BibCrit v3.0

Free, open-access web tool for biblical textual criticism at bibcrit.com. Compare MT, LXX, and Dead Sea Scrolls; reconstruct Hebrew Vorlagen; profile scribal tendencies; detect theological revisions; track patristic citations; model numerical discrepancies; detect literary structures (chiasm, inclusios, parallel panels); identify documentary source layers (J/E/D/P); and visualize manuscript genealogies — all in a browser, in English and Spanish.

Screenshots

Tools

Data Sources

License note: The ETCBC corpora (MT/BHSA, DSS, and Peshitta) are released under CC-BY-NC 4.0. The app code is Apache 2.0; the corpus data it ingests retains its own license terms. Do not use the ingested ETCBC data for commercial purposes without a separate agreement with ETCBC.

Tech Stack

Architecture

BibCrit/
├── app.py                      # Flask app factory; lazy _init() wires corpus + pipeline
├── state.py                    # Shared singletons: corpus, pipeline, i18n, TranslationProxy
├── requirements.txt
├── render.yaml                 # One-click Render deploy config
│
├── blueprints/
│   ├── textual.py              # /divergence, /backtranslation, /dss, /genealogy + APIs
│   ├── critical.py             # /scribal, /numerical, /theological, /patristic + APIs
│   ├── literary.py             # /chiasm, /source + SSE APIs
│   ├── discovery.py            # /discovery, /api/discovery/cards, /api/admin/discovery/flag
│   └── research.py             # /health, /guide
│
├── biblical_core/
│   ├── claude_pipeline.py      # ClaudePipeline: Claude calls, Supabase cache, budget tracking
│   ├── corpus.py               # BiblicalCorpus: loads MT, LXX, DSS, SP, GNT
│   ├── ref_utils.py            # Reference string parsing; per-tool verse-count limits
│   └── divergence.py           # parse_claude_response, format_sbl_footnote, format_bibtex
│
├── data/
│   ├── i18n.json               # All UI strings (en, es)
│   ├── prompts/                # Versioned prompt templates ({tool}_{version}.txt)
│   ├── cache/                  # Disk-based analysis cache fallback ({sha256}.json)
│   └── corpora/
│       ├── mt_etcbc/           # Masoretic Text (ETCBC/BHSA morphology)
│       ├── lxx_stepbible/      # Septuagint (Rahlfs)
│       ├── dss/                # Dead Sea Scrolls (ETCBC, primarily 1QIsaᵃ)
│       ├── sp_etcbc/           # Samaritan Pentateuch (dt-ucph/sp via ETCBC)
│       └── gnt_opengnt/        # Greek New Testament (SBLGNT)
│
├── scripts/
│   ├── precache_all.py         # Seed 91 featured passages in English
│   ├── precache_es.py          # Translate all 91 passages to Spanish
│   ├── push_cache_to_supabase.py  # Push disk cache → Supabase
│   ├── ingest_mt.py            # ETCBC/BHSA → CSV
│   ├── ingest_lxx.py           # LXX (STEP) → CSV
│   ├── ingest_lxx_rahlfs.py    # LXX (Rahlfs) → CSV
│   ├── ingest_dss_1qisaa.py    # ETCBC/DSS (1QIsaᵃ) → CSV
│   ├── ingest_sp.py            # SP (dt-ucph/sp) → CSV
│   └── ingest_gnt.py           # SBLGNT → CSV
│
├── templates/                  # Jinja2 templates extending base.html
└── static/                     # CSS (bibcrit.css, style.css), JS per-tool, SVG assets

Key design decisions

• state.py holds no blueprint or app imports, preventing circular dependencies. Blueprints read state.corpus and state.pipeline directly.
• app._init() runs before the first request (thread-safe double-checked locking) to keep startup fast.
• SSE (Server-Sent Events) streams real-time progress steps (step / done / error frames) to the browser while Claude analyzes.
• Budget checks happen before every API call; once spend_usd >= cap_usd the endpoint returns an error frame without calling the API.

Cache system

Every analysis result is keyed by:

cache_key = SHA256("{reference}|{tool}|{prompt_version}|{model}")

English (EN): written to both Supabase (analysis_cache table) and disk (data/cache/{sha256}.json). Supabase is the primary read path; disk is the fallback.

Spanish (ES): stored in Supabase only (analysis_cache_es table). Generated by scripts/precache_es.py, which translates cached EN analyses rather than re-running the full Claude pipeline.

