What We Got Wrong

¶ What We Got Wrong

Short answer. This is a standing record of the things this project got wrong and what it did about them. Four working theses were tested internally and rejected or reframed before they were published; one circulated figure had to be taken back and replaced with a number 22 points lower; and one of the project's own instruments failed an adversarial re-score of its own results. The arithmetic is given where there is arithmetic. The page exists because a method that is never seen failing is not a method, it is a brand.

One precision that governs everything below, and it cuts against the page: most of these were caught in internal triage before publication rather than retracted after it. Catching a bad thesis before printing it is ordinary diligence, not heroism. The genuinely uncomfortable item is the third one, where a figure that had already circulated was wrong and had to be replaced. Each entry says which kind it is.

¶ A thesis that was backwards

Caught in triage, never published. The working thesis was that Revelation functioned as the church's fear-and-control instrument. Half of that is strong and survives: Revelation had the most contested canon history of any New Testament book. Dionysius of Alexandria argued around 250 CE, on grammar, style and vocabulary, that the John of Revelation cannot be the John of the Fourth Gospel, which is sophisticated literary criticism seventeen centuries early. Eusebius files it among the disputed books. It is absent from the original Peshitta, from the canon list of the Council of Laodicea, and from Gregory Nazianzus's list. It is still not read liturgically in the Orthodox lectionary. Luther, in his 1522 preface, called it “neither apostolic nor prophetic.”

The other half is the opposite of what happened. Revelation is an anti-imperial resistance document written under Roman power. Babylon is Rome, about as plainly as a persecuted author could say it. A text written by the persecuted against an empire was later put to work as an instrument of control, which is a claim about its reception, not about the document. Stated the original way, a competent critic would have used the false half to discredit the true half, and would have been right to.

¶ A framing that recycled a trope we have a Book against

Caught in triage, never published. An early working formulation ran that Paul was unreliable, “being the Pharisee that he is.” Two things are wrong with it, and the second is worse than the first.

The chronology problem. Paul's undisputed letters are the earliest Christian documents in existence, written around 50 to 60 CE, before any gospel. “Paul corrupted the original message” requires access to a pre-Pauline Christianity we do not have except through Paul himself. You cannot use the earliest witness to convict the earliest witness of being late.

The trope problem. Using Pharisee as an epithet descends directly from a Christian caricature of Judaism as legalistic works-righteousness, the caricature E. P. Sanders demolished in 1977 in Paul and Palestinian Judaism by showing that Second Temple Judaism ran on covenantal nomism rather than merit-earning. A corpus that carries a whole Book on the de-Judaizing of Christian origins has no business recycling the trope that made Paul's Jewishness the thing wrong with him. It is also self-defeating: Paul's Pharisaism is precisely what makes him a reliable Jewish witness to the period.

¶ A number that had to be taken back

This is the real one. A figure of 87.4% for the Iranian influence case had already circulated internally when the model was extended with additional evidence lines. The posterior went up to 98.7%.

That jump was the problem. A move that large from six added nodes is the exact pattern the project had already documented as its own earlier failure mode, so it was checked rather than published. The symptom showed before the cause: correlation-aware aggregation started returning higher odds than naive aggregation, which is the wrong direction. The entire purpose of a correlation discount is to deflate double-counted evidence. If it inflates a case, something is structurally broken.

The cause was a cluster-assignment error, and it had been flattering the thesis. All counter-evidence had been filed into a single bucket. That is wrong on the model's own logic: objections are not correlated with each other, they are correlated with the claims they rebut, because they share those claims' failure modes. The objection that Iranian ritual vocabulary never crossed and the objection that Greek sources are Greek constructions have nothing in common except being objections. Lumping them together meant the sample-size discount treated ten counters as roughly three, while seven supporting clusters were each discounted separately. The opposition was being systematically under-weighted.

Corrected, the case is 76.8%, with a 90% credible interval of 52.0 to 93.5. That is 22 points below the number that had circulated. It is a good case and not an overwhelming one, and the bottom of that interval is the honest thing to quote beside the mean. Two further runs from the corrected model: the dating-immune subset alone gives 68.0%, and arguing the thesis from the popular eschatological parallels alone gives 0.8%, which is why this site does not argue it that way.

¶ An instrument that failed its own audit

Caught by the project's own adversarial re-scoring. The project built a composite measure intended to detect doctrinal overwriting across a text. It was then run against itself: the same material, re-scored by an adversarial pass instructed to find reasons the measure was wrong. It failed.

The precedent was already on the shelf and had been ignored. William O. Walker's Interpolations in the Pauline Letters (2001) is principled interpolation-hunting that was received coolly for lacking a control on the scorer's own subjectivity. The project's auditor reproduced Walker's exact wound. The measure is not in use as a load-bearing instrument, and it will not be until an inter-rater study exists.

