A genre-entropy gap between Ur III and Old Babylonian cuneiform tablets

Provenance notice. This article exists because a conjecture — a falsifiable public prediction, staked in advance against a named dataset — was pre-registered (its clause precedence fixed and published before the genre-by-period comparison was run) and supported, on the Inferpedia conjectures campaign's 1001-blind lane. The conjecture that produced this finding, fresh-fablemax-20260705-008, "The bureaucratic entropy floor", remains supported and untouched at its own L1 lead status; nothing here reopens or revises it. Unlike most findings promoted from this lane, this one is NOT calibration-flagged: the campaign's literature search located raw Ur III genre-share figures but no prior entropy computation or period comparison, so the triage verdict is "no prior formulation located" rather than "adjacent" or "leaked." That is not a novelty claim about the underlying historical reality — only about whether a stated formulation of this specific test was found in a dated search — and this article makes none. This article was published in beta after a recorded Fable publication judgement (docs/generated/publication_judgement_614_20260716.json); every figure was recomputed from the committed records, and once more from the live catalogue, before publication.

Epistemic status

This article reports a measured pattern in surviving records: the numbers are directly attested in the cited datasets; what they mean is an open interpretation.

That line is the standard notice for every "measured_finding" article on this site, and it is worth taking literally here: the two entropy values this article rests on — 0.1961 bits and 2.4879 bits — are not inferred from indirect traces. They are direct computations over CDLI's own genre labels, following a clause-precedence rule fixed before the computation ran. What is inferred is smaller: that the size and direction of the resulting gap is a real property of this catalogue (not a fluke of one or two large collections), and that its cause is NOT settled by this measurement. The conjecture's own proposed mechanism — "strong states make boring archives" — is one reading of the gap; a rival reading, that Ur III's administrative dominance is itself partly a modern discovery-and-publication artifact, is equally live, and this article does not adjudicate between them.

Summary

Among Ur III-period cuneiform tablets catalogued by CDLI (59,455 genre-labelled, 237 unlabelled), the Shannon entropy of genre composition — how evenly writing is spread across genres such as administrative records, letters, legal texts, and literature — is 0.1961 bits. Among Old Babylonian-period tablets (13,708 labelled, 1,267 unlabelled), the same measure is 2.4879 bits, an overall gap of 2.2918 bits. Ur III genre composition is close to a single category (administrative texts are 58,175 of 59,455 labelled tablets, about 97.8%); Old Babylonian genre composition is spread much more evenly across administrative, lexical, letter, legal, literary, and mathematical texts. The gap holds in the same direction, by at least 0.5 bits, in 8 of the 10 holding collections large enough to test (at least 200 genre-labelled tablets of each period) — the pre-registered check against the possibility that the pattern is an artifact of one or two dominant collections rather than a property of the record generally.

What is being inferred

Nothing about WHY the gap exists is asserted here. The conjecture's own proposed mechanism — that a maximally centralized bureaucratic state (Ur III) channels writing into administrative record-keeping while a fragmented one (Old Babylonian) lets writing diversify into letters, law, and literature — is a plausible reading this measurement is consistent with, but it is not the only one. CDLI's genre labels are modern cataloguing vocabulary applied retrospectively, not an ancient classification; and Ur III's administrative dominance in the catalogue is partly a discovery-and-publication phenomenon (large administrative archives, once excavated, were prioritized for cataloguing over less legible material). This article states the size and replication of the gap; it does not choose between "strong states make boring archives" and "catalogues make boring archives."

What is attested

A conjecture was pre-registered (fresh-fablemax-20260705-008, registered 2026-07-05T15:42:42Z) with clause precedence fixed in advance and evaluated strictly in that order: (1) KILLED if the Old-Babylonian-minus-Ur-III entropy gap is at or below 0.25 bits (including negative); (2) SUPPORTED if the gap is at least 1.0 bit AND at least 2 collections each holding 200+ genre-labelled records of each period show a same-sign gap of at least 0.5 bits; (3) otherwise INCONCLUSIVE, explicitly including the case where fewer than 2 collections qualify regardless of the overall gap.

