Ars Inquirendi

← All conjectures · Jewish book cultures

The order the forgers had to invent

Status: Falsified

The verdict’s fine print — quoted from the resolution record: “Killed on the registered unweighted-means ratio: 0.524 against SUPPORTED-iff-<0.2; the reported-only pooled alternative (0.725) is weaker still, so no aggregation choice rescues it.” Read the full caveats ↓

Status is derived only from the shepherd-authored triage/prediction data above -- community submissions and claims are a separate overlay and can never change it (see the participation panel below).

This is a conjecture imagined by a language model — drawn from its trained weights and held to falsifiability, novelty, and plausibility, not to any one method: it may join two or more fields, or none. It is not an article and not evidence: it sits below the evidence/publication boundary. A quantitative prediction and a named kill-dataset are attached (when registered) so the claim stays falsifiable rather than merely evocative.

Claim (verbatim)

The Palestinian Talmud (Talmud Yerushalmi) as transmitted carries gemara on only four of the Mishnah's six orders - Zeraim, Moed, Nashim and Nezikin - plus tractate Niddah; the entire order of Kodashim (the Temple-service tractates Zevahim, Menahot, Hullin, Bekhorot, Arakhin, Temurah, Keritot, Me'ilah, Tamid) has no Yerushalmi at all, and most of Tohorot is likewise absent. Two rival explanations have contended for a millennium: that a Yerushalmi Kodashim was composed and then lost, or that the Palestinian schools, cut off from a functioning Temple and dwindling under Christian Rome, never redacted it. The decisive evidence is citational: if the order once existed, the rishonim with access to a fuller Yerushalmi would quote it on Kodashim sugyot; if it never existed, their "Yerushalmi" citations must dry up precisely where the extant text stops. The cautionary control is the early-twentieth-century Friedlander forgery, which manufactured a "Yerushalmi" on Hullin and Bekhorot from scratch - a fraud that was only necessary, and only briefly plausible, because the genuine citation trail on those tractates is so nearly empty. Mechanism: a book leaves quotation-fossils in the commentators who used it, and the absence of such fossils on Kodashim is itself the datum. Prediction restated: in Sefaria's Tosafot and commentary corpora the per-daf rate of "Yerushalmi" citations on Babylonian tractates of Kodashim falls to under a fifth of the rate on tractates whose Yerushalmi survives - the signature of a text that was never there to quote.

Prediction clause (verbatim)

Prediction: string-counting the citation lemma "ירושלמי" (Yerushalmi) in Sefaria's Tosafot and base-commentary layers on the Babylonian Talmud, the mean per-daf occurrence rate on tractates of the order Kodashim (Zevahim, Menahot, Hullin, Bekhorot, Arakhin, Temurah, Keritot, Me'ilah, Tamid) will be below one-fifth (a ratio < 0.2) of the mean per-daf rate on tractates whose Yerushalmi is extant (the orders Moed, Nashim, Nezikin) - primary clause: the < 0.2 Kodashim-to-extant per-daf Yerushalmi-citation ratio; the verdict follows it. The test voids for coverage if fewer than 100 total "Yerushalmi" citations are recovered across the sampled tractates.

Kill-dataset (verbatim)

Kill: Sefaria's open Tosafot and commentary corpora on the Babylonian Talmud - string-match the "ירושלמי" citation formula, bin each hit by the tractate and Mishnaic order of its host daf, and compare per-daf rates for Kodashim against the orders whose Yerushalmi survives; the Venice/Vilna Yerushalmi and the Academy of the Hebrew Language critical text confirm which tractates are genuinely absent, and the Friedlander Hullin/Bekhorot volumes are flagged as forgery, not witness.