¶ A chain that erased a lineage

Caught in triage, reframed before publication. The working claim was that logos comes from Heraclitus and the Jews took it. The defensible part: Heraclitus gives logos its cosmic sense around 500 BCE, the Stoics develop it into an active ordering principle, and Philo of Alexandria, a Jew, fuses it with Jewish categories a century before John.

What is not true is the word “took.” Three streams converge at John 1 and the Greek one is merely the loudest. Jewish personified Wisdom is an independent and older stream, present at creation in Proverbs 8:22 to 31 and developed in Sirach 24 and Wisdom of Solomon 7. The Aramaic Memra of the Targums is a third, separate development. John 1 is a merge, not a borrowing. Saying the Jews took logos from the Greeks erases Proverbs 8, which a hostile reader would produce immediately and correctly.

¶ Why this page exists

Not as a display of humility. It is here because of a specific gap in how this subject is normally argued. The scholarship this project stands on is good, and the people who produced it are cited by name elsewhere on this site. But the working medium of the field is the trade book and the monograph, and neither format pre-registers a falsifiable prediction with a timestamp, publishes an adversarial re-score of its own instrument, or prints the arithmetic of a retraction alongside the claim it replaced.

That is not an accusation against anyone. It is a description of what books cannot do. It also means the practice is available, and this project has decided to occupy it: conventional company on evidence, and no company at all on method.

The commitment this page represents is narrow and checkable. When a load-bearing claim here is shown to be wrong, the correction appears on this page with the reasoning and the number, and the original stays visible rather than being quietly edited away.

¶ Common questions

¶ Has the Foundation ever retracted a claim?

Yes, and the arithmetic is on this page. A figure of 87.4% for the Iranian influence case had circulated internally when an extended model returned 98.7%. The jump was investigated rather than published, traced to a cluster-assignment error that had been systematically under-weighting counter-evidence, and corrected to 76.8% with a 90% credible interval of 52.0 to 93.5. That is 22 points below the figure it replaced. Most other items on this page are working theses caught in internal triage before publication, which is ordinary diligence rather than retraction, and the page says so.

¶ What was the biggest single correction?

The 22-point one, because it ran in the flattering direction. Adding evidence made the case look stronger, and the reason it looked stronger was a bug: all objections had been filed into a single correlation cluster, so ten counter-arguments were discounted as though they were about three while seven supporting clusters were each discounted on their own. Correlation-aware aggregation began returning higher odds than naive aggregation, which is the wrong direction and was the symptom that exposed the cause.

¶ What would change your mind about the Iranian influence thesis?

Specific things, and they differ by evidence line, which is the point of arguing it in four independent strands. The loanword case fails if the etymologies are wrong, which is a philological failure. The bridging lexemes fail if they are mis-tagged, and the project's own instrument already flags one of the three as its most contestable call. The Samuel-to-Chronicles substitution fails if the dating of those sources is wrong, which is stratigraphic. The early-comparandum argument fails if the Greek reports of Persian religion are Greek constructions, which is a real and standing objection. A skeptic needs all four to fail, and they fail for unrelated reasons.

¶ Was the Revelation thesis published and then withdrawn?

No. It was a working thesis rejected in internal triage before it was published, and the page says so rather than dressing it up as a retraction. What was wrong is that it read an anti-imperial resistance document written under Roman persecution as an instrument of control. The canon-history half of the thesis was sound and remains in use; the causal half was the opposite of what happened, and a competent critic would have used the false half to discredit the true one.

¶ Why publish your own errors at all?

Because a method that is never seen failing is indistinguishable from a brand. Any project can describe its standards; the only evidence that standards are real is a record of occasions when they cost something. This page is that record, and it is deliberately specific, because a vague admission of fallibility costs nothing and proves nothing.

¶ Isn't a page like this just public relations?

It is a fair suspicion and worth stating plainly: yes, publishing corrections is also good for credibility, and a project that benefits from looking honest has an obvious incentive to look honest. Two things make this checkable rather than decorative. The corrections are specific enough to verify, with the numbers, the mechanism of each error and the scholarship that caught it. And they run against the project's own interest, since the largest one lowered its headline figure by 22 points and the others rejected theses it would rather have been able to argue. Judge it on whether the next correction appears here too.

¶ Sources

The corrections on this page are drawn from the project's internal triage and modelling records. Scholarship referenced: E. P. Sanders, Paul and Palestinian Judaism (1977); William O. Walker, Interpolations in the Pauline Letters (2001); Elaine Pagels, Revelations; Adela Yarbro Collins, Crisis and Catharsis; Dionysius of Alexandria, preserved in Eusebius, Ecclesiastical History 7.25; Martin Luther, preface to the Revelation of St. John (1522); Proverbs 8:22-31, Sirach 24, Wisdom of Solomon 7; Philo of Alexandria on the Logos.

The volume behind this page: doi.org/10.5281/zenodo.21879071

¶ Related

More from the sourced library:

• How Every Claim Is Weighed
• In This Tradition
• Did Zoroastrianism Influence Judaism?
• The Persian Passband
• Did God or Satan Incite David?