The pre-registrations

A researcher who needs a particular outcome is unreliable — not out of ill will, but because whoever searches hard enough for a desired result will always find a proxy for it. The only antidote is to fix the criteria and the falsifier before the search, so that the outcome cannot afterwards be bent towards the wish. This appendix lays bare the three tests that were set up in this way: the search for the double construction (the measurement of ch. 3), the search for the intercalary month of AD 31 (the narrowest link of ch. 8), and the extra spring between John 2:13 and John 6:4 (the hinge between AD 30 and AD 31, ch. 9 and ch. 14). With each stands the prediction as it read beforehand, the falsifier, and the outcome as it afterwards turned out — even where that outcome is unwelcome. Then the findings that ran the other way, each with its counter-weight beside it; and the public deposit that gives the register a date not supplied by the author.

E.0 · Why beforehand, and not afterwards

Three of the book's load-bearing claims run the greatest risk of self-deception, because the author has a stake in them. The first is that the double construction is a measure (without it, no sign — ch. 3); the second, that AD 31 had an intercalary month (without it, no Wednesday — ch. 8); the third, that the ministry counted a fourth Passover (without it, AD 30 and not AD 31 — ch. 9, ch. 14). For all three the same holds: go looking for confirmation, and you will find it. For all three, therefore, the test was set up in reverse — not "seek support" but "seek the falsifier, and fix beforehand what counts as refutation". What follows is that set-up, with the outcome beside it. Two of the three came back short of what was hoped, and they are written down here in the form in which they fell.

E.1 · The double construction — the search for a single inclusive case (with ch. 3, appendix B)

Fixed beforehand. If "three days and three nights" is a measure, then every narrative duration statement of it counts a full, continuous number of full days, and not a single inclusive use will be found in the held-out corpora. The prediction was thus a null prediction, and it is falsifiable by one counter-example.

The falsifier, beforehand. One attested narrative use of the double construction in which the actual duration is shorter than the stated number of full days (for instance, a "three days and three nights" that comes out at one and a half full days). Such a case would turn the measure into a figure of speech and disarm the sign.

The outcome. The held-out corpora (Aramaic and plene Hebrew, via Sefaria and Maagarim) were searched after the registration: 14 + 22 raw hits, and zero inclusive. The counter-tradition even articulates the rule itself (Menot haLevi, the Maharal on Esther 4:16), and a pre-Christian calendar computation (4Q252) makes it recomputably full. The falsifier was sought and not found. The full working, with the binomial base rate, stands in appendix B. [DOCUMENTED]

E.2 · The intercalary month of AD 31 — the most dangerous question (with ch. 8)

This is the search where self-deception lay closest at hand, and it is therefore the most strictly registered. Without an Adar II in AD 31 there is no 14 Nisan on 25 April, and the whole thesis falls. Precisely for that reason the criteria, the falsifier and the expected outcome were written down beforehand — with the agreement that an unwelcome result would simply be written down.

Fixed beforehand — what counts as evidence. The heaviest test is the Babylonian calendar, for the Jewish month names are Babylonian and the two calendars ran largely in step; Parker & Dubberstein reconstruct it to AD 75. An Addaru II in the Babylonian year beginning in the spring of AD 31 counts as a strong independent indication for; its absence as a strong indication against. In addition: the barley (was there ripe aviv at the end of March for the omer?), and the proxies (volcanism, fissure infill, tree rings).

Fixed beforehand — the limits. Two counter-examples stand here, and they stand here because they work against the thesis. First: the Jewish calendar in the first century was observational, not a computation (Strack: "a decision, not a calculation"); a Babylonian Addaru II therefore does not force a Jewish Adar II — it is an indication, not proof, and equally so the other way round. Second: the omer could if need be be fetched from further afield (m. Menachot 10:2), and in the Jordan Valley barley ripens weeks earlier than in the Judean hills; a ripe aviv at the end of March is therefore not unthinkable, which weakens the barley argument. Both reservations were noted before the search.