Provenance

Run: Fresh agent generation · model: claude-fable-5

Fresh blind generation by claude-fable-5, 2026-07-17, Jewish text-culture wave instrument-anchored on the open Sefaria corpus and its cross-reference link data, with standard critical editions as controls: every kill names a real corpus and a countable operation (coverage maps, citation-formula counts, link-orphan shares, digest-fraction, recension divergence, citation-decay), thresholds far from 1 with explicit coverage guards distinguishing what Sefaria holds from what existed. Ground is CITATION-GEOMETRIC and disjoint from the 2026-07-10 w19 Jewish wave, which was material-culture-of-loss (colophons, parchment, genizah, masora, binding fragments): no material-culture re-posing here. Two candidates were dropped after a grep of all fresh packets: a Tosafot-density-by-tractate survival item (pre-empted by w02-philosophy #18, which already correlates per-folio Tosafot density with manuscript survival), and a piyyut liturgy-vs-anthology survival item (pre-empted by w01-literature #28 'Liturgy out-survives fame', keyed to Davidson's Thesaurus). The Mishneh Torah item here tests EXTERNAL source-attribution erasure vs the Tur (loss of source-geometry), deliberately distinct from minds-w02 #25, which tests the code's INTERNAL recall/promise cross-references. Two items are marked Kill (not yet built) where the deciding corpus (Kohut's Arukh apparatus; Lewin's parallel Iggeret recensions) is not yet digitized.

Novelty / leakage triage

anticipated in the literature — this exact test has never been run

The absence of a Yerushalmi on Kodashim and the spuriousness of purported witnesses to one is the published consensus: Strack-Stemberger states the missing orders and recounts the Friedlander 'Yerushalmi Kodashim' forgery, Sussmann's studies and his introduction to the Academy of the Hebrew Language Yerushalmi hold the order was never redacted, and Ratner's Ahawath Zion we-Jeruscholaim collected the rishonim's Yerushalmi citations tractate by tractate. But the instrument counts every 'Yerushalmi' lemma on Bavli Kodashim dapim, including genuine cross-order citations (Tosafot on Hullin or Zevahim do adduce Yerushalmi from Zeraim and Moed), so the per-daf rate on Kodashim is nonzero and uncounted, as is the extant-order baseline. The printed canvasses guarantee the direction, not the < 0.2 ratio; the specific arithmetic over Sefaria's commentary layers is un-run.

Sources cited by the triage
  • H. L. Strack & G. Stemberger, Introduction to the Talmud and Midrash (trans. M. Bockmuehl, 2nd ed., Edinburgh, 1996), chapter on the Yerushalmi (missing orders; the Friedlander forgery)
  • B. Ratner, Ahawath Zion we-Jeruscholaim, 12 vols. (Vilna, 1901-1917)
  • Talmud Yerushalmi According to Ms. Or. 4720 (Scal. 3) of the Leiden University Library, with introduction by Y. Sussmann (Jerusalem: Academy of the Hebrew Language, 2001)

Predictions

Killed registered 2026-07-19 variant: v1 calibration prediction (parent triage: leaked/adjacent)

Resolution: Killed

Caveats: Killed on the registered unweighted-means ratio: 0.524 against SUPPORTED-iff-<0.2; the reported-only pooled alternative (0.725) is weaker still, so no aggregation choice rescues it. The DIRECTION held solidly -- Kodashim tractates cite 'Yerushalmi' at roughly half the rate of tractates whose Yerushalmi survives (0.33 vs 0.63 per-daf) -- and the registrant's openly-disclosed confound explains the gap between direction and threshold: the hit definition counts EVERY Yerushalmi lemma, and the Tosafot's cross-order citation practice floods the Kodashim numerator (Menachot 0.65, Bekhorot 0.55 per-daf; the concordance sample catches a Yerushalmi Berakhot citation inside Bava Batra in the act). The registration called this confound direction conservative-against-support ex ante, and it was. CALIBRATION: another multiplicative-separation death (<0.2 = a 5x separation wagered; world gave 1.9x), joining the dead class exactly where the registration's own calibration note placed the risk; the categorical mechanism (a corpus that never existed) did not produce categorical separation because the MEASUREMENT is of citation habit, not corpus existence. Blinding note: the compute worked under opaque A/B labels with no thresholds -- the strongest firewall of the batch.