Getting Started

1. Clone

git clone https://github.com/Jossifresben/bibcrit.git
cd bibcrit

2. Install dependencies

python -m venv .venv
source .venv/bin/activate      # Windows: .venv\Scripts\activate
pip install -r requirements.txt

3. Create .env

cp .env.example .env           # or create from scratch — see table below

4. Run

flask run

The app is available at http://127.0.0.1:5000.

For production-like local testing:

gunicorn app:app --workers 1 --threads 2 --timeout 120

Environment Variables

Corpus Ingestion

Run ingestion scripts once to populate data/corpora/. Each script pulls from Text-Fabric or a local file and writes normalized CSV:

python scripts/ingest_mt.py
python scripts/ingest_lxx_rahlfs.py
python scripts/ingest_dss_1qisaa.py
python scripts/ingest_sp.py
python scripts/ingest_gnt.py

Text-Fabric downloads corpora on first run (~several hundred MB). The ETCBC and SP corpora require acceptance of their respective licenses before use.

Pre-caching Featured Passages

The repo ships with analyses for featured passages across all 9 tools. To seed or refresh:

# Seed all missing EN analyses (safe to re-run; skips already-cached)
python scripts/precache_all.py

# Seed a specific tool only
python scripts/precache_all.py --type numerical

# Dry run — show what would be seeded without calling the API
python scripts/precache_all.py --dry-run

# Push disk cache to Supabase
python scripts/push_cache_to_supabase.py

# Generate Spanish translations of all cached EN analyses
python scripts/precache_es.py

URL Routes

Pages

Analysis API (SSE streaming)

All stream endpoints emit step (progress), done (full JSON result), and error frames.

Open Data API

BibCrit's analysis corpus is publicly readable:

GET /api/cache
GET /api/cache?tool=divergence
GET /api/cache?tool=theological&ref=Isaiah+7:14
GET /api/cache?discovery_ready=true&limit=50&offset=0

All data is released under Apache 2.0. If you use BibCrit analyses in research, please cite:

Fresco Benaim, J. (2026). BibCrit: AI-assisted biblical textual criticism. ORCID: 0009-0000-2026-0836

Corpus Browser API

Export API

Discovery API

Hypothesis Voting

Internationalization

UI strings live in data/i18n.json. The lang query parameter (?lang=es) selects the active language. state.TranslationProxy (exposed as _t() in templates) falls back to English if a key is missing.

To add a language: add a top-level key to data/i18n.json matching all en keys, and add a button to the language picker in templates/base.html.

Scholar / Student Mode

A toggle in the navbar (book icon) switches between Scholar mode (default — full technical analysis) and Student mode (plain-language explanations highlighted, technical text hidden). The preference is persisted in localStorage under bibcrit-mode. No server changes are needed; the mode is purely client-side CSS (body.mode-student .technical-only { display: none }).

Supabase Schema

-- English analysis results cache
CREATE TABLE analysis_cache (
  cache_key       TEXT PRIMARY KEY,
  reference       TEXT,
  tool            TEXT,
  prompt_version  TEXT,
  model_version   TEXT,
  data            JSONB,
  cached_at       TIMESTAMPTZ,
  discovery_ready BOOLEAN DEFAULT FALSE
);

-- Spanish analysis results cache
CREATE TABLE analysis_cache_es (
  cache_key       TEXT PRIMARY KEY,
  reference       TEXT,
  tool            TEXT,
  prompt_version  TEXT,
  model_version   TEXT,
  data            JSONB,
  cached_at       TIMESTAMPTZ
);

-- Monthly API spend tracking
CREATE TABLE budget (
  month       TEXT PRIMARY KEY,   -- e.g. '2026-03'
  spend_usd   NUMERIC,
  cap_usd     NUMERIC,
  updated_at  TIMESTAMPTZ
);

-- Hypothesis voting
CREATE TABLE hypothesis_votes (
  reference   TEXT PRIMARY KEY,
  upvotes     INTEGER DEFAULT 0,
  downvotes   INTEGER DEFAULT 0,
  updated_at  TIMESTAMPTZ
);

All tables are optional — the app falls back to disk if Supabase is unavailable (EN only; ES requires Supabase).

Deploy to Render

A render.yaml is included. To deploy:

1. Push the repo to GitHub.
2. In the Render dashboard, click New > Blueprint and point it at the repo.
3. Set ANTHROPIC_API_KEY (and optionally SUPABASE_URL / SUPABASE_KEY / BIBCRIT_ADMIN_KEY) as environment variables.
4. Deploy.

The default spend cap is $10.00/month. Raise it via BIBCRIT_API_CAP_USD in the Render environment variables.

