OSKG-YahWeh

An Open Structured Knowledge Graph applied to the concept of Yahweh in the anthropological record — a data-driven investigation into the historical origins of the biblical God and its implications for an accurate understanding of Old Testament Judaism.

The OSKG-YahWeh knowledge graph: 723 claims, 154 chapter notes, 18+ primary sources, 9 canvases. Each node is a claim, note, or source. Each edge is a typed relationship — supports, contradicts, depends on, challenged by. The colors and clusters reflect topic domains: Yahweh's origins (green), comparative ANE religion (pink), monotheism's emergence (orange), and primary source evidence (light nodes).

---

What This Is

A multi-layered research pipeline that transforms 17 scholarly monographs into a queryable, edge-connected knowledge graph of 723 evaluated claims. Each claim is a first-class node tagged by topic, scholar, evidence type, and confidence. Edges are typed: supports, contradicts, depends on, challenged by. The graph can answer questions that reading summaries cannot — "show me every claim about Asherah," "what would break if Deut 32:8-9 is monotheistic," "how many scholars converge on Yahweh's southern origin at high confidence?"

The project applies the same structured evidence synthesis methodology used by the Open Research Knowledge Graph (ORKG) at Leibniz University Hannover — independently converged, validated by published literature, and recognized as a supported methodology for computational meta-analysis. See METHODOLOGY for the full methodological alignment.

Why Yahweh?

The central question: who was Yahweh, actually? Not the God of later Jewish and Christian theology. The deity as his earliest worshippers knew him. A southern storm god? A local Canaanite iteration? A consort to Asherah? The Most High from the beginning? The evidence is scattered across archaeology, epigraphy, philology, and biblical studies. No single scholar holds the full picture. This project synthesizes them all into a single structured corpus and asks: what does the weight of evidence actually show?

The implications extend beyond academic curiosity. If the Bible's own internal evidence reveals a polytheistic past later edited into monotheism, what does that mean for how we read the Old Testament? For how we understand the development of Judaism? For faith traditions built on the premise of a unique, unchanging revelation?

The Pipeline

17 books → 154 chapter notes → 723 claims → typed-edge graph → synthesis

Phase	Artifact	Size	Description
0. Acquisition	sources/books/_fulltext/ (gitignored)	19MB	17 monographs extracted to plaintext
1. Chapter Notes	notes/theology/	154 notes, ~466KB	Chapter-by-chapter critical analysis of every monograph. Every claim evaluated for evidence, confidence, stakes, and disagreement.
2. Claims Extraction	notes/claims/	723 claims	Each claim extracted into its own file with YAML frontmatter, typed edges, and primary source links. LLM + human-in-the-loop.
3. Knowledge Graph	Obsidian graph view	Fully connected	Claims are nodes. Edges connect evidence across scholars. Tags filter by topic, scholar, evidence type, confidence.
4. Synthesis	notes/synthesis/	5 files, 142KB	Hinge inventory, cascade trees, counter-position stress tests, convergence points, and a personal capstone.

Graph Structure

Canvases (Visual Maps)

Open these in Obsidian's Canvas view:

Canvas	What It Shows
Methodology Canvas	The full pipeline architecture and ORKG alignment
asherah-debate	The goddess-vs-symbol fault line across 6+ scholars
convergence-fault-line-map	Where scholars agree and where they genuinely disagree
counter-position-survival-map	Heiser/Schmid/Tigay/Kaufmann — what survives each challenge
deut-32-8-9-cascade-map	Full dependency tree of the project's central text
evidence-density-by-domain	Textual vs. archaeological vs. onomastic evidence distribution
scholarly-lineage-confessional-map	Who studied under whom, who's Jewish/Catholic/Evangelical/Atheist
chronology-source-timeline	Primary sources mapped to historical periods
kaufmann-paradox	Kaufmann's revolution thesis vs. the consensus

Claims (723 nodes)

Every claim file in notes/claims/ has:

• Frontmatter: claim ID, statement, confidence rating, topic tags, evidence tags, scholar tag, source tag
• Structured body: The Claim → Evidence → Confidence → Stakes → Disagreement → Edges → Assessment
• Typed edges: Supports, Contradicts, Depends On, Challenged By — each with descriptive rationale
• Primary source links: Wikilinks to the 18+ inscription files in sources/primary-sources/

Filter the graph by tag to see networks: topic/asherah (86 connected claims), topic/kenite-hypothesis (24 connected claims), topic/deut-32-8-9 (12 connected claims with cascades).

Primary Sources (18+ texts)

sources/primary-sources/ contains full text, translations, and scholarly significance for every major inscription cited in the project:

Academic genealogies and confessional commitments of the 45+ scholars in the OSKG. Colors represent scholarly traditions: green nodes = evangelical/archaeological perspectives, blue/purple = critical Jewish scholarship, red/brown = Copenhagen minimalists and threshold figures. Lines trace teacher-student relationships, intellectual debts, and methodological fault lines. The graph reveals which scholars cite each other — and which camps never engage.