The falsifier, beforehand. The hypothesis (an Adar II in AD 31) is refuted by the conjunction of three finds: (1) a Babylonian calendar that has no Addaru II in AD 31, and (2) proof that barley in the Jordan Valley is routinely ripe at the end of March, and (3) the absence of any independent indication of a cold or late spring. If all three hold, the Wednesday is a guess and not a hypothesis.

Fixed beforehand — the expected outcome. Undecided. No proxy with a resolution coarser than one year counts as evidence for a single spring, and nothing hard was expected.

The outcome. Undecided, as foreseen. There is no primary source that records the weather or the intercalation of AD 31; the Babylonian rule makes an April moon plausible but does not force the Jewish intercalation; the barley objection (Jordan Valley) still stands.

The Adar II leg of the test, made explicit. The pre-registered rule ran: an Addaru II in the Babylonian year that begins in the spring of AD 31 = a strong indication for; its absence = a strong indication against. The actual state of affairs must therefore be named, even where it works against the thesis. Parker & Dubberstein show no leap month in AD 31: that year (S.E. 342) counts twelve months and 355 days, with Nisanu 1 on 12 April. The April placement of Nisan does not run through an Adar II of AD 31, but through an Ululu II — a second Elul — in the autumn of AD 30, which brought the preceding year to 383 days and so made Nisan of AD 31 fall in April. [DOCUMENTED — Parker & Dubberstein, SAOC 24, table p. 46, seen at first hand on the scan: the line for S.E. 342 reads 4/12.]

To the letter of the pre-registered rule the strong-indication-against is thus partly fulfilled at the Babylonian level: the leap month is absent in AD 31 itself. But it is not fulfilled at the level the thesis actually needs. A Babylonian Ululu II is not a Jewish Adar II — the Jews intercalated only Adar, never Elul — and the Jewish intercalation was in the first century a separate, observational decision of the court, not a reading-off of the Babylonian cycle (see E.2's own first reservation: Strack, "a decision, not a computation"). The absence of a Babylonian Addaru II in AD 31 therefore weighs as a real, named burden on the thesis, not as its breaking: it deprives the Wednesday of the strongest form of Babylonian support, while the Jewish Adar II remains an in-itself-possible, non-Babylonian decision that is neither confirmed nor refuted. The falsifier of this dossier moreover required the conjunction of three finds (§ above); that conjunction has not occurred. [CONSTRUED]

That is why the book carries the Adar II of AD 31 as CONSTRUED and not as DOCUMENTED, and calls it the narrowest link — exactly what an honest, pre-registered test ought to yield here. [CONSTRUED; the search has been carried out, the falsifier not fully met and the thesis thus not broken, but not established either.]

