Datasets:

OpenChristianDataOrg
/

open-christian-data

@@ -1,169 +1,388 @@
----
-license: cc0-1.0
-language:
-- en
-pretty_name: Open Christian Data
-size_categories:
-- 100K<n<1M
-task_categories:
-- text-generation
-- question-answering
-- text-classification
-- feature-extraction
-tags:
-- christian
-- theology
-- bible
-- commentary
-- sermon
-- catechism
-- confession
-- church-fathers
-- public-domain
-- religion
-configs:
-- config_name: bible_text
-  data_files:
-    - path: data/bible_text.jsonl
-      split: train
-- config_name: catechism_qa
-  data_files:
-    - path: data/catechism_qa.jsonl
-      split: train
-- config_name: church_fathers
-  data_files:
-    - path: data/church_fathers.jsonl
-      split: train
-- config_name: commentary
-  data_files:
-    - path: data/commentary.jsonl
-      split: train
-- config_name: devotional
-  data_files:
-    - path: data/devotional.jsonl
-      split: train
-- config_name: doctrinal_document
-  data_files:
-    - path: data/doctrinal_document.jsonl
-      split: train
-- config_name: hymn_collection
-  data_files:
-    - path: data/hymn_collection.jsonl
-      split: train
-- config_name: prayer
-  data_files:
-    - path: data/prayer.jsonl
-      split: train
-- config_name: reference_entry
-  data_files:
-    - path: data/reference_entry.jsonl
-      split: train
-- config_name: sermon
-  data_files:
-    - path: data/sermon.jsonl
-      split: train
-- config_name: structured_text
-  data_files:
-    - path: data/structured_text.jsonl
-      split: train
-- config_name: topical_reference
-  data_files:
-    - path: data/topical_reference.jsonl
-      split: train
----
-# Open Christian Data
-A free, author/work-led dataset of public domain Christian literature — Bible text, commentaries, sermons, catechisms, confessions, theology, prayer, and reference works, projected into clean records with full provenance metadata. Intentionally designed for developers and AI/LLM training.
-Publication status: The authorized 2026-07-17 upload to [OpenChristianDataOrg/open-christian-data](https://huggingface.co/datasets/OpenChristianDataOrg/open-christian-data) is live. The release contains **408 work units**, **1,812 source files**, **544,961 top-level source records**, **1,811 exported source files**, and **805,151 flattened JSONL rows**. The verified all-data-upload revision is `c2b4c0e0dd6676020ee089139b665383a5bc498e` at `2026-07-17T02:41:41+00:00`. Card-only corrections do not alter that verified data payload.
-All content is written by authors who died before 1928. All texts are in the public domain. The dataset is CC0.
-## Export shape
-The live whole-corpus HuggingFace export is the per-schema JSONL export generated by
-`build/scripts/export_huggingface.py`. Each published config corresponds to one schema-specific
-JSONL file.
-## What's included
-The public inventory is counted by recognized work units: the title-level item a reader, library, citation, or source catalog would treat as present. Export records, verses, dictionary entries, topics, Q&A pairs, chapters, and quotations are subordinate technical units.
-Generated from current `data/` with `py -3 build/tools/count_dataset_records.py`.
-| Category | Work units | Top-level records |
-|---|---:|---:|
-| Bible text | 9 | 282,395 |
-| Catechisms | 15 | 3,509 |
-| Church Fathers quotation source | 1 | 70,164 |
-| Commentaries | 34 | 115,069 |
-| Devotionals | 2 | 1,464 |
-| Doctrinal documents | 38 | 38 |
-| Hymns | 1 | 34,904 |
-| Lexicon | 1 | 0 |
-| Prayers | 4 | 205 |
-| Reference works | 7 | 26,305 |
-| Sermons | 7 | 5,297 |
-| Structured texts | 288 | 289 |
-| Topical reference works | 1 | 5,322 |
-| **Total** | **408** | **544,961** |
-The live HuggingFace export flattens structured texts and doctrinal documents, so this release contains **805,151 JSONL rows**. The source repository includes the generated work metadata audit at `docs/WORK_CATALOG.md` and a browser-friendly review surface at `docs/WORK_CATALOG.html`.
