The dataset viewer is not available for this dataset.
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Sinew — a sourced Bible connection-graph
An open database of the Bible and its internal connections — text + canonical IDs + versification reconciliation + cross-references in one queryable artifact, every connection carrying its provenance.
"…the sinews and the flesh came up upon them… and they lived." — Ezekiel 37:8,10 Sinew is the connective tissue of the canon — the layer that links verse↔verse with provenance, so others build on reconnected Scripture instead of feeding a PDF to an AI.
🧑💻 Code, build & MCP server: https://github.com/LucasGalhardoLima/sinew 🔭 Live explorer (navigate by meaning, or search a theme): https://huggingface.co/spaces/LucasGalhardoLima/sinew-explorer
Why
Anyone building on the Bible (study apps, sermon tools, AI agents, research) redoes the same
foundational work — parsing text, canonical IDs, versification reconciliation, cross-reference
lookup — and worse, derives connections from text similarity, which is itself an interpretive
assumption (similarity ≠ connection). Sinew separates fact from interpretation: every
connection is type + source + weight, the dataset attributes, never asserts, and nothing is
silently dropped. v1 is entirely public-domain text + CC-BY data → frictionlessly redistributable.
The three tiers (never blurred)
- Tier 1 — Facts: verse text, canonical IDs, book metadata, versification map.
- Tier 2 — Sourced connections: every edge has
type + source + weight (+ review_status). - Tier 3 — Derived (optional): computed embeddings/similarity, "computed, not authoritative" — powers the live explorer's meaning-terrain and its in-browser theme search; built opt-in, never joined to the fact tables (and not part of this published dataset, which is Tier-1/2 only).
What's in v1
| Text | World English Bible (WEB), public domain — 31,095 verses / 1,189 chapters / 66 books |
| Canonical IDs | Book.Chapter.Verse (e.g. John.3.16), eng/KJV numbering base |
| Versification | eng (identity) + org (Hebrew) per verse, via the Copenhagen Alliance spec |
| Connections | OpenBible cross-references → 613,998 Tier-2 edges (type=cross_reference, weight=votes) |
| Formats | sinew.sqlite (all tables + joins) and per-table Parquet (this dataset's viewer) |
Files
| Path | What |
|---|---|
sinew.sqlite |
The whole dataset in one SQLite file — all tables, ready for joins |
parquet/*.parquet |
One file per table (powers the dataset viewer above) |
sources.lock.json |
Exact upstream retrieval URLs + sha256 pins (reproducible build) |
Load it
Parquet, via 🤗 datasets (each table is a named config):
from datasets import load_dataset
edges = load_dataset("LucasGalhardoLima/sinew", "connections", split="train")
verses = load_dataset("LucasGalhardoLima/sinew", "verses", split="train")
texts = load_dataset("LucasGalhardoLima/sinew", "texts", split="train")
SQLite, for joins (one file, all tables):
from huggingface_hub import hf_hub_download
import sqlite3
db = hf_hub_download("LucasGalhardoLima/sinew", "sinew.sqlite", repo_type="dataset")
con = sqlite3.connect(db)
# John 3:16 and its strongest sourced cross-references (with provenance)
con.execute("SELECT text FROM texts WHERE verse_id='John.3.16'").fetchone()
con.execute("""SELECT target_verse_id, source, weight FROM connections
WHERE source_verse_id='John.3.16' AND review_status='ok'
ORDER BY weight DESC LIMIT 5""").fetchall()
# Hebrew (org) numbering for an eng verse
con.execute("SELECT scheme_ref FROM versification_map WHERE verse_id='Joel.2.32' AND scheme='org'").fetchone()
# -> ('Joel.3.5',)
Schema (tables)
verses—verse_id (PK),book,chapter,verse,canonical_order,tier=1(stable, translation-independent address)books—book (PK),name,testament,book_number,chapter_counttexts—verse_id,translation,text,tier=1(PKverse_id+translation; v1: WEB only)versification_map—verse_id,scheme ∈ {eng,org},scheme_ref,status ∈ {present,merged,split},tier=1connections—source_verse_id,target_verse_id,type,source,weight,confidence,review_status,turpie_class (NULL in v1),tier=2dataset_meta—key,value
Rules: no connections row without source; endpoints reference the canonical address
(verses.verse_id), never a texts row; an edge whose endpoint doesn't resolve (≈311, versification
gaps) is flagged in review_status, never dropped — the "no silent failure" guarantee.
Sources & license
| Source | Provides | License |
|---|---|---|
| World English Bible (via getbible.net v2) | PD English text | Public Domain |
| Copenhagen Alliance versification | eng/org mappings | CC-BY / open |
| OpenBible.info cross-references | ~340k weighted edges | CC-BY |
This compilation is licensed CC-BY-4.0 and requires attribution to OpenBible.info; the
underlying Bible text is public domain. Exact retrieval URLs + sha256 pins are in sources.lock.json
and the dataset_meta table.
Roadmap (P1): typed NT→OT quotations (Turpie 1868, classified A–E); deeper Tier-3 semantic search beyond the explorer (a query/MCP-layer semantic endpoint — the explorer's in-browser theme search already ships); original-language (Macula) lemma/morphology; more PD translations + multilingual alignment.
- Downloads last month
- 13
