The corpus sweep of the double construction, and its base rate

This appendix makes the book's only true measurement recomputable: the claim (ch. 3) that the double construction "three days and three nights" is never counted inclusively anywhere in the searched corpus. It contains the search itself, the counter-tradition that articulates it, and the binomial base rate that gives it weight. The binomial computation can be reproduced with a small Python script.

B.0 · The question, put sharply

The ordinal "on the third day" demonstrably counts inclusively in Judaism: a part of the first day counts as a whole one (Jerusalem Talmud, Shabbat 9:3, the "onah" rule; ch. 3). The book's question is narrow: does the double construction — which names the nights separately — also ever count inclusively, or does it always name full days? If it ever counts inclusively, the sign says nothing; if it never does, it is a measure.

This is an empirical question, and it can be tested by sweeping the corpus for every place where the double construction describes a factual duration, and seeing whether that duration is ever shorter than the number of full days named.

B.1 · The search — registered in advance, then carried out (Test A)

The prediction was fixed before the search: if the double construction is a measure, then every narrative duration statement made with it is full/continuous, and the falsifier (a single inclusive case) will not be found. Only then were the held-back corpora searched. [CALCULATED/DOCUMENTED — seen in the primary sources]

Every narrative duration statement — the siege of Tur Malka, the ship and the fish, the lament, the furnace of Abraham, the fasts — counts full/continuous. Within the corpus, not a single inclusive use of the double construction has been found. The falsifier was sought and not encountered — with one honesty added, for the candidate has been put forward and deserves mention: outside the corpus it does exist. Augustine argues for the inclusive count in the fourth century, in Latin (§B.3c). He is an interpreter who argues for it, not a source that uses the idiom that way — and precisely for that reason the falsifier is formulated as it is. (The limit of this: the corpus is large but not exhaustive; see B.5.)

One place counts here toward neither side: 1 Samuel 30:12. It carries the double construction, but it is the only place where the duration cannot be checked independently. The single internal measure — verse 13, "today three" (hayyom sheloshah) — is a vague range, not a second measurement, and under every reading it is compatible with the exact statement of verse 12; nothing is compressed (the full analysis, with the cardinal "three" against the ordinal of verse 1, the parallel in 1 Samuel 9:20, Driver, and the independent confirmation by all four ancient versions — Targum Jonathan yemamin/yomin, Peshitta imamin/yawmin, Septuagint τρεῖς ἡμέρας καὶ τρεῖς νύκτας / τριταῖος, Vulgate tribus diebus et tribus noctibus / nudiustertius —, stands in ch. 3). For that reason 1 Samuel 30 is not a calibration point: it neither confirms the full count nor refutes it. It has not been counted above, nor in the base rate of §B.4, as a confirming case; the loss of this one borderline case falls within the 7–11 band with which §B.4 already reckons conservatively, and does not weaken the signal.

B.2 · The counter-tradition articulates the rule itself

The strongest point is not that the hits are silent, but that the Jewish tradition states the rule explicitly — and precisely at the place where the opposing side believes it holds its strongest card, Esther 4:16. [DOCUMENTED]

- Menot haLevi (Solomon Alkabetz, 1529) and Maharal, Ohr Chadash on Esther 4:16: had the actual fast been 72 hours, the verse would have had to say "שלושה ימים ושלושה לילות" (cf. 1 Samuel 30:12); it says "לילה ויום" (night and day) — therefore no 72 hours. The tradition avoids the double construction precisely because the fast was inclusive. - Ben Yehoyada (Yosef Chaim of Baghdad): "three days and three nights" = 72 hours (gematria חסד). - Midrash Tehillim 22 (before 1500): Esther broke off "while it was still day" — the actual fast was therefore inclusive.

Thus the Esther "falsifier" turns into confirmation. The tradition treats the double construction precisely as a measuring rod: where it stands, the duration is full (cf. 1 Samuel 30:12); where it is avoided, it is not. And Esther 4:16 avoids it — it reads not "three days and three nights" but "לילה ויום", "night and day" — while the actual fast ends "on the third day" (Esther 5:1). So Esther does not break the rule; Esther shows the rule existed and was known, and that one deliberately knew to avoid it where the duration was inclusive. The distinction between the full measure and the inclusive use is drawn by the tradition itself (ch. 3).

B.2b · The adverbial form — and the limit it sets on my own claim

Esther 4:16 is not the only verse with this build, and the second case is the one that binds me.

Leviticus 8:35 reads "וּפֶתַח אֹהֶל מוֹעֵד תֵּשְׁבוּ יוֹמָם וָלַיְלָה שִׁבְעַת יָמִים" — "and at the entrance of the tent of meeting you shall remain, day and night, seven days." [DOCUMENTED — the Masoretic text seen primarily] Day and night are named explicitly here, and yet to the seven days derived from this verse — the seven days of mourning (y. Moed Katan 3:5; Bereshit Rabbah 100:7) — the part-for-whole rule miktsat hayom kekhulo is applied without difficulty.