-## Loading the data
-```python
-from datasets import load_dataset
-# Load published configs
-commentary = load_dataset("OpenChristianDataOrg/open-christian-data", "commentary")
-structured_text = load_dataset("OpenChristianDataOrg/open-christian-data", "structured_text")
-```
-## Data structure
-Every record is a flattened row combining the resource-level metadata with entry-level fields. Key fields present on all records:
-| Field | Description |
-|---|---|
-| `_source_id` | Unique identifier for the parent resource (e.g., `matthew-henry-complete`) |
-| `entry_id` | Unique identifier for this entry (e.g., `matthew-henry-complete.Ezek.1.1-3`) |
-| `author` | Author name |
-| `tradition` | Array of theological tradition labels (e.g., `["reformed", "puritan"]`) |
-| `era` | Historical era (`patristic`, `medieval`, `reformation`, `post-reformation`, `modern`) |
-| `license` | `cc0-1.0` or `public-domain` |
-| `schema_type` | Content type — one of the config names above |
-Schema-specific fields vary by content type. Full field documentation is in the [source repository](https://github.com/OpenChristianData/open-christian-data/blob/main/docs/SCHEMA_SPEC.md).
-## Provenance
-Every record carries full provenance — the source URL, source format, download date, processing script, and a SHA-256 hash of the source file. This is stored in the `provenance_*` fields in the flattened export.
-## Verse references
-All verse references use [OSIS format](https://crosswire.org/osis/) (`Gen.1.1`, `Rom.9.1-Rom.9.5`). Where sources use non-standard reference formats, both the raw form and the normalized OSIS form are stored.
-## License
-- **Data**: [CC0 1.0 Universal](https://creativecommons.org/publicdomain/zero/1.0/) — all underlying texts are public domain; the structuring and provenance tracking are dedicated to the public domain
-- **Bible text (BSB)**: [CC0 1.0](https://berean.bible/licensing.htm)
-- **Code and schemas**: [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) — free for non-commercial use; commercial use requires a license (see [source repository](https://github.com/OpenChristianData/open-christian-data))
-## Citation
-```bibtex
-@dataset{open_christian_data_2026,
-  title        = {Open Christian Data},
-  author       = {OpenChristianData},
-  year         = {2026},
-  publisher    = {HuggingFace},
-  url          = {https://huggingface.co/datasets/OpenChristianDataOrg/open-christian-data},
-  license      = {CC0-1.0}
-}
-```
-## Source repository
-[github.com/OpenChristianData/open-christian-data](https://github.com/OpenChristianData/open-christian-data)

+---
+license: cc0-1.0
+language:
+- en
+pretty_name: Open Christian Data
+size_categories:
+- 100K<n<1M
+task_categories:
+- text-generation
+- question-answering
+- text-classification
+- feature-extraction
+tags:
+- christian
+- theology
+- bible
+- commentary
+- sermon
+- catechism
+- confession
+- church-fathers
+- public-domain
+- religion
+configs:
+- config_name: bible_text
+  data_files:
+    - path: data/bible_text.jsonl
+      split: train
+- config_name: catechism_qa
+  data_files:
+    - path: data/catechism_qa.jsonl
+      split: train
+- config_name: church_fathers
+  data_files:
+    - path: data/church_fathers.jsonl
+      split: train
+- config_name: commentary
+  data_files:
+    - path: data/commentary.jsonl
+      split: train
+- config_name: devotional
+  data_files:
+    - path: data/devotional.jsonl
+      split: train
+- config_name: doctrinal_document
+  data_files:
+    - path: data/doctrinal_document.jsonl
+      split: train
+- config_name: hymn_collection
+  data_files:
+    - path: data/hymn_collection.jsonl
+      split: train
+- config_name: prayer
+  data_files:
+    - path: data/prayer.jsonl
+      split: train
+- config_name: reference_entry
+  data_files:
+    - path: data/reference_entry.jsonl
+      split: train
+- config_name: sermon
+  data_files:
+    - path: data/sermon.jsonl
+      split: train
+- config_name: structured_text
+  data_files:
+    - path: data/structured_text.jsonl
+      split: train
+- config_name: topical_reference
+  data_files:
+    - path: data/topical_reference.jsonl
+      split: train
+---
+# Open Christian Data
+Open Christian Data (OCD) aims to be the single unified collection of all public domain Christian text. It exists to bring this collection together from across the internet and to structure it in a useful format for public use.