Roadmap

✅ Completed (v2.5 → v3.0)

• 13 analysis tools across textual, critical, literary, and discovery categories
• Chiasm & Literary Structure Detector (/chiasm) — first open tool of its kind
• Source Criticism Tool (/source) — J/E/D/P attribution with Wellhausen / Friedman / Baden grounding
• Full OT corpus coverage — MT (ETCBC) + LXX (Vaticanus) across all books
• 1QIsaᵃ (Dead Sea Scrolls) corpus — DSS Bridge Tool runs on real scroll data
• Samaritan Pentateuch corpus (5 books, 114,889 words)
• MorphGNT / SBLGNT corpus (27 NT books, 137,554 words)
• Peshitta real corpus — ETCBC/peshitta via Text-Fabric (SEDRA); 39 OT books, 308,863 Syriac word tokens
• NT Use of OT Analyzer — citation-form determination across MT and LXX
• Spanish translation layer (analysis_cache_es) — full UI + analysis output
• 141 featured passages pre-cached across all tools (instant load)
• Persona-based home page — Scholar, PhD Candidate, Student entry points
• Scholar / Student mode toggle (technical vs. plain-language view)
• RIS and TEI XML export for Divergence Analysis
• BiblIndex deep-links in Patristic Citations tool
• Open Data API

🔜 Phase 1 — Months 1–2: Foundation

Corpus

• Peshitta real corpus — ETCBC/peshitta via Text-Fabric; 39 OT books, 308,863 Syriac word tokens in pesh_etcbc/
• MT/LXX expansion — all 39 MT books and 38 LXX books already present (complete)
• Extended DSS witnesses — 4QSamᵃ (4Q51), 11QPaleoLev (11Q1), 4QDeutn (4Q41) added; note: 1QpHab excluded (ETCBC DSS has no MT-aligned verse coordinates for this scroll)

🔜 Phase 2 — Months 3–4: New Traditions

Corpus

• Targum corpus (Onkelos + Jonathan) — Sefaria API; 26 books; 240,297 Aramaic word tokens in targ_sefaria/
• Vulgate corpus (Jerome) — scrollmapper/bible_databases Clementine Vulgate; 66 books; 569,588 Latin word tokens in vul_clementine/
• LXX variant MSS — add Sinaiticus and Alexandrinus alongside Vaticanus; unlocks three-way LXX manuscript comparison

New tools

• Targum Comparator (/targum) — MT vs. Targum word-level comparison; expansion types: theological, halakhic, messianic, divine-name substitution (Memra/Shekhina), haggadic; Sperber / McNamara methodology
• NT Textual Tradition Analyzer (/nt-text) — classify NT variants across Byzantine, Alexandrian, and Western text types; UBS/NA apparatus data + AI analysis; Metzger methodology

Infrastructure

• Hebrew RTL UI (he locale) — full RTL layout; makes BibCrit usable by Israeli biblical scholars
• True Anthropic token streaming — replace blocking messages.create() with messages.stream(); sections appear as Claude writes them, 10–20s earlier on first queries

🔜 Phase 3 — Months 5–6: Synthesis

Corpus

• Second Temple literature — 1 Enoch, Jubilees, Sirach, 4 Ezra, Tobit from CLTK / Open Scriptures; register stl_cltk/ tradition
• Peshitta NT — Syriac NT (Aramaic Primacy tradition); third NT tradition alongside SBLGNT

New tools (capstone — require all prior corpora)

• Second Temple Literature Bridge (/stl) — map allusions from 1 Enoch, Jubilees, Sirach, 4 Ezra to canonical texts; critical for DSS and NT intertextuality; Nickelsburg / VanderKam / Collins methodology
• Intertextuality Mapper (/intertextuality) — full allusion network for any passage: inner-biblical allusions, NT echoes, patristic citations, DSS parallels, Second Temple parallels; exportable as JSON-LD / RDF; Hays / Beale / Fishbane methodology

Infrastructure

• Full open API v1 — versioned endpoints, API key management, rate limiting, Swagger docs at /api/docs
• Dutch UI (nl) and Portuguese UI (pt)
• JOSS paper v3 + Zenodo DOI update — reflect 15 tools and 9 corpus traditions

v3.0 target state

License

Apache 2.0 — see LICENSE.

Corpus data retains its own licenses: ETCBC/BHSA and ETCBC/DSS are CC-BY-NC 4.0; SP (dt-ucph/sp) and SBLGNT have their own terms. See each upstream repository for details.

Credits

Built by Jossi Fresco. Analysis powered by Claude (Anthropic). Corpus data: ETCBC (MT, DSS), dt-ucph/sp (Samaritan Pentateuch), Rahlfs (LXX), SBLGNT (GNT).