What remains is narrower and grammatical — it turns on the repetition of the numeral:

[DOCUMENTED — all fifteen places seen primarily. The count itself is CALCULATED and is a claim about this corpus, not a measurement of Biblical Hebrew.]

⚠ And the objection that belongs here: two cases are few on which to build a class, and the adverbial row is therefore too thin to carry anything on its own. It does not need to. Its function in this argument is purely limiting: it shows which claim is too broad, and it denies the opposing side the option of entering Esther 4:16 as an instance of the double construction. The load is carried by the thirteen.

B.2c · The defenders and their proof texts — one basis of two verses

The defenders of the inclusive reading are usually cited as a series. Set their proof texts side by side and something else appears.

Baron limits himself on each of his three: of Jonah he says "we have no means of accurately ascertaining" how long it lasted; 1 Sam. 30 requires on his reading that verses 12 and 13 measure one span, on the ground of "the impression left on the mind"; and Esther 4:16 he writes out as "three days—night and day" — the adverbial form, reproduced verbatim and not recognised. Of his three places, then, one actually uses the form: 1 Sam. 30:12.

⭐ And there is a turn that comes from Bullinger himself. In the Companion Bible, appendix 144, he writes: "when the number of 'nights' is stated as well as the number of 'days', then the expression ceases to be an idiom, and becomes a literal statement of fact." That is word for word the distinction of §B.2b — the numeral with day and with night. Bullinger thus states the very criterion on which his own Esther example fails, for Esther 4:16 states no number of nights. ⚠ Source status: this appendix has been seen as a transcription, not in the printed Companion Bible, and the posthumous editorial history of the appendices (Bullinger d. 1913; the final volume appeared in 1922) has not been established. The observation therefore stands here as [DOCUMENTED — transmitted via transcription], and it carries nothing that §B.1 does not already carry.

⚠ The limit of this section, and it is emphatic: this concerns the defenders read for this book. It is not a statement about the literature as a whole, and anyone who points to a fifth defender with a fifth proof text simply adds a row to the table above. That is exactly what appendix E records as a standing invitation.

B.3 · The hardest support: a pre-Christian calendar computation (4Q252)

The Qumran flood text 4Q252 works the "forty days and forty nights" out explicitly in the calendar; the three independent calibration points in that computation close only if they are forty full days (García Martínez & Tigchelaar, 1997–1998). Here the double construction is not merely presumed full but made recomputably full — before the Christian era, and outside any Christian interest. [DOCUMENTED]

B.3b · And the rabbinic tradition computes it just as exactly — the zero is not empty

A sweep that finds only silence is weak; it can always rest on a gap in the corpus. That is why it counts double that the tradition also computes the double construction positively, and that the outcome of that computation can be checked independently.

Seder Olam Rabbah, chapter 6 computes Moses' stay on Sinai: forty days and forty nights, and then once more forty days and forty nights. The midrash counts those periods exactly — full days, nothing rounded — and thereby arrives at the tenth of Tishri as the day on which Moses descends with the second tablets: the Day of Atonement. That is not a claim but a sum with a checkable endpoint: whoever counts the forty days and nights inclusively misses Yom Kippur. And the limit right with it, for ch. 3 draws it itself: Seder Olam counts whole days and counts the first day in — that is an inclusive count of endpoints, and not the question at issue with the sign, whether a half day counts as a whole. On that question this source is silent. The double construction functions here as what it is — a measure on which one dares to build a calendar. [DOCUMENTED; the computation is recomputable]

And the rule is also stated, by a Jewish authority, at precisely this construction. Rashi notes on Genesis 7:12 and Exodus 32:1 that where a day without its night is meant, that appears from the absence of the doubling — "אין לילו עמו", "its night is not with it". For him the doubling is thus the marker that the day is full; it is read not as ornament but as measure. Honestly added, and ch. 3 says it too: Rashi has to add the word shelemim, "full", to it in order to cut out the partial day — the form itself does not compel that. He testifies that the construction was read that way, not that it could not be read otherwise. [DOCUMENTED — the quotation; CONSTRUED — the inference]

With that the nature of the finding changes. The sweep not only finds no inclusive use; the tradition computes the construction where it meets it, and it names why. A silence can be a gap; a computation with a recomputable outcome is a measurement.

B.3c · The honest exception — and it is Latin, not Hebrew

One witness does count the double construction inclusively, and it must stand here, for a critic finds it within a day. Augustine, in De consensu evangelistarum III.24.66, applies the principle a parte totum — the part for the whole — expressly to "three days and three nights", and so arrives at the Friday. The claim "the double construction is nowhere counted inclusively" is therefore, in that general form, false, and this book does not pretend otherwise. [DOCUMENTED, read in the Latin]

But see where he stands and what he himself says. Augustine is a Latin witness from the late fourth century, and he does not speak of Hebrew or Aramaic usage but of the harmonisation of the Gospels; he thereby falls outside the corpus that §B.1 searched, and he cannot fall within it either. And he himself admits that the solution does not close: it comes out here, he writes, "non ad liquidum" — not clearly. Whoever brings him forward as counter-evidence brings a fourth-century Latin exegetical intervention against a measurement in the Hebrew and Aramaic corpus. That is allowed, but it is something other than a counter-example.