The computation (2026-07-05T15:44:39Z), read directly from the committed extract: Ur III entropy 0.1961 bits (59,455 labelled tablets, 237 unlabelled); Old Babylonian entropy 2.4879 bits (13,708 labelled, 1,267 unlabelled); overall gap 2.2918 bits. Clause (1) does not fire (2.2918 is far above 0.25). Clause (2) fires: the overall gap clears 1.0 bit, and of the 10 holding collections large enough to test, 8 show a same-sign gap of at least 0.5 bits — the "?" (unattributed) stratum (gap 1.8821), the Musée d'Art et d'Histoire, Geneva (1.9381), the National Museum of Iraq (2.2637), the Penn Museum (1.7225), the Yale Babylonian Collection (1.2809), the British Museum (2.0213), the de Liagre Böhl Collection (1.9369), and the Royal Ontario Museum of Archaeology (0.7307). Two collections do not clear the 0.5-bit replication bar: the Schøyen Collection, whose gap runs the opposite sign (−0.2138 — its Ur III holdings are, atypically, genre-diverse), and the J. Pierpont Morgan Library Collection (held at the Yale Babylonian Collection, catalogued there as a distinct sub-collection), whose gap is positive but small (0.447, just under the 0.5-bit bar — this is the collection the resolution record's own narrative shorthands as "Yale 0.447," distinct from the main Yale Babylonian Collection's own qualifying gap of 1.2809). The verdict follows the clause precedence exactly: SUPPORTED.

Why infer this entity

This article promotes a measurement, not an entity reconstructed from indirect traces — the usual Inferpedia case. It is included in the promotion ladder because the conjectures campaign's kill-loop (register a falsifiable, clause-ordered test before computing; compute; report the verdict even where it is inconvenient) produced a disciplined, denominator-carrying, cross-collection-replicated result, and this is the lane's only SUPPORTED finding that is not itself calibration-flagged (the literature search located no prior entropy-by-period computation on this catalogue). That absence-of-prior-formulation is a fact about a dated search, not a claim that the underlying pattern is new to history or to scholarship generally, and no such claim is made.

Evidence ledger

Two primary records, both committed to this site's own repository, were read directly: the pre-registration record for fresh-fablemax-20260705-008 (docs/generated/conjecture_prediction_entropy_floor_20260705.json, row 0) and its resolution/verdict record (docs/generated/conjecture_resolution_entropy_floor_20260705.json, row 0), which in turn point to the full per-collection computation artifact (docs/generated/conjecture_extracts/entropy_floor_20260705.json). No live model call read or summarized any external web page for this promotion; the entropy computation was performed once, in house, against CDLI catalogue metadata already resident in this project's database.

Counterarguments

The catalogue's sampling regime is not established as random: the underlying CDLI harvest is a continuation of an initially ID-ordered batch, and whether later growth (from roughly 60,000 to 126,000 records at computation time) preserved or corrected that ordering is not verified here. CDLI's genre labels are a modern cataloguing vocabulary, not a category the ancient scribes themselves used, so "genre diversity" measures how modern cataloguers classified the record, one further step removed from ancient practice. Ur III's administrative dominance in the catalogue is plausibly inflated by a discovery-and-publication bias: the large administrative archives from sites such as Umma, Drehem, and Girsu were excavated in bulk and catalogued efficiently, while more heterogeneous or fragmentary material may be under-catalogued regardless of period. And the two non-replicating collections are a genuine counter-case, not noise to be waved away: the Schøyen Collection's Ur III holdings are, on this catalogue's labels, more genre-diverse than its Old Babylonian holdings, the opposite of the aggregate pattern — evidence that collection formation history can invert the gap locally even where it holds in most of the record.

Confidence scores

• Direct attestation of the entropy values and the collection-level replication: 90
• Existence warrant (this is a real, reportable pattern in the catalogue): 88
• Specificity: 84
• Reconstruction dependence (how much rests on interpretation beyond the two entropy numbers and the replication count): 20
• Counterevidence pressure: 25

What would change the score

A version of this test run against a catalogue with a documented, verifiably random or complete sampling frame (rather than CDLI's harvest-continuation history) would raise the attestation score by closing the sampling-regime caveat. A genre classification built from ancient rather than modern categories — where recoverable — would address the cataloguing-vocabulary caveat and could move the reconstruction-dependence score in either direction depending on what it showed. And a study of cataloguing/publication intensity by genre and period, independent of the entropy measurement itself, would let a future article discriminate between the conjecture's "strong states make boring archives" reading and the rival "catalogues make boring archives" reading that this article deliberately leaves open.