+Beyond the Bible, Christian writing is poorly represented as a cohesive dataset or as data structured for AI training. This Hugging Face release is the AI-focused publication of the collection: consistent, downloadable JSON for model training, evaluation, retrieval, and other computational uses.
+**Nearly 150 million words of Christian texts across 12 categories, including
+363 books.** The collection also includes 34,904 hymns, 5,297 sermons, nine
+Bible translations, and extensive commentary and reference material. That is
+roughly 196 million tokens across 805,146 records.
+## Why the collection exists
+Most public domain Christian literature beyond the Bible is scattered across
+scanned books, old websites, specialist file formats, and separate
+digitization projects. Even when a text is online, it may not be available as
+part of a collection in which its author, work, edition, structure, and source
+history are clearly preserved.
+Being online is not enough. Without that context, texts are difficult to find
+together, cite accurately, compare across editions, correct against their
+sources, credit properly, or use responsibly in research, study, software, and
+AI systems.
+The collection stands on the shoulders of the faithful people who preserved,
+digitized, and published these texts before us, and it stays true to their
+ethos: information about Christianity and historical Christian texts should be
+freely and widely available to all. Advances in AI have made a collection of
+this scale practical in a way it has not been before.
+## Three questions shape the data
+This dataset was the project's original goal: public domain Christian text
+structured for AI use. Building it raised three questions that now govern how
+every record is made:
+1. **Is the text a faithful copy of its source?** A record must be checkable
+   against the digital source it was taken from, so source, edition, and
+   processing history travel with the text.
+2. **Is the text a faithful copy of the original published work?** A digital
+   source can itself diverge from the edition it transcribes. The wording and
+   the structure and formatting of the original both carry information for a
+   serious reader of the text, so both are recovered and checked.
+3. **Is the text free to use?** Everything published here is intended to be
+   public domain with no rights attached, in keeping with the public-good
+   spirit of the project. Where rights questions remain, they are recorded
+   openly rather than passed on silently.
+Code alone cannot answer the first two questions, because its output is not
+easy for a person to review. Each work is therefore normalized into a shared
+intermediate representation that can be rendered as readable pages and
+visually verified against the source. That verification work incidentally
+produces human-readable editions of the texts, useful well beyond the
+dataset's original AI purpose; it lives in the
+[GitHub project](https://github.com/OpenChristianData/open-christian-data).