Registered against the Sefaria API (www.sefaria.org/api/... — known-PERMITTED instrument, the house's proven route for five prior resolutions), text layers 'Rashi on <Tractate>' and 'Tosafot on <Tractate>' of the Babylonian Talmud. PREDICTION VERBATIM: "Prediction: string-counting the citation lemma \"ירושלמי\" (Yerushalmi) in Sefaria's Tosafot and base-commentary layers on the Babylonian Talmud, the mean per-daf occurrence rate on tractates of the order Kodashim (Zevahim, Menahot, Hullin, Bekhorot, Arakhin, Temurah, Keritot, Me'ilah, Tamid) will be below one-fifth (a ratio < 0.2) of the mean per-daf rate on tractates whose Yerushalmi is extant (the orders Moed, Nashim, Nezikin) - primary clause: the < 0.2 Kodashim-to-extant per-daf Yerushalmi-citation ratio; the verdict follows it. The test voids for coverage if fewer than 100 total \"Yerushalmi\" citations are recovered across the sampled tractates." FULL OPERATIONALIZATION PINNED: LAYERS: for each tractate T, the layer set = the Sefaria indices titled exactly 'Rashi on T' and 'Tosafot on T' ('base-commentary' = Rashi, the printed base layer; Sefaria's titling is followed mechanically, so pseudo-Rashi layers printed and titled as Rashi count as registered). Instrument-existence fact verified at registration from a single /api/index table-of-contents call: both layers exist for every pinned tractate EXCEPT Tamid, which has neither 'Rashi on Tamid' nor 'Tosafot on Tamid' — Tamid stays in the prediction's verbatim Kodashim list but contributes no layer text on this instrument and is excluded from the Kodashim mean with its absence reported (effective Kodashim n=8). TRACTATE GROUPS PINNED: Kodashim = ["Zevachim", "Menachot", "Chullin", "Bekhorot", "Arakhin", "Temurah", "Keritot", "Meilah", "Tamid"]; extant-Yerushalmi group (verbatim 'the orders Moed, Nashim, Nezikin', realized as the Bavli tractates of those orders): Moed = ["Shabbat", "Eruvin", "Pesachim", "Rosh Hashanah", "Yoma", "Sukkah", "Beitzah", "Taanit", "Megillah", "Moed Katan", "Chagigah"] (Shekalim excluded — it has no Bavli gemara; the text printed in the Bavli cycle is the Yerushalmi's own), Nashim = ["Yevamot", "Ketubot", "Nedarim", "Nazir", "Sotah", "Gittin", "Kiddushin"], Nezikin = ["Bava Kamma", "Bava Metzia", "Bava Batra", "Sanhedrin", "Makkot", "Shevuot", "Avodah Zarah", "Horayot"]. HIT DEFINITION: over the concatenated Hebrew text of a tractate's layer set (default Hebrew version returned by GET /api/texts/<ref>, versionTitle recorded), hits(T) = substring count of 'ירושלמי' PLUS substring counts of the abbreviation forms 'ירוש' immediately followed by one of the three abbreviation marks {U+0027 apostrophe, U+05F3 Hebrew geresh, U+2019 right single quote} where the match is not part of a 'ירושלמי' match (prefixed forms בירושלמי, הירושלמי etc. are caught by substring matching; the mark requirement keeps ירושה/ירושלים out). DENOMINATOR: amudim(T) = number of top-level sections of the BASE Talmud text index for T as returned by the API; dapim(T) = amudim(T)/2. The identical convention applies to every tractate, so any daf-vs-amud convention choice cancels in the cross-group ratio. RATES: rate(T) = hits(T)/dapim(T); group mean = UNWEIGHTED arithmetic mean of rate(T) over the group's tractates possessing at least one pinned layer (pinned realization of 'mean per-daf occurrence rate on tractates'; the pooled alternative sum(hits)/sum(dapim) is computed and reported, non-verdict, stated ex ante). RATIO = mean_Kodashim / mean_extant.

Resolution criteria — the registered fine print

Resolution criteria: COMPUTE RECIPE (blind agent): (1) GET /api/index existence check for the 70 pinned layer titles; (2) for each existing layer, fetch the full Hebrew text via the API (whole-node or per-section calls, >=1s spacing; record versionTitle per layer); (3) count hits per the pinned definition; (4) fetch each base tractate's index for amudim(T); (5) compute rate(T), the two unweighted group means, and RATIO. CLAUSE PRECEDENCE: (1) INCONCLUSIVE_BY_ACCESS if any pinned layer that exists in the index cannot be fetched in full, or any base-tractate section count is unavailable; (2) INCONCLUSIVE_BY_DESIGN if total hits across all sampled tractates (both groups, all fetched layers) < 100 (the prediction's own void, quoted verbatim); (3) SUPPORTED iff RATIO < 0.2; (4) REFUTED otherwise. REPORTED, NON-VERDICT: per-tractate hits, dapim, and rates; the pooled-ratio alternative; the abbreviation-form share of hits (ירוש'-type vs full ירושלמי); Tamid's layer absence; per-layer versionTitles; and a 30-hit random concordance sample (40 chars context) for face-validity audit. Compute firewalled from Ratner's Ahawath Zion we-Jeruscholaim (the printed tractate-by-tractate Yerushalmi-citation canvass), from the house's prior Sefaria computes, and from this batch's other rows.

