Inferpedia - an encyclopedia of the missing

AI-generated conjecture · below the evidence/publication boundary

← All conjectures · Institutions, law & bureaucracy

The bureaucratic entropy floor

Status: Supported

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 proposed connection between two domains, generated by a language model. 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 bureaucratic entropy floor. Joins Shannon information theory to Mesopotamian political cycles: strong states make boring archives. Peak bureaucratic centralization crushes the genre diversity of the written record toward an entropy floor, and political fragmentation shows up as an entropy rebound.

Prediction clause (verbatim)

In CDLI catalogue metadata, the Shannon entropy of genre composition (in bits, over the standard genre labels) for Ur III-period tablets sits at least 1.0 bit below the Old Babylonian-period entropy, with administrative genres exceeding 85% of Ur III items; and the sign of the Ur III versus Old Babylonian entropy gap replicates within at least 4 of the 5 largest holding collections computed separately, addressing the non-random-sample caveat.

Kill-dataset (verbatim)

Kill: CDLI cuneiform catalogue metadata (genres by period; current local batch is a non-random sample, hence the within-collection replication requirement). An Ur III entropy within 0.3 bits of Old Babylonian, or a sign flip in most collections, kills it.

Provenance

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

Generated by a fresh Fable-tier instance at maximum effort with generation-first blindness (no repo reads, no searches, no DB queries); title list supplied at launch, titles only, no verdicts or dossiers seen; prompt pre-committed in docs/GOAL_CONJECTURES_BATCH2_20260705.md (7e55eb8). Novelty unverified by construction.

Novelty / leakage triage

no prior formulation located (search dated 2026-07-05)

Raw genre shares consistent with the premise exist (a CDLI-derived dataset reports ~93.6% administrative for Ur III) and computational archive analysis is active (Murasu SNA), but no Shannon-entropy computation of genre composition by period was located. No prior formulation located (search dated 2026-07-05). In-house resolvable in principle (CDLI catalogue genres by period), with the non-random-sample caveat binding.

Predictions

Supported registered 2026-07-05

Resolution: Supported

Caveats: Calibration verdict on a descriptive gap whose INTERPRETATION stays open: the pre-registered claim (Ur III genre entropy at least 1 bit below OB) is robustly true in this catalogue - overall gap 2.29 bits, replicated in 8 of 10 collections holding 200+ of each period - but the conjecture's mechanism ('strong states make boring archives') is one reading among several. CDLI genre labels are modern cataloguing vocabulary; Ur III administrative dominance is partly a discovery/publication phenomenon (massive administrative archives excavated and prioritized); the harvest's sampling regime remains unverified-random (Wave 3a ID-ordered caveat; continuation regime not established); and the Schoyen counter-case (-0.21) demonstrates concretely that collection composition can invert the gap locally - what replicates across MOST collections is the catalogued record's shape, not necessarily ancient production. Registrant's prior (SumTablets ~93.6% Ur III admin share) pointed the right direction; the OB side and the replication structure were genuinely unseen at registration.

In-house test (triage: novel_unlocated; raw Ur III genre-share figures exist in a dataset description but no entropy computation or period comparison was located): in CDLI catalogue metadata, the Shannon entropy of genre composition for Ur III tablets sits materially below Old Babylonian entropy - peak bureaucratic centralization compressing the written record's genre diversity - replicated within collections to guard against the non-random-harvest caveat.

Resolution criteria: Population: tradition=cuneiform, period_canonical_name in {Ur III, Old Babylonian}; genre = raw genres_for_ws_e string, with empty/null excluded from the entropy and their share reported. H = Shannon entropy in bits of the genre-label distribution per period. CLAUSE PRECEDENCE, evaluated strictly in this order: (1) KILLED if H_OB minus H_UrIII <= 0.25 bits (including negative). (2) SUPPORTED if H_OB minus H_UrIII >= 1.0 bit AND at least 2 collections each holding >= 200 genre-labelled records of EACH period show a same-sign gap >= 0.5 bits. (3) Otherwise INCONCLUSIVE - explicitly including the case where fewer than 2 collections qualify, regardless of the overall gap. Narrative (non-binding): full genre distributions per period, per-collection gaps, unlabelled shares.

Known priors disclosure: Seen at registration: period counts (Ur III 59,692; Old Babylonian 14,975 - the local CDLI harvest has grown to 126,000 records via concurrent sanctioned scriptome work since Wave 3a's 60,000) and top collection names/counts. NOT seen: any genre distribution, any genre-by-period or genre-by-collection cross-tab. Known priors: the triage dossier's SumTablets figure (~93.6% administrative among Ur III transliterated tablets) implies LOW Ur III entropy is likely - which is why the kill/support hinges on the OB comparison and the within-collection replication, not on Ur III concentration alone. Standing caveats acknowledged: the harvest's randomness is unverified (Wave 3a's batch was ID-ordered; the continuation's sampling regime is not established here); CDLI genre labels are modern cataloguing vocabulary; Ur III administrative dominance is partly a discovery/publication phenomenon.

Exactly as pre-registered, clause precedence explicit: Shannon entropy (bits) of genre labels per period, kill clause first, then support (overall gap >= 1.0 AND >= 2 collections with >= 200 labelled records of each period showing same-sign gap >= 0.5), else inconclusive. Script preserved at docs/generated/conjecture_extracts/entropy_floor_20260705_script.py.

Dataset: CDLI catalogue metadata in-house (126,000 records at computation; concurrent scriptome harvest continuation since Wave 3a): Ur III 59455 genre-labelled (+237 unlabelled), Old Babylonian 13708 (+1267). Genre = CDLI's genres_for_ws_e label; collections from collections_raw. Read-only.

computed 2026-07-05

Weigh in

No community feedback yet.

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.