A second finding on this leg, and it cuts both ways. Robert Klever's reconstruction of the Jewish calendar for 63 BC – AD 72 makes the equinox rule depend on the position in the sabbatical cycle. In the sixth, seventh and first years of the cycle the omer may not fall before the equinox; because the seventh and first years are not intercalated, the intercalary month that the sixth year receives as a rule guards that boundary. In the second to fifth years no fixed rule applies, and the weather and the ripeness of the barley decide alone; the second year does carry a tendency towards intercalation in his model, to regain distance from the equinox, but no rule (pp. 65–72, esp. p. 65 with n. 252; p. 218). On Wacholder's placement of the sabbatical years — the placement Klever himself adopts (pp. 147–149, 217) — the spring of AD 31 is the third cycle year; on Zuckermann's it is the fourth. On both grids, and therefore on the grid Klever himself works with, the spring of AD 31 falls in the band where his model lays down no equinox rule. More than that: his source dossier contains no ancient witness he accepts for the strict reading of 1 Nisan on which an automatic April placement would rest; Philo's wording that Moses sets the first month "at the beginning of the spring equinox" he expressly declines to take literally (p. 58, n. 215). b. Sanhedrin 12b–13b (Mar Samuel, 3rd century) on his reading permits 1 Nisan up to fifteen days before the equinox, though he notes that he is not sure he has grasped the passage correctly (pp. 58–59, n. 215). His own criterion is the omer variant — 16 Nisan on or after the equinox (pp. 57, 65, 70–71) — which the March option already satisfies, its omer falling on 29 March against an equinox of 23 March. His note 254 adds that he suspects the barley "in most cases" was not ripe before the equinox (p. 65). His table does not turn that into a preference for March: for the spring of AD 31 it gives both moons as main options, 1 Nisan on 14 March (14 Nisan on Tuesday 27 March) or on 12 April (14 Nisan on Wednesday 25 April) (p. 275). On no reading within Klever's material does the equinox force AD 31 into April. That is a second, independent burden on top of the Babylonian one above, and it is entered here as such. The same table gives AD 30 and AD 33 a 14 Nisan on a Friday as main option, 7 April and 3 April, with the later moon only as a secondary option (pp. 275, 277); for AD 33 Klever thereby stands against the strict equinox rule that chapter 7 follows for that year (Saturday 2 May). That too is a burden for this book, and a heavy one: in Klever's reconstruction AD 33 also has a Friday.

What it yields for the thesis weighs just as much. First, a convergence that does not come from this book: the registered position on this leg is that the protection rests on the barley alone, and Klever's model says the same thing for a third cycle year, arrived at along a route of his own (p. 65). Second, an arithmetical constraint that is recomputable: without an intercalation in one of the two springs, Nisan of AD 32 would begin about 2 March and the omer fall on 17 March, before the equinox of 22 March — so one of the two springs, 31 or 32, must carry the intercalary month, and Klever's model decides that by the barley in both. His table follows that constraint: for AD 32 it knows a single 1 Nisan, 31 March or 1 April, and the moon of 2 March is Adar or Adar II there, depending on whether AD 31 was intercalated (p. 277). Third, the April moon of AD 31 stands in that table as a main option, not a secondary one: 1 Adar II on 14 March, 1 Nisan on 12 April, 14 Nisan on Wednesday 25 April (p. 275), the dates of chapter 7. Klever thereby defends no crucifixion year, no weekday and no thesis of this book. Thursday 26 April, the reconstruction an opponent has already advanced, appears in the table as 15 Nisan, not as 14 Nisan.

The balance. On balance, Klever's table moves the calendar for all three candidates at once. In an independent Jewish reconstruction the Adar II of AD 31 is no longer a marginal case but a main option beside the March moon; so are the Fridays of AD 30 and AD 33. The calendar thus excludes none of the three years and chooses none of them: the choice returns to the sign and the weekday (ch. 3, ch. 7). That does not make the Wednesday of AD 31 stronger than the Friday of AD 30 or AD 33, but it does make it equally admissible, and that is more than the Babylonian series alone showed. [CONSTRUED] [DOCUMENTED for Klever's wording and his table — Zur Chronologie der Geschichte Israels im Zeitalter des Zweiten Tempels, vol. I, half-vol. 1 (BoD, 2025; August 2026 version), consulted at first hand: pp. 57–72 (the rules), 147–149 and 217 (the sabbatical years), 214–219 (the explanation) and the table on pp. 274–277, the latter in the page image, because there the type size alone separates main from secondary options. CALCULATED for the Nisan-32 arithmetic and the month lengths.]

E.3 · The extra spring between John 2:13 and John 6:4 — the test that did not compel (with ch. 9, ch. 14)

The reading that inserts a spring between the Passover of John 2:13 and that of John 6:4 carries, on this book's route, the whole difference between AD 30 and AD 31: with the baptism in the autumn of 27, three Passovers give AD 30 and four give AD 31. It was therefore put to a registered test on the same pattern as E.1 and E.2.