Known-priors disclosure — what the registrant already knew

Known priors disclosure: Triage 2026-07-17 graded ADJACENT: that Yerushalmi Kodashim never existed (and its purported witness was the Friedlaender forgery) is published consensus — Strack-Stemberger; Sussmann's introduction to the Academy of the Hebrew Language Yerushalmi; Ratner's Ahawath Zion we-Jeruscholaim collects rishonic Yerushalmi citations tractate by tractate — but the triage explicitly notes the instrument counts EVERY Yerushalmi lemma on Kodashim dapim, including genuine cross-order citations (Tosafot on Hullin and Zevahim do adduce Yerushalmi from Zeraim and Moed), so the Kodashim rate is nonzero-by-mechanism and the <0.2 ratio is a live, un-run wager, not a tautology. INSTRUMENT-EXISTENCE PROBE DISCLOSED: one GET /api/index call (the full public TOC) verified layer existence — 34 of 35 pinned tractates carry both layers, Tamid carries neither; NO text, hit counts, or lengths were fetched at registration, so no effect quantity was glimpsed. SCOUTED MASSES: none beyond layer existence — daf masses are derived in-run from the API's own section counts; the >=100-hit coverage floor is the prediction's own and its mass cannot be pre-scouted without counting effect hits (stated per lesson 5; qualitatively the registrant expects Tosafot's Yerushalmi citations to run to many hundreds corpus-wide, from the standing scholarship, with no number known). HOUSE-PRIOR DISCLOSURE: five house Sefaria computes exist (Mekhilta coverage, Yerushalmi Nezikin segment lengths, targum expansion, Rif coverage, MT/Tur citation density); none touched Tosafot or Rashi citation lemmata; no effect quantity for this item exists in-house. CALIBRATION CLASS NOTE (binding lesson 6): the <0.2 primary is a MULTIPLICATIVE-separation pin — the historically dead class (nearest kin: the <0.67 Bavot-segment pin died at 0.87); registered faithfully per the anti-softening rule; selection accepted the risk because the mechanism (a wholly absent target corpus versus merely cross-order echo) is categorical rather than marginal, unlike the dead pins, which compared two existing corpora. The registrant states the confound direction openly: cross-order citation traffic can only RAISE the Kodashim numerator, i.e. push AGAINST the conjecture — the faithful test is conservative-against-support, not rigged toward it.

Method and dataset — how it was measured

Firewalled blind compute (claude-sonnet-5) fetched and counted under NEUTRAL group labels (no direction inferable) per the pinned hit definition (full lemma + mark-form abbreviations; 1,220 full / 8 mark-form) and amudim/2 denominators; all group aggregation performed at verdict time by claude-fable-5 per the registered operationalization (UNWEIGHTED tractate-rate means; pooled alternative reported-only). Clause precedence: access passed, coverage passed, SUPPORTED iff mean_Kodashim/mean_extant < 0.2, else REFUTED.

Dataset: Sefaria API, Rashi+Tosafot layers on 34 of 35 pinned Bavli tractates (Tamid carries neither layer, excluded per registration; effective Kodashim n=8). 1,058 API calls, zero failures; 1,228 total lemma hits vs the registered 100-hit coverage floor (12x clearance). versionTitles recorded (Vilna Edition throughout).

computed 2026-07-19

Weigh in

No community feedback yet.

New here? Create an account first

Create an account or sign in and your feedback is tied to you — you can track it, get replies, and claim this conjecture so others know you’re working on it. Prefer not to? Just leave your take below as a guest — only the name you type is shown.

Add your take

Posted immediately (spam is removed). Community feedback is never an adjudicated verdict and never changes this conjecture's triage label or status above.