Old Testament textual analysis bundle: reader apparatus, full critical apparatus, witness manuscript catalog, verse-level witness texts and alignments, syntax graph index, and Vulgate Vorlage apparatus. Based on the Westminster Leningrad Codex and Biblia Hebraica apparatus.

Configs

Config	Rows	Columns	Description
`reader_apparatus`	2,817	5	OT reader apparatus, SBLGNT-style: one note per variation unit with WLC word-span anchors (`inline_anchor.wlc_span`); SP (von Gall) units, DSS variants, LXX/known watches, SP transposition watches (⸉)
`apparatus_full`	2,817	22	Complete critical apparatus for OT witnesses: all variant note types, mirror categories, and full witness attestation details
`witness_catalog`	122	14	Registry of OT manuscript witnesses: source family, type, textual tradition, and geographic provenance
`witness_verse_texts`	7,217	22	Per-verse reconstructed text for each OT manuscript witness with attestation kind and textual confidence
`witness_verse_alignment`	7,217	8	Per-verse alignment records: attestation presence flags and deviation type from WLC base text
`witness_apparatus_links`	2,817	11	Cross-reference table linking OT witness readings to their apparatus entries by verse
`witnesses_public`	122	9	Public-facing OT witness registry: sigla, formal names, and source family assignments
`syntax_graph_index`	1,904	8	Index of MACULA Hebrew syntax trees by sentence: dot and render file paths for visualization
`vulgate_vorlage_apparatus`	27	5	Simplified Vulgate Vorlage apparatus: cases where Jerome's Hebrew source differed from WLC, in reader-note format
`vulgate_vorlage_apparatus_full`	27	22	Full Vulgate Vorlage apparatus with categorized divergence types, confidence ratings, and all mirror-category fields
`vulgate_vorlage_registry`	1	7	Summary registry of Vulgate Vorlage analysis: hotspot count, divergence type distribution, and confidence tier breakdown

Staged verse-level LXX configs (computed locally, NOT YET PUBLISHED)

The LXX analyzers already compute four per-verse findings from HF upstream, but the published apparatus only retains them as coarser span/point watch rollups inside apparatus_full / reader_apparatus. Per the granularity rule these verse-grain rows are emitted alongside the rollups (never replacing them); every row carries match_grain='verse' + an osisRef anchor. They are staged to current_HF/out/<config>_factual/ and are not uploaded (the YAML configs: block above intentionally omits them until upload).

Build/refresh: python scripts/build.py --factual-only

Config (staged)	Rows	Columns	Source fn	Description
`lxx_mismatch_verses`	4,753	9	`compute_lxx_mismatches` (full rows)	One row per LXX↔MT content-word Strong's mismatch pair (incl. legitimate-LXX-difference slice); back-linked to the published watch that consumed it. Cols: osisRef, match_grain, greek_strong, hebrew_strong, hebrew_uStrong, mismatch_class, crosswire_target_uStrong, in_published_watch, watch_apparatus_id
`lxx_divergence_verses`	21,015	8	`_compute_verse_substitutions`	Per-verse content-word substitution count (DNA alignment) with book-conditioned threshold (9 for Ps else 5) and flag. Cols: osisRef, match_grain, book, subs, threshold, flagged, in_published_watch, watch_apparatus_id
`lxx_addition_verses`	997	7	`build_lxx_addition_watches` (flagged set)	Per-verse LXX content-word excess vs MT (excess≥6), before cluster/singleton filtering. Cols: osisRef, match_grain, greek_content, hebrew_content, excess, in_published_watch, watch_apparatus_id
`lxx_omission_verses`	23,213	6	`build_lxx_omission_watches` (per-verse view)	One row per WLC verse: LXX presence + versification-gap flag + omission-cluster membership. Cols: osisRef, match_grain, lxx_present, is_versification_gap, in_omission_cluster, cluster_apparatus_id

Note: is_versification_gap is uniformly False in the current lxx-wlc-bridge snapshot because verse_bridge covers the full WLC grid (no WLC ref is absent from the bridge); the true LXX-omission signal is lxx_present=False (2,109 verses, 2,000 in contiguous ≥3-verse clusters). The flag is retained per the plan schema and would surface gaps if a future bridge dropped offset verses outright.

Reproducibility

This dataset is fully reproducible from HF T0 sources:

python scripts/build.py              # Build all 11 configs from HF
python scripts/build.py --verify     # Compare against published HF data
python scripts/build.py -v           # Verbose output

HF Dependencies

HF Dataset	Role
`NuBerea/macula-hebrew`	WLC verse grid (23,213 verses)
`NuBerea/samaritan-pentateuch-gall`	SP witness verse text (von Gall 1918, PD; config `text`) + MS-context apparatus
`NuBerea/aleppo-codex`	Aleppo witness verse text
`NuBerea/vulgate-clementina`	Vulgate verse text
`NuBerea/lxx-wlc-bridge`	LXX witness text + lexical bridge
`NuBerea/ot-apparatus-primitives`	DSS material witness (non-derivable)
`NuBerea/macula-hebrew-syntax`	Syntax graph index
`NuBerea/crosswire-greekhebrew`	Greek↔Hebrew Strong's reference
`NuBerea/stepbible-tbesg`	Greek lexicon metadata
`NuBerea/stepbible-tbesh`	Hebrew lexicon metadata

Usage

from datasets import load_dataset

# Load a specific config
ds = load_dataset("NuBerea/ot-apparatus", "reader_apparatus")

Attribution

Upstream Data Sources

Source	License
Westminster Leningrad Codex 4.20, J. Alan Groves Center	Public Domain
IAA Leon Levy Digital Library imagery (read by us; no images redistributed)	our transcriptions, PDDL
Dead Sea Scrolls in Text-Fabric, ETCBC	cited facts-only (edition-disagreement register); no ETCBC text shipped
Samaritan Pentateuch, ed. A. von Gall 1918 (NuBerea/samaritan-pentateuch-gall)	Public Domain — SP witness text + MS context; the NC DT-UCPH set is fully retired from this pipeline
MACULA Hebrew Linguistic Datasets	CC BY 4.0
Clementine Vulgate, NOCR	Public Domain

NuBerea project. Licensed CC BY 4.0.

Downloads last month: 84