Kuntillet Ajrud · Khirbet el-Qom · Deut 32:8-9 + 4QDeut · Soleb Shasu · Merneptah Stele · Mesha Stele · Tel Dan Stele · Lachish Ostraca · Ketef Hinnom Amulets · Elephantine Papyri · Ugaritic Baal Cycle (KTU 1.1–1.6, Keret, Aqhat) · Black Obelisk · Kurkh Monolith · Siloam Inscription · Gezer Calendar · Ekron Inscription

Scholars Covered (17 monographs from 15 scholars)

Tier 1 (core consensus + counters)	Tier 2 (extensions + challenges)	Reference Works
Smith — Early History of God (2002)	Albertz — History of Israelite Religion Vol I–II (1994)	Keel & Uehlinger — Gods, Goddesses, and Images (1998)
Smith — Origins of Biblical Monotheism (2001)	Day — Yahweh and the Gods of Canaan (2000)	Kaufmann — The Religion of Israel (1960)
Römer — The Invention of God (2015)	Fleming — Yahweh Before Israel (2021)	Schmid — Historical Theology of the Hebrew Bible (2019)
Dever — Did God Have a Wife? (2005)	Lewis — Origin and Character of God (2020)
Sommer — The Bodies of God (2009)	Tigay — You Shall Have No Other Gods (1986)
Heiser — The Unseen Realm (2015)	Cross — Canaanite Myth and Hebrew Epic (1973)
Stavrakopoulou — God: An Anatomy (2021)

See notes/theology/scholarly-directory-yahweh-origins.md for a 45+ scholar reference directory. See notes/theology/meta-analysis-scholars.md for analysis of how scholars' confessional commitments, academic genealogies, and institutional incentives shape their conclusions.

Key Findings (Synthesis)

The four-phase synthesis in notes/synthesis/ identifies what the graph reveals:

Hinge claims (phase 1-2): The top 5 load-bearing claims control cascades of 28-112 downstream claims each. Schmid's late dating of the Pentateuch is the single most destructive counter-position — if correct, ~41% of the consensus graph loses its textual foundation.

Convergence points (phase 4): Scholars agree at HIGH+ confidence on: Yahweh's southern origin, El as Israel's original god, monotheism's emergence in the exile, and the divine council as a biblical reality.

Genuine unknowns (phase 4): Scholars genuinely disagree (with HIGH confidence on both sides) about: whether Asherah was a goddess consort or a cult symbol, whether Deut 32:8-9 describes one God or two, and whether early Israel was polytheistic or monolatrous.

Capstone (notes/synthesis/capstone-what-does-this-mean.md, 49KB): Personal synthesis — what the evidence means for faith, scripture, and the God you're left with.

Key Documents

Document	Description	Size
sources/source-inventory	Complete inventory of 17 monographs, 18+ primary sources, and supplementary materials — every source from which 723 claims were extracted. Includes copyright/fair use framework.	24KB
METHODOLOGY	Full methodological alignment with ORKG at Leibniz University Hannover. Independent convergence, pipeline architecture, comparison table.	5KB
notes/pipeline-overview	Phase-by-phase pipeline architecture: acquisition → notes → claims → graph → synthesis.	7KB
notes/claims-architecture	Claim file format, tag taxonomy, edge types, extraction process.	10KB
notes/claims-audit-report	Post-extraction audit results: 723 claims, zero broken edges, 269 source links added.	—

Document → Canvas Map

Each document has corresponding visual canvases that map its argument structure:

Document	Canvases	What the Canvas Shows
Source Inventory	canvases/chronology-source-timeline	Primary sources mapped to historical periods
	canvases/evidence-density-by-domain	Textual vs. archaeological vs. onomastic evidence distribution
METHODOLOGY	canvases/Methodology Canvas	Full pipeline architecture and ORKG alignment
Phase 2 (Cascade Trees)	canvases/deut-32-8-9-cascade-map	Dependency tree of the project's central hinge text
Phase 3 (Stress Tests)	canvases/counter-position-survival-map	Heiser/Schmid/Tigay/Kaufmann — what survives each challenge
Phase 4 (Unknowns)	canvases/convergence-fault-line-map	Where scholars agree and where they genuinely disagree
	canvases/asherah-debate	The goddess-vs-symbol fault line across 6+ scholars
Scholarly Directory	canvases/scholarly-lineage-confessional-map	Academic genealogies and confessional commitments
Kaufmann vs. Consensus	canvases/kaufmann-paradox	Revolution thesis vs. evolutionary model

LLMs + OSKG: Grounded Reasoning Over Structured Knowledge

This repository is designed to be LLM-navigable. Unlike general-purpose chatbots that answer from training data — often hallucinating citations, flattening scholarly disagreement, or defaulting to theological platitudes — an LLM pointed at OSKG-YahWeh can traverse the graph to produce evidence-grounded, scholar-attributed, confidence-weighted answers.

How It Works

The graph provides three layers an LLM can follow:

1. Claim files (notes/claims/) — 723 atomic, falsifiable assertions. Each claim has a scholar, a source work, a confidence rating, the evidence presented, who disagrees, and what's at stake. An LLM can read the claim, check its edges, and follow the trail.