+## What is in the collection
+| Category | Configuration | What the collection contains |
+|---|---|---|
+| Books and long-form works | `structured_text` | 288 works, including theology, church history, treatises, and devotional classics, divided into 259,198 passages for publication |
+| Bible translations | `bible_text` | 9 translations containing 282,395 verses |
+| Commentaries | `commentary` | 34 commentaries containing 115,069 entries |
+| Dictionaries and encyclopedias | `reference_entry` | 6 dictionaries and encyclopedias containing 25,682 entries |
+| Topical Bibles and indexes | `topical_reference` | *Nave's Topical Bible* and *Torrey's New Topical Textbook*, containing 5,945 topics |
+| Devotionals | `devotional` | 2 devotionals containing 1,464 readings |
+| Sermons | `sermon` | 5,297 sermons from 7 collections |
+| Creeds, confessions, and doctrinal documents | `doctrinal_document` | 33 documents published as 1,314 articles and clauses |
+| Catechisms | `catechism_qa` | 15 catechisms containing 3,509 questions and answers |
+| Prayers and liturgies | `prayer` | 4 sources containing 205 prayers and collects |
+| Hymns | `hymn_collection` | 34,904 hymn texts |
+| Church Fathers quotations | `church_fathers` | 70,164 scripture-linked quotations |
+Representative works include Matthew Henry's *Exposition of the Old and New
+Testament*, Spurgeon's *Metropolitan Tabernacle Pulpit*, the *Catholic
+Encyclopedia*, *Nave's Topical Bible*, Augustine's *City of God*, Calvin's
+*Institutes of the Christian Religion*, and editions of the *Book of Common
+Prayer*.
+Hugging Face publishes the collection through 12 configurations. Long-form
+works are divided into passages for search, retrieval, and model use, producing
+805,146 downloadable records. That record count is a technical measure rather
+than the size of the library: a record may be a verse, commentary entry, sermon,
+hymn, question and answer, prayer, reference entry, or passage from a book.
+## What makes the collection different
+- **Unified.** Books, commentaries, sermons, hymns, catechisms, prayers, and
+  reference works that were scattered across archives, websites, and
+  specialist formats can be found and used together.
+- **Free to use.** The published dataset is released under CC0 for research,
+  publishing, software, model development, and other reuse. Source-specific
+  rights and the remaining audit questions are recorded below rather than
+  hidden behind the collection-wide license.
+- **Traceable provenance.** Source, license, edition, contributors, and
+  processing history travel with the text, so users can inspect where it came
+  from and credit the people and organizations that made it available.
+- **Faithful to the source text.** The goal is accurate recovery of the words
+  and structure of the represented edition, not merely data that passes a
+  schema. Corrections are made against source evidence.
+- **Texts remain connected to their context.** A passage stays attached to its
+  author, work, edition or translation, place in the work, and literary form.
+  That context is what makes a passage citable, historically intelligible,
+  comparable with other editions, and correctable when an error is found.
+## How the collection is made
+Each work begins with a particular public domain source and edition. Texts that are already digital are carefully parsed; texts that survive only as scans are transcribed using OCR and corrected against the page images. The aim is an accurate digital text, not simply a large quantity of extracted words.
+The text is then organized according to what it is. Books retain their divisions, sermons and hymns remain whole works, commentaries stay connected to the Bible passages they discuss, and authors, contributors, editions, translations, and source history remain attached.
+The Hugging Face files are the AI-ready publication of that work. Longer books and documents are divided into usable passages, while shorter forms remain natural units such as verses, sermons, hymns, questions and answers, prayers, quotations, and reference entries. Each record carries enough information to reconnect it to its parent work and source.
+The [GitHub project](https://github.com/OpenChristianData/open-christian-data) contains the texts, provenance, schemas, and construction tools, together with the developing TEI intermediate representations and visual reading and verification work from which future publication formats can be produced.
