Dra-Go-Nite Omni Knowledge

Dataset Description

A deduplicated, provenance-preserving knowledge corpus for retrieval-augmented generation.

Every record carries an immutable identifier, its origin, a content hash, a license, and a trust classification — so any retrieved passage can be traced back to exactly where it came from.

Dataset Details

• Curated by: ORCHORDS.COM LAB
• Funded by: ORCHORDS.COM LAB
• Shared by: ORCHORDS.COM LAB
• License: See Licensing
• Status: 14 of 17 planned sources ingested as of 2026-10-13

Uses

• Retrieval-augmented generation and question answering
• Training and evaluating retrieval systems
• Provenance and traceability auditing
• Reproducing source-grounded evaluations

Dataset Format

Field	Type	Description
record_id	string	Immutable record identifier, stable across corpus revisions
dataset	string	Originating source dataset
upstream_revision	string	Exact upstream revision the record was taken from
source_record_id	string	Native ID in the source dataset
source_url	string	Locator the record can be recovered from
content	string	Full text of the record
content_sha256	string	SHA-256 of normalized content
dedup_status	string	Deduplication outcome
license	string	License and usage restrictions carried over from the source
attribution	string	Author or citation, where supplied
trust_class	string	documented_reference, general_reference, scientific_educational, generated_reasoning, or restricted_quarantined
corpus_revision	string	Corpus version the record belongs to
recovery_path	string	Location the full passage is recoverable from

trust_class governs retrieval. generated_reasoning, restricted_quarantined, and records with dedup_status = evaluation_only are excluded from the default trusted retrieval path.

How to Use It

from datasets import load_dataset
ds = load_dataset("ORCHORDS/Dra-Go-Nite", split="train", streaming=True)

Dataset Quantification

• Record count — 62,449,035 indexed records
• Passage-level locator rows — 228,764,790
• Indexed Parquet shards — 3,319
• Orphan locators — 0
• Records processed for deduplication — 65,779,843
• Exact duplicate groups — 700,492
• Records excluded as evaluation_only — 54

Licensing

No single blanket license is asserted for this corpus. Original terms are carried at the record level: the license and attribution fields on each row hold the terms and attribution that govern that specific record, and they travel with the data.

Use of this corpus is subject to the terms recorded in those fields. For anything not covered there, contact crm@orchords.com.

Important

• Coverage is incomplete. 14 of 17 planned sources are present.
• Retrieval over this corpus has not demonstrated aggregate improvement over a model operating without it. Retrieval works; retrieval benefit is not established.
• generated_reasoning records are model output and can be wrong. They are labelled, not certified.
• Caselaw content is not legal advice.
• No latency figures are published. Measuring them under corpus build load has not been completed, so none are claimed.

Citation

@misc{orchords2026dragoniteomniknowledge,
      title={Dra-Go-Nite Omni Knowledge},
      author={{ORCHORDS.COM LAB}},
      year={2026},
      howpublished={Dataset}
}