# 2026 GitHub / Dataset Source Review

Updated: 2026-07-09

This note documents how the site uses the external repositories and dataset links supplied for the OWL source refresh. These materials are useful for discovery, search, OCR comparison, downloader review, and visualization ideas. They are not official evidence sources by themselves.

## Inputs checked

- FULL_EPSTEIN_INDEX repository: https://github.com/theelderemo/FULL_EPSTEIN_INDEX
- FULL_EPSTEIN_INDEX Hugging Face dataset: https://huggingface.co/datasets/theelderemo/FULL_EPSTEIN_INDEX
- FULL_EPSTEIN_INDEX raw-file folder: https://drive.google.com/drive/folders/18tIY9QEGUZe0q_AFAxoPnnVBCWbqHm2p?usp=drive_link
- GitHub topic page sorted by forks: https://github.com/topics/epstein-files?o=desc&s=forks

## Handling rules

- Treat GitHub topic entries as discovery leads, not evidence.
- Treat RAG/chat projects as interface references only unless they preserve document citations.
- Treat OCR text as searchable working material; verify against the source image/PDF before using it in claims.
- Preserve victim dignity, redactions, and privacy limitations. Do not mirror sensitive raw material into the public site.
- Keep repository stars, forks, and update dates as freshness/context signals only.
- Do not import generated claims from third-party tools unless the underlying file ID and source image can be checked.

## Promoted to the site source review layer

| Entry | Why it is useful | Site role | Main caution |
| --- | --- | --- | --- |
| GitHub topic: epstein-files | Shows 35 public repositories matching the topic, with visible language and update metadata. | Repository discovery index | Topic descriptions are maintainer supplied. |
| theelderemo/FULL_EPSTEIN_INDEX | Aggregates links and index material around public releases; README links dataset and raw files. | Cross-reference index | README warns about OCR noise and unverified material. |
| FULL_EPSTEIN_INDEX Hugging Face dataset | Provides a CSV/text dataset surface observed at 8.53k rows. | Search/OCR comparison | Dataset text is not a substitute for file review. |
| FULL_EPSTEIN_INDEX raw file folder | Provides a retrieval mirror linked from the repo README. | Mirror for file retrieval | Confirm file IDs and release context against issuing agencies. |
| AnkitNayak-dev/EpsteinFiles-RAG | Popular visible RAG pipeline on the topic page. | Tooling reference | RAG answers need citations before use. |
| stonesalltheway1/Epstein-Pipeline | Document processing pipeline for OCR, entities, dedupe, and export. | Processing reference | Entity extraction can over-link names and mentions. |
| AbhisumatK/Epstein_Files_RAG | RAG exploration platform for unsealed court documents. | Search-interface reference | Confirm corpus boundaries before comparing outputs. |
| prizmatik666/epstein-ripper | DOJ PDF download tooling. | Retrieval reference | Downloader outputs need ledgers and checksums. |
| division-labs/epstein-maxwell-analysis | Claims a citation-forward analysis approach. | Method comparison | Imported claims still need direct EFTA citation review. |
| division-labs/epstein-maxwell-netviz | D3 document similarity and entity visualization project. | Visualization reference | Similarity networks are navigation aids, not findings. |

## Screened but not promoted

Some visible topic entries were not added to the main card rail because their topic descriptions were too thin, too broad, or not directly useful to this site's source-review workflow. These can still be revisited later:

- ishumilin/epstein-chat: potentially useful chat interface, but no site card until citation behavior and corpus boundaries are reviewed.
- guilyx/epsteinexposed and guilyx/epsteinexposed-mcp: unofficial API/MCP wrappers; defer until API provenance and terms are reviewed.
- IAmSoThirsty/Hub_of_Epstein_Files_Directory: directory hub with automation claims; defer until file provenance is inspected.
- imdvz/AskJeffrey: RAG project; defer until citation behavior is checked.
- r1c0n/doj-epstein-crawler and usmanx03/epstein_doj_downloader_python: downloader references; defer unless OWL needs another retrieval method.
- LineThunderFactory/epstein-documents-analysis-system: AI analysis platform; defer until outputs and citations are inspected.
- jpuckett11/ObsidianGroup_Epstein: documentary/evidence-edition repository; defer because the site should not import claims without source-file review.
- Entertainment, templates, or thin description entries were not promoted.

## Implementation notes

- `data/source-index.json` now has an `external_sources` array.
- `js/app.js` renders `externalSourcesGrid` separately from official sources.
- `js/command-palette.js` searches external sources as their own type.
- `index.html` keeps the official source audit first, then shows the external source review layer beneath it.