+## Using the dataset
+```python
+from datasets import load_dataset
+commentary = load_dataset(
+    "OpenChristianDataOrg/open-christian-data",
+    "commentary",
+)
+structured_text = load_dataset(
+    "OpenChristianDataOrg/open-christian-data",
+    "structured_text",
+)
+```
+To reproduce a particular release, pin its Hub tag after the tags are created:
+```python
+commentary_v020 = load_dataset(
+    "OpenChristianDataOrg/open-christian-data",
+    "commentary",
+    revision="v0.2.0",
+)
+```
+### Size of each configuration
+| Configuration | Records | Tokens | File size |
+|---|---:|---:|---:|
+| `bible_text` | 282,395 | 9.0M | 128.9 MB |
+| `catechism_qa` | 3,509 | 0.9M | 6.5 MB |
+| `church_fathers` | 70,164 | 16.9M | 119.9 MB |
+| `commentary` | 115,069 | 54.3M | 332.9 MB |
+| `devotional` | 1,464 | 0.5M | 3.3 MB |
+| `doctrinal_document` | 1,314 | 0.2M | 2.0 MB |
+| `hymn_collection` | 34,904 | 7.9M | 54.8 MB |
+| `prayer` | 205 | 0.1M | 0.3 MB |
+| `reference_entry` | 25,682 | 29.8M | 164.0 MB |
+| `sermon` | 5,297 | 36.7M | 163.6 MB |
+| `structured_text` | 259,198 | 39.2M | 616.1 MB |
+| `topical_reference` | 5,945 | 0.2M | 9.4 MB |
+| **Total** | **805,146** | **195.7M** | **1.60 GB** |
+Token counts cover the text of the works themselves — verse text, commentary,
+sermon and hymn text, questions and answers, entry bodies, and passages. They
+exclude JSON structure, identifiers, references, URLs, and the inlined metadata
+fields, so they measure the corpus rather than the file. They are computed with
+the `o200k_base` encoding; another tokenizer will give a different number. File
+size is the uncompressed JSONL on the Hub; the whole collection is roughly
+333 MB gzipped.
+Each configuration currently provides a single `train` split. The split name
+describes how the files are exposed on the Hub; it is not a recommendation
+that every record should be used for model training.
+## Data structure
+Every row combines information about its parent work with fields for the
+particular verse, entry, sermon, hymn, question, or passage.
+| Field | Meaning |
+|---|---|
+| `_source_id` | Identifier for the parent work or collection |
+| `entry_id` | Identifier for the individual record |
+| `author` | Author name, when known |
+| `tradition` | Descriptive theological-tradition labels |
+| `era` | Historical period |
+| `license` | Rights status of the source text |
+| `schema_type` | Kind of content and corresponding Hub configuration |
+Fields specific to commentaries, sermons, hymns, Bible texts, and long-form
+works vary by configuration. Full definitions are in the
+[schema documentation](https://github.com/OpenChristianData/open-christian-data/blob/main/docs/SCHEMA_SPEC.md).
+Bible references use [OSIS](https://crosswire.org/osis/) notation, such as
+`Gen.1.1` and `Rom.9.1-Rom.9.5`. Where a source uses another reference system,
+the original form is retained alongside the normalized reference when
+available.
+Where records carry `summary` and `key_quote` fields, those fields currently
+ship empty with `summary_review_status` set to `"withheld"`. They are
+placeholders for reviewed summaries, published empty rather than filled with
+unreviewed generated text.
+## Intended uses
+Open Christian Data is intended for:
+- research and digital-humanities work on historical Christian writing;
+- search, retrieval, reading, and reference applications;
+- comparison across authors, works, periods, and traditions;
+- language-model training, evaluation, retrieval augmentation, and other
+  computational uses that need openly reusable Christian texts;
+- building better editions or metadata by tracing records back to their
+  sources and contributing corrections.
+## Coverage and limitations
+- The collection is English-language and historical. It is not a balanced
+  representation of global Christianity, present-day Christian belief, or
+  every theological tradition.
+- Coverage reflects what survives in usable public-domain editions. Some
+  authors and traditions are much better represented than others.
+- Historical works preserve the language, assumptions, disagreements, and
+  prejudices of their authors. Inclusion is not endorsement.
+- Source transcriptions and metadata can contain errors. Validation can catch
+  structural problems, but it cannot prove that every reading or editorial
+  description is correct.
+- Editions and translations matter. Records from different editions should
+  not be treated as interchangeable without checking their provenance.
+- Tradition and era labels are aids to discovery, not authoritative judgments
+  about contested theological or historical boundaries.
+- The dataset is a research and development resource, not a doctrinal
+  authority or a critical edition of every represented work.