Fixed beforehand — the criteria. Three criteria that would have to be met for the extra spring, three that would tell against it, written down before the sources were read. No criterion was adjusted along the way.

Fixed beforehand — the consequence clause. If the criteria for were not all met, the extra spring would be carried as possible, not compelled, and AD 31 would stand on this axis as admissible and not as indicated — with the same said of AD 30, which needs exactly three Passovers.

The outcome: possible, not compelled.

Two things were learned that the design had not expected, and both are entered as they fell. The seasonal reading of John 4:35 turned out stronger than expected — no four-month sowing proverb is attested anywhere, and the commentary tradition is divided without dividing in a way that settles it — so criterion 2 fails not because the seasonal reading is weak but because it does not exclude its rival. And the weakest limb turned out to be the order, not the season: Matthew places the plucking of the grain after the Baptist's question from prison, Mark and Luke long before it, so at least one of the three is not ordering chronologically here, and Mark 2:1–3:6 is a thematically arranged series of controversies.

Per the consequence clause, literally: on this axis AD 31 is admissible and not indicated — and AD 30 no more so. The axis is undecided and the decision falls elsewhere.

What it costs: the four-Passover reading may no longer be carried as compelled, and earlier versions of this book carried it more strongly than the test bears. What it does not cost: the reading is not special pleading; it is a respectable reading that merely does not force.

The commentary corpus, completed. The minimum corpus fixed beforehand has now been read on both verses. Not one of the five commentators reads John 5:1 as a Passover. Carson declines to decide — "the truth of the matter is that we do not know what feast John has in mind", and he notes in the same place that a Passover would give a ministry of about three and a half years and anything else about two and a half. Keener does not list Passover among the candidates at all (Purim, Pentecost, Tabernacles, Rosh Hashanah) and reads the feast as no more than the reason Jesus is in Jerusalem; his own note adds that Bowman's Purim reading "would fit between 4:35 and 6:4 (but only assuming a strict and unbroken chronology)". Hoehner prefers Tabernacles "with great hesitation". Godet calls Passover "positively excluded" and reads Purim; Wieseler likewise reads Purim, resting it on the sabbath of John 5:9. On John 4:35 the split is real: Keener prefers an otherwise unattested proverb, Carson notes that no such proverb is attested elsewhere and takes the four months as a genuine interval "by ordinary reckoning", and Hoehner, Godet and Wieseler read a season. [DOCUMENTED — Carson, 1991, and Keener, 2003, I:623–625 and I:634–635, both consulted at first hand; Hoehner, 1977, pp. 55–58; Godet, trans. Dwight, 1886, I:452–454; Wieseler, 1843, p. 219 (scanned page)]

⚠⚠ And here the corpus cuts both ways, which is why it is set out in full. Godet and Wieseler are the two clearest seasonal readers of John 4:35 — and both of them spend that reading in the opposite direction. Godet puts the conversation in December and Passover of 6:4 in April, concludes to Purim in March, and draws from it that "two years and a half are sufficient". Whoever cites them as support for the seasonal reading of 4:35 brings in, in the same breath, two witnesses who use it to shorten the ministry. That is not an argument this book may carry one-sidedly.

Hoehner in 1977 and in 2011, and the difference is instructive. His Brill entry (Holmén & Porter, 2011, pp. 2337–2339) has now been read at first hand, and he has not changed his mind: John 4:35 still places Jesus in Samaria in January/February, John 5:1 is still Tabernacles, the ministry is still three years. What changed is the argument underneath. In 1977 the proverb reading was named, attributed to Bernard, Barrett, Dodd, Brown and Lindars, and answered with three numbered grounds, among them Dalman's harvest months and m. Ta'anit 1:7. In 2011 that entire discussion is gone; the January/February dating stands there without a single footnote. And on John 5:1 the movement runs the other way: 1977 says "to identify this feast with certainty is impossible" and chooses Tabernacles "with great hesitation" and with "a slight edge"; 2011 says "it seems most likely" and drops both qualifications. The honest citation of Hoehner on 5:1 is therefore the 1977 page, not the later one — his own earlier formulation is the sharper counterweight to his later. [DOCUMENTED — Hoehner, 2011, pp. 2337–2339, and 1977, pp. 55–59, both seen at first hand]

