You need to agree to share your Hugging Face information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

This dataset is distributed under CC BY-SA 4.0 (the horizon_* and divine_epithet* configs are computed fact tables). Access is gated for attribution and share-alike stewardship: any Adapted Material you share must carry the same license and attribute NuBerea. Please review the per-config license table in the dataset card.

Log in or Sign Up to review the conditions and access this dataset content.

NuBerea signal_theology

An emergent, within-Bible operator kernel and narrated-event control surface. The kernel is a set of typed doctrine / boundary-line statements that surfaced bottom-up from corpus-intrinsic signals (composition dating, intra-canon citation, reception weighting, embedding geometry) over the canonical Bible, with the narrated events that ground them and an intra-canon reception keystone. No doctrine taxonomy was declared in advance; benchmarks were used only post hoc.

Everything here is STRUCTURE in the received text, not a claim about truth: "reception" means within-Bible cross-attestation (how heavily the rest of the canon cites or reuses a verse, pericope or event), never patristic or reformer reception, and salience is not endorsement.

Configs

Config Grain Rows License Description
operators_kernel operator 80 cc-by-sa-4.0 Unified operator kernel: statement text plus register {apex (argued) | ground (presupposed) | both}, type {doctrine | boundary}, type_flag, contestation 0-1, grounding {event | assertion}, era and composition-date span, intra-canon reception weights (recv, recv_w).
operators_conviction operator 59 cc-by-sa-4.0 Pole-pure conviction operators (what the canon presupposes rather than argues), synthesized over 2,409 received grounds: anchor_thesis, anchor_unit, lineage_units, support_size, weight_mass, corpus_mix.
operator_relations operator pair 2,547 cc-by-sa-4.0 Operator-operator edges: edge_type {dialectic (emergent co-occurrence) | supersession (dated OT-boundary to NT-doctrine) | event_cogrounding (shared salient event)}, weight, orientation_basis.
event_salience event 3,392 cc-by-sa-4.0 Narrated events with reception-derived salience (peak and sum over pericope spans), n_spans, type and discovery class.
event_operator_edges event x operator 1,913 cc-by-sa-4.0 Which events ground which contention operators (op_rank 1-57), with recv and n_spans support.
event_edges event pair 49,417 cc-by-sa-4.0 Event-to-event intertextual graph rolled up from verse-level intertextuality; keeps match_grain and src_osis / tgt_osis anchors so verse level stays reconstructible.
reception_verse verse 29,244 cc-by-sa-4.0 Intra-canon reception keystone per verse: recv_raw, recv (template-discounted), recv_pagerank, formulaicity; the granularity floor.
reception_pericope pericope 2,317 cc-by-sa-4.0 Pericope rollup of reception_verse (peak / sum / mean, top_verse), kept alongside the verse table, never instead of it.
horizon_verse verse x field 94,652 computed Per (verse, cited surface, citing field): first / last cited year and citation counts. Fact scalars only, no text.
horizon_verse_summary verse 51,493 computed Cross-field rollup of horizon_verse: number of engaging fields, field list, total citations, earliest year.
operator_all operator 139 (expected) cc-by-sa-4.0 Unified operator registry: operators_kernel and operators_conviction gathered under one facet value (kernel | conviction), joined by a shared operator_key display column (kernel's own op_id, conviction's own anchor_unit) alongside every native column kept under its own name. No Bible-verse anchor (match_grain constant segment).
operator_edge_all event/operator pair 53,877 (expected) cc-by-sa-4.0 Unified edge table: event_edges, event_operator_edges, and operator_relations gathered under one edge_domain value (event_event | event_operator | operator_operator), re-keyed onto a shared src/tgt pair. match_grain is per row: pericope on event_event (carries its own src_osis/tgt_osis anchor), segment elsewhere; event_edges' own native grain column is kept alongside as source_match_grain.
reception_signal_all verse/pericope 31,561 (expected) cc-by-sa-4.0 Unified reception table: reception_verse and reception_pericope gathered under one grain value (verse | pericope), every member column kept under its own name. The underlying verse-anchor scheme for this repo is not yet resolved to a standard OSIS scheme, so osisRef_std is left null on every row rather than asserted.
event_recurrent_type type 12 cc-by-sa-4.0 Recurrent event types on the event layer: battle or siege, plague, famine or drought, earthquake, theophany, covenant ceremony, census, feast observance, exile or captivity, anointing or enthronement, raising or resurrection appearance, and the parent type conflict. One row per type, never an occurrence: label, parent, own-words definition, evidence tier, and the title rules that make the edge derivation reproducible from this table.
event_recurrent_type_edge event x type 6,214 cc-by-sa-4.0 Event-to-type edges keyed on event_id, both grains side by side: 6,023 verse rows (a verse cited under a seed dictionary headword or Nave's topic that lies inside the event's span; standard-grid osisRef) and 191 pericope rows (the event-level typing derived from them and the title rule; osisRef/osisRef_end carry the event span). event_form_class carries the event tables' existing event_type form class for joins.
divine_epithet epithet 82 computed Registry of designations of the divine referent other than the proper name (Most High, Almighty, LORD of hosts, Holy One of Israel, Rock as a designation, ...): class, referent pointer and scope (contested referents are stated, no stance taken), a cross-link to the title concepts, evidence tier and computed counts. Titles stay separate from persons.
divine_epithet_alias epithet x form 411 computed Surface forms per language: English terms in our own wording plus one anchor alias per rule alternative (Hebrew, Aramaic or Greek lemma, Latin lemma) with its Strong or lemma id. Dictionary lemmas and alias terms only, never running text.
divine_epithet_attestation verse/pericope 6,372 computed 4,831 verse rows (3,889 designation, 942 ambiguous) across MT, SBLGNT, LXX (Rahlfs) and the Vulgate, with the standard-grid osisRef (TVTMS bridge) and the native reference beside it, plus 1,541 derived pericope rows beside them. Ids, references and codes only.
divine_epithet_all epithet / alias / attestation 6,865 computed One table for the whole divine-epithet registry: divine_epithet, divine_epithet_alias and divine_epithet_attestation gathered under one kind value (epithet | alias | attestation), every member column kept under its own name. Every alias and attestation row carries its epithet's label, class, referent, scope and evidence tier, and every row carries alias_forms (all surface forms of the epithet), so one filter on a label or a surface form reaches the attestations without a join. match_grain is verse or pericope on attestations and segment on registry and alias rows. The members stay published.