And a second name belongs with it, for it shows how the matter really stood in that century. Tyconius, in the fifth rule of his Liber regularum, "De temporibus", discusses the same question, and he does so as someone arguing against an existing strict reading: of the three days, he says, "solus medius plenus" — only the middle one full — and the others were "sed dies duo". He has to make that effort because the strict count is there, and he names its consequence in so many words: then "erit dominicus dies quartus" — then Sunday is the fourth day. The loose reading is here not the self-evident Jewish convention but a patristic intervention that had to be brought against the strict count. [DOCUMENTED]

So it stands honestly: there is an inclusive count of the double construction, it is Latin and late fourth-century, its own author calls it inconclusive, and a contemporary testifies that it had to defend itself against the full count. That does not lower the value of the zero in §B.1; it places it.

B.4 · The base rate — how much does "zero inclusive" weigh?

Grant the opposing side its strongest form (the idiom null): the double construction was a flexible idiom which, like the ordinal, could be counted inclusively in about half of its uses. How improbable is it, then, to find it inclusive zero times in n independent narrative uses? The probability is (1 − p)ⁿ. The number of genuinely independent narrative duration statements I estimate conservatively at between 7 and 11. [CALCULATED]

And here a correction to this book belongs, with the reason why it changes nothing. The list of biblical places in ch. 3 stood at twelve and should have stood at thirteen: Deuteronomy 10:10 — arba'im yom we'arba'im laylah — was missing from it. The omission has been repaired, and the sweep redone over the whole consonantal text (ch. 3). But n does not move with it, and it would be arithmetic sleight of hand to let it. n does not count occurrences of the idiom; it counts independent durations. Deuteronomy 10:10 is a further mention of Moses' stay on the mountain, alongside Exodus 24:18 and 34:28 and the four places in Deuteronomy 9 — the same event or the same series, not a new one. The Sinai passages together supply, as before, two or three independent durations, not seven. So the base rate stands unchanged, and the correction adds an occurrence to the list without adding a case to the count.

Read the table as a sensitivity analysis, not as a single number. Under the neutral idiom null (p = 0.5) the p-value lies between 0.008 and 0.0005 — a strong signal. Even under a reserved null (p = 0.3, inclusive counting in only three cases out of ten) it stays between 0.08 and 0.02 — marginal to significant. On the book's fixed p-value scale (< 0.01 strong; 0.01–0.05 significant; 0.05–0.10 marginal) this is the only pillar that delivers a strong signal on its own.

What the table does not give is the other side of the same zero: how high can the inclusive fraction still be, given that it was found zero times in n cases? That upper bound belongs here, because without it "zero inclusive" reads as "never", and that is not what the measurement says. For zero successes, the one-sided 95 % upper bound (Clopper–Pearson) and the posterior mean under a flat prior are:

[CALCULATED] ⚠ Read this too as a series and not as a single number — B.4 works with the same band 7 · 9 · 11, and naming one upper bound without its n is the mistake this appendix avoids elsewhere.

And one thing about the rule all of this turns on, because it belongs in the weighing. Miktsat hayom kekhulo is no dead rule; it lives, and it can be located. It dwells in the law of mourning, in the periods of purity, in the widow's waiting time and in the agricultural terms — in short in status periods, where a day opens or closes a term. In narrative statements of duration it does not occur, and with the double construction zero times. That is not silence for want of material: it is a rule that appears everywhere on its own ground and nowhere on this one. [DOCUMENTED]

B.5 · What this appendix does not deliver

- The corpus is not exhaustive. Sefaria and Maagarim cover the classical rabbinic corpus and a large part of the medieval Hebrew/Aramaic, but not every manuscript, and not every spelling variant of the double construction was searched separately. The claim is "not found where I looked thoroughly", not "exists nowhere". - The "effective n" is an estimate, not a count. The 14 + 22 raw hits contain repetitions, quotations and prescriptions; how many genuinely independent narrative duration statements there are is a judgement (7–11). The base rate is therefore a band, not a point. - The null p = 0.5 is a choice. It is maximally favourable to the idiom reading; whoever finds it too high reads the table at p = 0.3 or 0.4. The signal weakens then, but does not vanish. - The articulation is partly late. The explicit formulation (Menot haLevi, Maharal) is sixteenth-century; it builds on older midrash (Tehillim 22, before 1500) and on 4Q252 (pre-Christian), but the sharpest wording is not ancient.

Reproduction: a small Python script prints the binomial computation. The corpus hits were consulted at first hand via Sefaria and the Maagarim text corpus (Academy of the Hebrew Language).