NuBerea/segments

Analytical unit definitions for biblical, Second Temple, rabbinic, and early Christian corpora: pericope boundaries for the Hebrew Bible and Greek New Testament, segment boundaries for the Dead Sea Scrolls, Talmudic literature (Mishnah, Tosefta, Bavli, Yerushalmi), early Christian writings, Nag Hammadi codices, Old Testament pseudepigrapha, and Migne's Patrologia Latina, together with a cross-corpus event topology (canonical biblical events, their aliases, relations, and mappings onto pericope spans).

Pericopes — the traditional reading units of scripture — are the natural analytical grain between the verse and the book, and this dataset supplies a consistent set of such units across very different corpora so that segment-level scholarship (source criticism, synoptic comparison, reception history) can be carried out on a common footing. The full specification (per-config documentation, schemas, and methodology) is available in the gated SPEC accompanying this dataset.

License and attribution

Estate-curated segmentation and event-topology tables are released under CC BY 4.0. Several configs carry per-source terms or derive from sources requiring attribution:

Source	Link	License / terms
MACULA Greek (SBLGNT) and MACULA Hebrew, Clear Bible	https://github.com/Clear-Bible	CC BY 4.0
STEPBible data (TVTMS versification), Tyndale House	https://github.com/STEPBible/STEPBible-Data	CC BY 4.0
Sefaria (Talmud reference structure; facts only, no source text redistributed)	https://www.sefaria.org	CC BY-NC 4.0 (structure references)
Matthew Barron, NET Bible Synopsis of the Four Gospels (bible.org)	https://bible.org/assets/pdf/White_ntsynopsis.pdf	Editorial content used as external T0 reference
Migne, Patrologia Latina (segment derivation)	https://patristica.net/latina/	Public domain
Migne, Patrologia Graeca (segment derivation, pg config)	https://patristica.net/graeca/	Public domain
Dead Sea Scrolls transcription lineage (dss_pericopes)	https://huggingface.co/datasets/NuBerea/dead-sea-scrolls	CC BY-NC 4.0 lineage — this config includes reconstructed text and is licensed CC BY-NC 4.0, not CC BY 4.0
Secondary/non-biblical source union (secondary_text_all; Josephus, Philo, classical-Greek segments, Nag Hammadi Coptic, CSEL heretical Latin, ECW apocryphal English)	https://huggingface.co/datasets/NuBerea/secondary-sources	CC BY-SA 4.0 (composite, per-row k) — this config carries a ShareAlike part and is licensed CC BY-SA 4.0, not covered by this card's CC BY 4.0 header

Note: the previous card revision declared a blanket CC BY 4.0; the dss_pericopes exception above corrects that against the estate's source-of-truth licensing records.

pg config (T6, 2026-09-12)

Greek pericope-scale segments over patristics-migne-pg's clean_text (Migne's Patrologia Graeca), the pg analog of pl/sch — 295,198 rows, MainText_ColGreek only (lang=grc), built by huggingface/candidates/build/patristics-migne-pg/scripts/build_pg_segments_driver.py (build_pg), published under commit 933a93d1. osisRef is unresolved at this rung (0/295,198 filled), same pattern as pl and sch — scripture anchoring is a later unit's job. Licensed public domain (Migne, out of copyright); see the license table above.

entity_index config

MACULA-derived named entity index — one row per entity-verse occurrence, 33,985 rows / 2,969 distinct entities (entity_id/canonical_name/entity_type/testament/ strong_id/lemma/osisRef/match_grain=verse, plus legacy osis_ref/book/ chapter/verse). Verse grain native (no pericope rollup). Built by ontology/scripts/build_entity_index.py, wired live per the 2026-07-24 owner ruling. entity_type is heuristic (referent-prefix + LN/SDBH domain codes); entity_id is a name-slug key, not a person-identity key (distinct same-named individuals merge; one individual can fragment across typing variants). Licensed under the MACULA CC BY 4.0 terms in the license table above (computed from macula-greek + macula-hebrew, no NC/SA in the lineage).

entity_gazetteer / entity_mentions / entity_config_edges — planned, not live

Three GAZ-ASSEMBLY-v3-REF candidate configs (ASSEMBLY_V3DRY4_RECEIPT under ~/Documents/OCR/ecrp_v1/W53_parity/lexicon/gazetteer/assembly/, supersedes the REF-excluded ASSEMBLY_V3DRY3_RECEIPT), extending this repo's existing entity_index config per GAZETTEER_PLAN_v3.md §4 card #1's recommended placement option — not yet published, not in the configs: front-matter above, same posture as this card's other not-yet-live rows.

• entity_gazetteer (66,749 rows, planned) — one row per entity_id (index, not an occurrence table): entity_type, canonical_name_en, aliases (lang/form/form_norm/source/is_lemma across en/la/grc/hbo/arc/syr/cop/de plus a long tail of real ISO codes recovered from the seed's own lang_raw column, with an honest und fallback where none is carried), external ids, and a derived n_mentions_by_rail map.
• entity_mentions (1,980,629 rows, planned) — one row per mention, any rail, never rolled up (granularity HARD RULE): bible rows resolve real OSIS verse refs; pl/schol/apoc/pseud_la/clsgrc/pg/ref rows carry a rail-scoped anchor string with osisRef empty/non-OSIS/null depending on rail (documented in the manifest node's address.unresolved_ok). ref (Reformation clean-text, 354,793 rows, new this leg) folds only REF_LANG_SCREEN_v0.csv verdict=LAUNCH (Latin) volumes, entity_id-linked rows only, per the GAZ-OCC-REF-v1 + REF-LA-EYEBALL-200 precision-gate ACCEPTs (>=90% on 320 combined adjudicated rows): confidence=high uniformly, review_status=eyeball_320_92pct for the 294,329 major/patristic rows and sample_tail_83 for the 60,464 tail rows. mention_role is name on every row except the 610 rows linked as demonyms (an adjective or people-name formed from a place name, linked to that place), which carry mention_role=demonym and review_status=relinked_demonym.
• entity_config_edges (9,333 rows, planned) — a DERIVED entity_id x config_repo x config x rail x mention_count rollup (KG Option B, relation mentions), never a replacement for entity_mentions' own row-level data. Now includes the reformation config_repo (+2,396 rows) alongside the pre-existing bible/latin/greek cells.

ALTERNATIVE (not applied, owner decision pending): a new NuBerea/entities repo instead of extending segments — see GAZETTEER_PLAN_v3.md §4 card #1 option (a). Placement and publish are both owner-gated (card #1 GO); this section documents the candidate shape only.