6/11. Read these as an ordering, not as precise rates.

About the preaching column

high These components are fully measured in preaching, scripture and statement and have no tradition or condemned counterpart, because the seven axes record no position on them. That is the reference having no field, never a tradition being silent -- every tradition here confesses the resurrection.

About the doctrine axes

high A claim resting on a quoted confessional article and a claim resting on one pass of a noisy classifier are both printed here, side by side. They are not equally strong and the columns must not be read as if they were.

About every column

high Two different gaps are published and neither is picked for the reader: the statement addresses a part and no sermon contains it, and the statement addresses a part and no sermon develops it. The second includes the first, so the two must never be added together. Both can be recounted from the preaching figures.

About the doctrine axes

high Which axis a gospel component belongs on is a judgement, not a measurement. Every link is recorded with its reason in this project's data, where anyone can check it and dispute it.

About the doctrine axes

high Five columns of evidence, not a pipeline that emits a verdict. Each layer has a different instrument behind it and a different failure mode; nothing here composes them, and no score, rank or alignment figure exists to be read off it.

About every column

info An axis appears once per component that links to it, carrying that component's own anchor verses, sermon evidence and confessional citations. Those rows are not duplicates and must not be collapsed.

About the doctrine axes

info The treatment totals count classifications across components. They are not a score and must never be presented as one: components are not weighted, not ranked, and not summed into a verdict.

About the preaching column

info These are counts of components, not a grade. No ratio is computed from them and none may be: a gap count over an eligible count is a fidelity percentage with a different name, and this project's founding rules forbid one.

About the doctrine axes

high Presence and teaching are different questions, and both are asked here, because a component can be present in every sermon of a window and developed in none of them. All five treatments are published per component and none is summed away: "taught" means worked out rather than merely named, "reframed" carried in the sermon's own vocabulary, "formulaic" recited near-verbatim across this church's other sermons. Whether a reframed part counts as preaching it is the reader's call and this site does not make it, which is why the numbers are shown instead of a verdict. Measured example: over 24 Lakewood sermons the call to repent is absent in none and taught in none (16 reframed, 6 formulaic, 2 in passing), and "taught" occurs twice in total across all eleven components.

About the preaching column

high Reference figures are in this table. An influential preacher's archive is not a congregation's year and the two are not peers.

About the preaching column

high A statement of faith is an act of assent, not a claim about sermon content. The project's standing ruling is that affirmation is universal while teaching is not: a church may genuinely hold that Christ rose and not preach the resurrection every Sunday. What is reported here is a gap between two documents over a stated window -- never insincerity, hypocrisy or failure, and nothing here may be read as any of those.

About the statement and the preaching side by side

high Each statement was classified once. The clause set below is one classifier's reading of one document on one pass, under one version of its instructions; no spread across runs exists.

About the statement column

high A short statement has fewer axes on which it CAN diverge. Lakewood's runs 1,060 characters against Apologia's 87,370, so a low gap count from a silent document is not agreement. The number of axes the statement addresses at all is published beside every gap count for that reason.

About the doctrine axes

high The tradition reference records positions on seven axes chosen because sermons reveal them -- authority, God and Christ, soteriology, anthropology, sacraments, eschatology, ecclesiology. Some gospel components have no axis there for ANY tradition, and others have an axis that a given tradition's file carries no row for. Both read "not recorded", and both are facts about the reference rather than findings that a tradition is silent: every tradition in this corpus confesses the resurrection, and Trent teaches Christ's death for sins, whether or not a row records it.

About the tradition column

Limits most churches share

Worded identically on many church pages but not on all of them, so none of these is asserted of every church: each applies to a church only where that church’s page links to it, and each says how many of the 101 pages compared carry it in these words.

worded identically on 97 of 101 pages

high One classification pass. Nothing in this window was classified twice, so its numbers carry no spread across runs -- and one pass is one pass, not agreement. A second run on the same sermons would be recorded beside the first.

About the preaching column

worded identically on 51 of 101 pages

high A statement exists for these churches and has not yet been read by the classifier. Again not "not addressed".

About the statement column

worded identically on 90 of 101 pages

medium Some components carry runs made under a different rubric or over a different text. Those are NOT pooled into the spread: runs under a different rubric answer different questions, and runs over a different text read different words, so only the most recent comparable group is counted. Nothing was deleted.

About the preaching column

About this site

corrections, takedowns, privacy

What this site is and is not, the errors a transcript can carry, how to ask for a correction or a takedown, and privacy.

This site measures and describes what churches said in public recordings. It does not accuse anyone of anything, and it does not rule on anyone’s faith: it sets what was said beside the Bible text, the church’s own statement and its tradition, and leaves the reader to judge.

Quotes come from transcripts — automatic captions or our own transcription of the audio — and transcripts can contain errors: a misheard word, a name, a sentence given to the wrong speaker. Before relying on a quote, check it against the linked video at the moment given.

If something here is wrong, misquoted, or names a private person, please report it. Every report is read by hand.

Copyright: if you hold the rights in material quoted here and want it removed, use the same form with the reason copyright or a takedown request.

Privacy: visits are counted without tracking cookies or trackers. If you send a request or a correction, your email is used only to reply and is never shared.

Earlier work

Pew 2019, Dodd 1936, the critics

Who counted sermons' Bible references before, who drew a list like this one from the same speeches, and who disputes it.

  • Pew Research Center, The Digital Pulpit: A Nationwide Analysis of Online Sermons (16 December 2019). 49,719 sermons from 6,431 churches, delivered 7 April to 1 June 2019, with the Bible books they name counted. This site’s citation count is the same kind of measure, over far fewer churches. Pew’s churches skew larger and more urban than U.S. congregations as a whole. The churches here were chosen by hand and are not a sample at all. the report
  • C. H. Dodd, The Apostolic Preaching and Its Developments (London: Hodder & Stoughton, 1936). He drew a list of the early preaching’s elements from the speeches in Acts and from Paul, as this site’s eleven parts are drawn. He also wrote that much of the preaching in church in his day “would not have been recognized by the early Christians as kerygma. It is teaching” (Lecture I, read in the 1937 Chicago edition). the 1936 record the 1937 text
  • Critics of a single kerygma. D. A. Templeton (Glasgow PhD thesis, 1967) argued that the New Testament holds not one kerygma but many, some of which contradict one another. the thesis Robert C. Worley, Preaching and Teaching in the Earliest Church (1967), disputed Dodd’s line between preaching and teaching, as cited by page in Ministry, March 1994. the book Ministry C. F. Evans, “The Kerygma”, Journal of Theological Studies n.s. 7 (1956) 25–41, is listed by title only: it has not been read here. the article
  • What this site does with that. It uses one list and says so. It shows the speeches the list came from (/acts). It does not claim the list is the only reading.