⚠ What remains open. Barrett and Brown still stand only through Hoehner's own footnote 51 — a witness against his own interest, and therefore usable, but not their own pages. Origen's Commentarii in Joannem XIII has been read in the Greek (Preuschen's text): he does not pose the question of proverb or season at all, but his own phrase ἡ ἐλπιζομένη τετράμηνος, "the expected four months", presupposes a real term still to run. He cannot be carried as a witness for the proverb, and only weakly and indirectly for the season. [SECONDARY for Barrett and Brown; DOCUMENTED for Origen — Greek text, Perseus/Scaife]

E.4 · The same measure on the other side

The pre-registration also fixed that every demand made of AD 31 would be made of the rivals in the same breath. AD 33 requires that that year had no intercalary month (has anyone ever recorded that, or assumed it because it suited?); AD 30 requires a weekday that depends on the visibility criterion (Thursday or Friday — a choice, not a measurement). If the consensus does not account for its own calendar assumptions, then its certainty is a majority and not a measurement — and that, beforehand and afterwards, is the heart of the book.

⛔ And one undertaking belongs here, for it is the price of a freedom this book takes.

This book repeatedly uses an author's measurement without that author's verdict — Steinmann's count without his exclusion of AD 31 (ch. 9), Strashny's barley series within his own limitations (ch. 8), Stern's pp. 159–160 together with his judgement on the calculated scheme (appendix D.1), Schaefer's R values together with his own confidence ratio (appendix C), Milgrom's sentence together with the sentence behind it (ch. 8), and Newton's exclusion of AD 31 together with the half-sentence preceding it (ch. 1). The rule that permits this is written out in chapter 2: take the measurement where it rests on its own material, leave the verdict where it rests on other material, and name the difference.

What is recorded here is what that rule costs, so that it can be brought against this book. Whoever may use another's measurement without that author's conclusion gives the same right away. A reader may take over Strashny's dataset and leave chapter 8's reading of it; he may accept the crescent computation of appendix C and reject the calendar rule of chapter 8; he may use Parker and Dubberstein's tables and reject this book's choice of intercalation. That is permitted, and the book will not complain about it. ⛔ What the book must supply in return is that its own measurements be separable from its own conclusions — and that is exactly what the three marks exist for. A chapter in which `[COMPUTED]` and `[CONSTRUED]` cannot be pulled apart breaks this undertaking.

E.5 · Findings that returned in favour of the case

The deposited register of E.6 lists fourteen searches that returned against this book. A register that lists only those is not symmetry; it is a display of one kind of honesty, and a case that never records what came out in its favour books its profit in silence — the very thing this book charges against the field. Two findings ran the other way, and they stand here under the same rule that governs the others: on the record, with their counter-weight beside them, so that they cannot be quietly dropped if they later weaken.

1 · The reckoning of Tiberius' regnal years — and the same article against. Steinmann's survey of the first- and second-century literary, numismatic and epigraphic evidence finds no support for a count from the co-regency and none for an inclusive reckoning of Tiberius' years ("no evidence for inclusive reckoning of Tiberius's reign", 2022a, p. 115). AD 30 needs an early fifteenth year to reach a baptism in the autumn of 27 with exactly three Passovers; the finding therefore takes the reckoning away as a support for AD 30. It is favourable in that narrow sense and no wider, for it supplies no support here either: the AD 27 baptism of this book needs the same count, which chapter 9 carries as unattested and not refuted (ch. 14).