+Corrections and better source evidence are welcome through the
+[source repository's issue tracker](https://github.com/OpenChristianData/open-christian-data/issues).
+## Sources and acknowledgments
+Open Christian Data brings together texts made available through the
+[Christian Classics Ethereal Library](https://www.ccel.org/),
+[Internet Archive](https://archive.org/),
+[Project Gutenberg](https://www.gutenberg.org/),
+[Standard Ebooks](https://standardebooks.org/),
+[CrossWire SWORD](https://www.crosswire.org/sword/),
+[HelloAO](https://bible.helloao.org/),
+[HistoricalChristianFaith](https://github.com/HistoricalChristianFaith/Commentaries-Database),
+[Scrollmapper's bible_databases](https://github.com/scrollmapper/bible_databases),
+[Berean Bible](https://berean.bible/),
+[The Kingdom Collective](https://thekingdomcollective.com/spurgeon/),
+[New Advent](https://www.newadvent.org/cathen/), Wikisource, and other archives
+and source projects.
+The hymn collection was provided by [Hymnary.org](https://hymnary.org/) at
+Calvin University. If you build with it, please link to Hymnary.org and tell
+them about your project through [their contact page](https://hymnary.org/contact).
+Many books and other texts were **sourced via
+[CCEL.org](https://www.ccel.org/)**, which asks to be acknowledged for making
+its files available.
+The [principal-source and acknowledgment ledger](https://github.com/OpenChristianData/open-christian-data/blob/main/docs/SOURCES.md)
+distinguishes historic works from modern digitizations and databases, records
+source-specific credit, and links to third-party notices. Specific source and
+edition information also travels with individual records.
+The v0.2.0 source audit checked the rights behind the release rather than
+assuming them. It found five doctrinal records whose text came from a
+copyrighted modern Bible translation and so could not be released under CC0.
+Those records have been removed rather than kept under a license they did not
+fit. The passages themselves remain available in the public-domain or CC0 Bible
+translations included here.
+Three questions remain open and are recorded in the source ledger rather than
+settled quietly: the license covering the JWBickel structured data, the
+Wikisource transcription layer, and a translation warning carried by the
+HistoricalChristianFaith source. Credit is not treated as a substitute for
+permission, and anything that cannot be released cleanly will be removed or
+replaced rather than relabeled.
+## Versions
+| Version | Date | Summary |
+|---|---|---|
+| **v0.2.0** (corrected upload pending) | 2026-07-17 | A substantially larger collection, with richer descriptions of works and editions and a broad correctness pass. [Full notes](https://github.com/OpenChristianData/open-christian-data/blob/main/docs/releases/v0.2.0.md) |
+| v0.1.0 | 2026-04-12 | Initial public release: 11 configurations and 247,649 rows. |
+Hugging Face versions are Git revisions. Once the release tags are present,
+`v0.2.0` and `v0.1.0` can be used in `revision=`; immutable Hub commit hashes
+can also be used for exact reproduction.
+## License
+- **Published data:** [CC0 1.0 Universal](https://creativecommons.org/publicdomain/zero/1.0/)
+- **Berean Standard Bible:** [CC0 1.0](https://berean.bible/licensing.htm)
+- **Code, schemas, and tooling:** [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/)
+The CC0 dedication applies to the published dataset. Code, schemas, and tools
+in the source repository have a separate license and are not part of that CC0
+dedication.
+## Citation
+```bibtex
+@dataset{open_christian_data_2026,
+  title        = {Open Christian Data},
+  author       = {OpenChristianData},
+  year         = {2026},
+  version      = {0.2.0},
+  publisher    = {Hugging Face},
+  url          = {https://huggingface.co/datasets/OpenChristianDataOrg/open-christian-data},
+  license      = {CC0-1.0}
+}
+```
+## Source repository
+[github.com/OpenChristianData/open-christian-data](https://github.com/OpenChristianData/open-christian-data)