Major Site Updates — February 2026
We’ve shipped a series of significant upgrades to Epstein Exposed over the past week. Here’s everything that’s new and how to use it.
Hybrid Search with 2.67 Million Vector Embeddings
Every document chunk in our database now has a vector embedding, enabling semantic search alongside traditional keyword search. This means you can search by meaning, not just exact words.
How to use it:
- Go to epsteinexposed.com/search
- Type any query — the system automatically runs keyword (BM25) + semantic (vector similarity) + Cohere AI reranking
- Results show badges: keyword (blue), semantic (purple), or both (green)
- Try natural language queries like “financial connections between Epstein and Deutsche Bank” instead of just keywords
Stats: 2,669,382 chunks embedded with OpenAI text-embedding-3-small (1536 dimensions)
Ask AI — Document Q&A
A new Ask AI button appears on document pages. Click it to ask questions about any specific document and get cited answers powered by Claude.
How to use it:
- Open any document page
- Click the gold “Ask AI” button in the bottom-right corner
- Ask questions like “What financial transactions are described in this document?” or “Who is mentioned in this deposition?”
- The AI will cite specific passages from the document in its response
Mobile Network Graph Overhaul
The network graph is now fully usable on phones and tablets.
What changed:
- Compact header — controls that cluttered the mobile screen are now behind a “Filters” toggle
- Touch-optimized — tap nodes to see details in a bottom sheet, pinch to zoom, no accidental page scrolling
- Performance-tuned — mobile defaults to ~400 nodes (vs ~1,200 on desktop) for smooth interaction
- Bottom sheet details — tapping a person opens a slide-up panel with their stats, connections, and dossier link
- iOS Safari compatible — proper viewport handling with scroll locking
- Desktop improved too — graph area is 78% larger with filters collapsed by default
EFTA Document ID Search
You can now search by EFTA document ID directly. Previously, typing EFTA01249597 in search returned nothing because the search engine couldn’t handle document IDs.
How to use it:
- Go to the search page or the documents browser
- Type any EFTA ID (e.g.,
EFTA01249597,EFTA00020457) - The system detects ID-shaped queries and does a direct database lookup instead of full-text search
- Results appear instantly
Document Filter Fixes
Fixed 11 broken filter buttons on the documents page:
- DS9 (Dataset 9) was missing from the Source filter buttons — now shows as “DOJ Data Set 9”
- 7 other source filters had blank labels (showing as unmarked buttons)
- 3 category filters were also mislabeled
- All source and category filters now display correctly
Thanks to J4Vic for reporting the DS9 button issue!
Enhanced Document Review System
The review desk has been upgraded:
- Person name suggestions — the system automatically detects person names in OCR text and suggests them for tagging
- Location spotting — geographic references are automatically highlighted
- New review tags — financial records, witness testimony, travel records, and more granular categorization
- Improved fallback — the “Start Reviewing” button now reliably serves documents (fixed empty queue issue)
Technical Artifacts Browser
New page at /artifacts — browse extracted entities, redaction analysis scores, and OCR quality metrics across documents. This is useful for researchers doing quantitative analysis of the document collection.
Self-Hosted Forum
The community forum has been migrated to a self-hosted server on DigitalOcean for better performance and reliability. Along with the migration, several new plugins have been activated:
- Voting — upvote topics and posts to surface the best research
- Gamification — earn badges and recognition for contributions
- Reactions — react to posts beyond just likes
- Calendar — community events and research milestones
- Data Explorer — run custom queries on forum data (for power users)
- Solved — mark definitive answers on research questions
Investigative News Section
Seven new investigative articles published at /news, covering:
- Bruce Moskowitz — Epstein’s physician and Mar-a-Lago Trio member
- Dana Burns — dual assistant to Maxwell and Brunel
- Peter Mandelson arrest, Deepak Chopra emails, Brian Vickers ties
- Congress six names analysis, FBI stand-down pattern (1996-2019)
All articles cite specific document IDs — click through to verify every claim.
Data Quality
- 249 garbage person entries removed — MySpace usernames, generic court roles, single first names
- 26,410 false document links deleted from a fabricated “Richard Epstein” entry
- 40 persons received new detailed descriptions backed by document evidence
- 18 wrong person photos corrected (thanks to forum member Wja for the audit)
- Person count: 1,723 verified identities
- Document count: 2.1M+ with full OCR text searchable
Full Changelog
For the complete list of changes, visit epsteinexposed.com/changelog.
Questions? Bugs? Post in Bug Reports or Feature Requests.
Updated February 25, 2026
Report Inaccuracies
You can now report issues directly on any person or document page. Click the “Report Issue” button to flag:
- Wrong photo — the image shows the wrong person
- Wrong bio/description — factual errors in the summary
- Wrong person linked — a document is incorrectly tied to someone
- Bad OCR text — garbled or incorrect scanned text
- Broken link/PDF — dead links or inaccessible files
- Incorrect dossier info — errors in AI-generated dossiers
Reports go directly to the admin team with email notifications for fast triage.
Review Guide
New comprehensive guide at /review/guide covering:
- Rating Decision Tree — step-by-step flowchart to classify any document
- Concrete examples for each rating (Significant, Routine, Junk, Unreadable, Needs Expert)
- All 13 review tags with descriptions
- Person spotting tips — when to tag existing people vs. suggest new ones
- Keyboard shortcuts reference table
- FAQ — answers to common reviewer questions
Inspired by feedback from forum member Moth — thank you!
Person Alias Search
The person search (used in the review desk and across the site) now matches known aliases. For example, searching for a known alias will show the person with an “(alias: …)” hint.
Review Desk UX Improvements
- Rating help toggle — click the
?button next to “Rating” to see concrete examples for each category. This resolves the “Routine vs Junk” confusion. - Suggest New Person is now highlighted in a green box with an icon — much easier to find
- Section dividers (“People” / “Locations”) in the review panel for clearer visual grouping
- Guide link added to the review desk and keyboard shortcuts modal