And in the same breath, the second argument of the same article, which is not favourable and which names this year: Steinmann concludes that the evidence "calls into question AD 30 or 31 (or any year before AD 30) as possible years of Jesus's crucifixion", and that the Gospel testimony for a ministry of about three and a half years "supports AD 33 as the correct year of the crucifixion" (2022a, p. 116). The two arguments rest on different material and neither derives from the other — the first is a count of coins, papyri and inscriptions, the second a claim about the length of the ministry that enters his concluding sentence as a premise (ch. 9). Whoever accepts the first is not bound to the second. That does not make the second go away, and it is registered against this book by name. [DOCUMENTED — Steinmann, 2022a, pp. 115–116, seen at first hand]

2 · A duration is not a numbering — and the offset stands beside it. The literary witnesses to Tiberius' reign (Philo, Josephus, Tacitus, Dio) give its length, some twenty-two and a half to twenty-three years. A length says how long a reign lasted; it does not say how a dating formula numbers the fifteenth year of that reign. The conclusion drawn from those witnesses against an inclusive numbering is therefore weaker than it is usually stated to be, and chapter 9 draws that distinction — and applies it against its own side as well, since the same witnesses are cited there for the epoch of 14. The offset, in the same breath: Antiquitates 18.106 has a Jewish narrator numbering a Roman emperor's regnal year non-inclusively, and that is the nearest thing in the whole dossier to the case this book would need. The finding removes an argument; it does not supply one. [CONSTRUED for the distinction; DOCUMENTED for Ant. 18.106 — ch. 9, with Niese's apparatus]

The asymmetry of the count is itself a finding and is not smoothed over: two entries here against fourteen there. Nothing has been added to this rubric that has not been measured for this book, and a favourable finding whose counter-weight cannot be stated beside it does not enter it.

E.6 · The deposited register

A pre-registration verified only by its author's word is not a pre-registration. For the tests of E.1, E.2 and E.3 that objection stands, and E.7 leaves it standing. For the falsification criteria of the day-axis it no longer does.

The falsification criteria for the weekday, together with the year-axis appendix, are deposited publicly and with a digital object identifier: doi:10.5281/zenodo.22042826, versions 1 to 3, each version remaining visible and citable beside the next. The deposit was made before the weighting study for which the document was written had been run, and before any result of it was known here.

What stands in it: seven falsifiers on the day-axis, each with what would count, what would not, and the consequence committed to if it is met — up to the withdrawal of a chapter and of the claim itself; a closed list of what would count as confirming; an advance ruling on the three sources an opponent is most likely to submit, so that adjudication cannot happen after the fact; the fourteen searches already run that returned against the case; and an ordered, exhaustive table of outcomes with the commitment attached to each.

The year-axis is registered there but not under test, in six entries:

Two things are stated there that cost: that this book's own scoring model, under the premise the correspondent holds, predicts the rival result (appendix F); and that a favourable outcome on the weekday would not establish the year.

E.7 · What this appendix does not deliver

- A pre-registration is no guarantee against error, only against self-deception after the fact. It binds the author to his own criteria; it does not by itself make the criteria correct. Whoever disputes the criteria themselves reads the search and judges anew. - The outcome of E.2 is and remains undecided. That is not rhetorical modesty but the literal state of play: the Wednesday rests on this link as a hypothesis, not as an established fact, and the book says so throughout. - The outcome of E.3 is possible, not compelled, and its corpus is now complete but its edges are not. All five commentators fixed in the minimum corpus have been read on both verses. What remains outside: Barrett and Brown stand only through Hoehner's own footnote 51 and not on their own pages, Godet only in the English translation and not in the French, and Wieseler only on his page for John 5:1 and not on those for 4:35. The test may not be booked as finished until those edges are closed. - The timestamp covers the weekday register, not every test in this appendix. The deposit of E.6 gives an independent date to the falsification criteria for the day-axis and to the year-axis appendix. That the criteria of E.1 and E.2 were fixed before those searches has no such external seal, and rests on the author's word. The value of a pre-registration lies in its content — that the falsifier is sharp and testable — not in a seal of approval.