Their Reply to a Critic, Scored

Four points need to be separated from the question of whether the book’s historical conclusions follow. On these points, White’s criticism does not establish what he claimed.

Philemon. White argued that the book's claim about Philemon falling below 400 words was false, since "in most English versions of the Bible, Philemon is somewhere between 430 and 490 words." That is the English. The book states that it runs the Greek, and the Greek text they use is about 336 words. White works this out later in the same article, after leading with the accusation.

The NIV. White wrote that the authors "do claim to be using the NIV translation." The line he is reading is the standard front-matter notice that scripture quotations come from the NIV. It is boilerplate about quoting, not a statement about what was fed to the software.

Which distance measure. White complained that he could not tell whether Eder's Delta or Eder's Simple was used. The reply points to the caption printed beneath one of the book's own dendrograms, which reads "100 MFW Culled @ 0% Eder's Simple distance." The parameters were under the picture.

Scope. Stylometry occupies two of the book's nine chapters and does real work in one of them. White's series addresses that chapter while being framed as a refutation of the book. The complaint is fair.

None of this is trivial. A critic who gets the word count wrong by reading the English, misreads a copyright notice as a methodology claim, and misses a caption has handed his opponents three free wins, and they took them.

The centrepiece is a technical disagreement about most frequent words, and it repays slow reading.

White argued that because Eder's Delta performs best somewhere around 1,000 to 1,500 most frequent words, and the samples here are under 1,000 words and average closer to 400, "there is going to be a serious problem." The reply calls this a failure of understanding, and on the wording it is one:

The source site: "a text containing 1,000 words, isn't going to have 1,000 MFW's. In reality, it will have far fewer than that… 1 Corinthians 15 contains 843 words but only 397 are unique."

They then work through the arithmetic of that chapter, note that 272 of its words appear exactly once, and conclude that a setting of 100 most frequent words is reasonable.

Here is the difficulty. In stylo, the most-frequent-word list is not the vocabulary of the text being analysed. It is computed across the whole corpus and then applied as a single fixed feature set to every sample. The package's own authors describe the workflow plainly: the frequent items are harvested from the corpus, and then users "extract a vector for each text or sample, containing the relative frequencies of these features, and combine them into a frequency table for the corpus." The unique-word count of 1 Corinthians 15 is not a quantity the software consults.

So the arithmetic answers a question nobody asked. And the real objection survives untouched, because it was never about whether a short chapter contains enough distinct words. It is that estimating relative frequencies for a hundred shared features from roughly 372 tokens produces mostly sampling noise, which is the actual subject of the Eder paper both sides keep citing.

The fair verdict is uncomfortable for everyone. White stated the objection loosely and got corrected. The correction is not responsive to the objection properly stated. They win the exchange and lose the argument, and the Stylometry page sets out why at length.

Away from that exchange, the page recycles the errors the book is criticised for.

Dating. An independently established author can constrain a text's date. The difficulty is establishing that identity rather than assuming it from a cluster. A modern known-answer example does not validate the proposed ancient attribution. Our stylometry article distinguishes authorship, chronology and the evidence required to connect them.

The fifty-word study. Robert Gorman's 2020 study reports short-text identification using syntactic information. It shows why a blanket claim that short samples can never work is wrong. Its reported accuracy cannot simply be transferred to a different clustering method and corpus.

Eder on sample size. Eder's 2017 conference paper makes the dependence on text and comparison corpus explicit. General thresholds cannot replace a test of the actual procedure. Both criticism and defence of the book need evidence about its actual procedure, rather than a selective quotation for or against a universal minimum.

Genre. A dendrogram of Latin authors grouping by writer rather than by genre is offered as evidence that genre does not influence results. One favourable demonstration on large samples of well-attested authors does not establish genre-independence for 400-word chapters of Koine Greek.

Peer review. The charge that "peer-reviewed" is used misleadingly gets this answer: "It's in the statement, there is no asterisk or footnote. There is no intention to deceive." As a reply to the specific complaint that is a denial of intent rather than an argument, and it sits oddly beside the book's promotional copy, which advertises "proven, peer-reviewed mathematics" without the qualification.

It would be unfair to leave it there, though, because Matthew Britt had already made the distinction himself, in public, unprompted, a year before this page was written. Pressed on exactly this point by Peter Kirby on the Biblical Criticism & History Forum in August 2024, he wrote:

Matthew Britt, 14 August 2024: "We try to word it specifically that Stylo itself is peer-reviewed and that the mathematics behind it are as well, not that our specific results have been peer reviewed. I do see the concern there, but we tried to word it carefully. Maybe we failed in that aspect."

In the same post he says the book "is meant for a popular audience, not a scholarly audience" and that "we never claim to be scholars nor do we try to masquerade as such." That is a good deal more candid than the website's version of the same defence, it concedes the objection is reasonable, and it should be on the record here.

What survives the concession is narrow but real. The distinction Britt drew accurately in 2024 is not the one a listener takes away from the phrase "peer-reviewed" repeated on a livestream, and the publisher's copy still carries the unqualified claim. The wording was the concern; the wording did not change.

Running through the page, returned to four or five times, is a challenge that deserves a straight answer rather than a rebuttal:

The source site: "If stylometry was as unreliable as Dustin makes it out to be, he should have taken a page from his own critique and been transparent by providing screenshots, disclosing the data, and providing the parameters used to debunk Christ Before Jesus' findings. It's very eye-opening to notice that he didn't."

As logic this does not hold. You are not required to build a machine in order to explain why it cannot work, and a method whose inputs were never published in the book is a strange thing to be told you should have replicated. As a challenge to critics generally, though, it has force, and it is the most effective thing on the page.

It also happens to be answerable. The software is free, the Greek New Testament is available in open editions, and the comparison texts are on public repositories. Nothing prevents anyone from running the pipeline at the stated settings and publishing what comes out, including the known-answer controls that would settle whether chapter-length attribution works on this corpus at all. This site intends to do exactly that, and to publish the result whichever way it falls. Until then, the card stands.

One detail is worth recording because it is unusual. The page explains why it does not link to the articles it is answering:

The source site: "The articles are not directly linked to because they do not deserve additional attention. Searching for them will help his search engine results, but linking to them may convey legitimacy."

The page itself is reachable from a single line in a list of frequently asked questions, far down a separate page, where the link is carried by the words "many flaws" rather than by the critic's name. It appears in no navigation.

That is a choice about distribution rather than about argument, and it cuts against the material. The reply contains the authors' best technical writing and four solid hits on a critic. Filing it where almost nobody will encounter it, while declining to link the work it answers on the stated grounds of search rankings, treats a methodological dispute as a reputational one.

This site takes the opposite view, which is why the page you are reading exists, why it opens by conceding what they got right, and why their reply is linked here in full.