2. Typed edges — The Edges section of every claim file contains wikilinks to other claims. These are not hyperlinks — they are typed semantic relationships. Supports: means the target provides evidence for this claim. Contradicts: means the two claims cannot both be true. Depends on: means this claim collapses if the target is falsified. An LLM can trace these chains: "Show me everything that depends on the Soleb inscription reading 'Yhw' as a divine name."

3. Primary sources (sources/primary-sources/) — The ground truth layer. When a claim says "the Kuntillet Ajrud inscriptions read," the claim file has a wikilink to the actual text and translation. An LLM can verify the claim against the source — not against its training data.

What This Enables

Before (training-data-only LLM):

"Did Yahweh have a wife?" "The Bible presents God as unique and without consort. Some fringe scholars have suggested otherwise based on limited evidence."

After (OSKG-grounded LLM):

"Did Yahweh have a wife?" "Six scholars at HIGH+ confidence argue yes, based on the Kuntillet Ajrud and Khirbet el-Qom inscriptions (both read 'Yahweh and his Asherah'), the 3,000+ Judean Pillar Figurines (Dever), and the Ugaritic precedent of Athirat as El's consort. One scholar (Smith) dissents at HIGH confidence, arguing 'his Asherah' refers to a cult symbol, not a goddess. The evidence is contested — see topic/asherah (86 claims) for the full fault line. Scholarly consensus leans toward goddess (HIGH), but Smith's philological case is methodologically serious (MEDIUM). This is a genuine unknown, not a settled question."

The difference: the second answer has scholars, evidence, confidence ratings, primary sources, and honest uncertainty. It doesn't just recite a position — it maps the debate.

Using an LLM with OSKG-YahWeh

With Obsidian + Copilot/ChatGPT plugins: Point the LLM at the vault. Ask a question. The LLM reads the claim files, follows the wikilinks, and synthesizes from the graph. Because every claim has typed edges, the LLM can do things its training data can't — like calculating how many claims would break if Schmid's late dating of the Pentateuch is correct (answer: ~41%).

With Hermes Agent (the toolchain used to build this project): Point a session at the repo, ask a synthesis question, and the agent reads claim files, follows edges, checks primary sources, and reports with citations. This is how the Argument Dependency Map's four phases were produced — the agent didn't hallucinate from training data. It read the graph.

With any RAG system: The repo is plain Markdown with YAML frontmatter and wikilinks. Any document-store RAG pipeline can ingest notes/claims/ as a corpus and use the frontmatter tags + edges for retrieval. The typed edge structure means the retriever doesn't just find relevant documents — it can follow argument chains.

Why This Matters

LLMs trained on the open internet have absorbed centuries of theological apologetics, popular misconceptions, and flattened scholarly consensus. Ask one about Yahweh's origins and you'll get a mix of Sunday school, Wikipedia, and whatever blog posts dominated the training corpus. Point that same LLM at OSKG-YahWeh and it reads 723 claims extracted from 17 monographs, follows typed edges through dependency trees, checks primary sources, and reports with scholar-attributed, confidence-weighted answers. The graph doesn't tell the LLM what to believe. It tells it what the evidence shows — and what remains genuinely unknown.

How to Use

This repository is an Obsidian vault. Clone it, open it in Obsidian, and you get:

• Graph view: Filter by tag to see networks of claims, scholars, and evidence
• Canvas view: Open the visual maps in canvases/
• Wikilink navigation: Every claim, note, and source is interlinked
• Search: Full-text search across 154 notes, 723 claims, and 18 primary sources

If browsing on GitHub, start at Home.

Philosophy

1. Primary sources first. Every claim links to the inscription or text it cites.
2. Confidence, not certainty. Every claim rated VERY HIGH through LOW with explicit rationale.
3. Steelman every position. Consensus, counter-positions, and minimalists all represented in their strongest form.
4. Typed edges. "Supports" is not "depends on" is not "contradicts." Precision in disagreement.
5. Human in the loop. LLM extraction, human verification. The graph is machine-assisted but human-curated.

Methodology

The four-phase pipeline (left, green/blue nodes): 17 books extracted → 154 chapter notes → 723 claims → synthesis. The ORKG parallel (right, blue nodes): independently converged on the same architecture — structured claim extraction → typed semantic edges → graph querying → evidence synthesis. The pipeline was developed organically through iterative refinement of the Truth Project's book note process. The convergence with ORKG validates the approach: when a humanities researcher and a computer science lab independently arrive at the same architecture, the pattern has genuine methodological validity.

The pipeline architecture converges with the Open Research Knowledge Graph (ORKG) at Leibniz University Hannover (Auer, D'Souza, Farfar et al., 2019–present). Both systems use: structured claim extraction → typed semantic edges → graph querying → evidence synthesis. The independent convergence validates the approach. OSKG-YahWeh differs from ORKG in scale (17 books deep rather than millions of papers shallow) and domain (humanities/biblical studies rather than scientific literature). See METHODOLOGY for the full analysis.

---

"The first principle is that you must not fool yourself — and you are the easiest person to fool." — Richard Feynman