The config list above matches the live file listing on the repo exactly.

Known limits (read before ranking anything)

  • Reception weighting is load-bearing but multiplicity-driven: its relevance signal collapses when controlling for n_spans (0.28 to 0.08). Treat recv as an attestation count, not an independent importance measure.
  • 1,933 of 3,392 events (57%) ground no operator; about 37% once zero-content stubs and redundant Gospel parallels are removed. Grounding tracks cumulative cross-attestation, so single-locus high-value passages are under-grounded. Absence from event_operator_edges is not evidence that a passage carries no doctrine.
  • The conviction pole's near-complete ground coverage is partly tautological (a received-ground thesis exists for nearly every pericope); it is coverage, not validated correctness, and its heresy-recovery check is post hoc with no null class.
  • The operator set is a sample weighted toward argued, high-reception claims; event-mode analyses are chapter-grain; the pipeline depends on a single sentence-encoder family.
  • Three configs (operators_kernel, operator_relations, event_operator_edges) are frozen products: their local build inputs drifted after publication, so the published tables are reproduced from a pinned seed rather than regenerated from scratch.
  • event_recurrent_type_edge thresholds are provisional. A type is never promoted to an event-level (pericope) edge on Nave's references alone; Nave's-only verse rows stay, flagged as weak evidence. The verse rows skip the Psalter and every chapter whose numbering differs between the standard grid and the Hebrew or Greek scheme, because the event spans carry no single declared numbering; the title rule still applies there.
  • divine_epithet_attestation records designation uses only. An anchor word judged ordinary (a boulder, a human redeemer, daylight) is counted on divine_epithet.n_ordinary_hits, never shipped as a row; ordinary means no designation evidence was found, not that a use was proved non-divine. ambiguous rows are never promoted.

Attribution

Derived entirely from NuBerea's own Bible-scoped claim, contention/conviction, intertextuality, pericope and event layers over public-domain and openly licensed editions of the canonical text. The event-type and divine-epithet configs also read NuBerea's own token tables of the Hebrew, Greek, LXX and Latin texts and verse-reference lists from a Bible dictionary index and Nave's Topical Bible (references and topic names only, no prose). Operator statements are LLM-synthesized English paraphrases; no verse text from any restricted edition ships here.

NuBerea project. Licensed CC BY-SA 4.0 (horizon_* and divine_epithet* configs: computed fact tables).

What this is not

Not a doctrinal taxonomy, not a truth-verdict on any doctrine, and not a record of patristic or reformer reception. It is a descriptive map of what the canon itself argues, presupposes and narrates, weighted by how the canon cites itself.

Downloads last month
1,730