PAiNT by Numbers #1.5 — Epstein Files Update

🔹 What’s new as of September 11, 2026

• New batches of documents were added to the DOJ Epstein disclosures site in the weeks leading up to September 11, 2026. These batches include additional court filings, supplemental exhibits, and newly processed media (images and short video clips).
• Targeted supplements: recent uploads appear to focus on materials tied to state‑level inquiries and previously withheld exhibits that required additional redaction review.
• Active production: filings from court dockets and last‑minute submissions (redacted) indicate the archive remains a living dataset rather than a closed dump.
• Legislative pause: the House’s decision to postpone action until after the election reduces immediate congressional pressure, but it does not halt judicial or DOJ administrative processes that can produce more releases.

🔹 The newest file types to watch for

• Court filings and docket supplements — PDFs with exhibits attached. These often contain the most legally relevant material.
• Exhibit bundles — large ZIPs or grouped PDFs containing images, scanned documents, and transcribed interviews.
• Media packages — images and short video clips extracted from devices or surveillance.
• Indexed spreadsheets — metadata exports that list filenames, Bates ranges, and basic metadata (date, source, file type). These are gold for researchers because they let you triage without opening every file.
• Redaction logs — sometimes included as separate documents; they explain why certain content was withheld.

🔹 How to find the newest files on the DOJ site (step‑by‑step)

1. Start at the DOJ Epstein disclosures landing page.
   • Look for a “Latest Releases” or “Recent Uploads” section. The DOJ often lists batches by date or docket number.
2. Use the site’s search or filter tools.
   • Filter by date range (most recent 30 days) and by file type (PDF, ZIP, image).
3. Download the metadata index if available.
   • Many releases include a CSV or Excel index. Download that first — it saves time.
4. Sort the index by upload date or Bates range.
   • Identify the newest Bates ranges and note the filenames and docket references.
5. Open docket‑level PDFs for context.
   • Docket entries often explain why a batch was produced (e.g., “supplemental production per court order”).
6. Capture redaction logs and exhibit lists.
   • These explain what was withheld and why — essential for interpreting gaps.
7. Record the exact filenames and timestamps.
   • Save them in a research log so you can cite the precise file if you publish.
8. If the DOJ site is slow or unsearchable, use the site: query on search engines (e.g., site:justice.gov/epstein "Bates"), then cross‑check filenames against the DOJ index.

🔹 The newest files (what to prioritize)

1. Exhibit lists and indexes — they tell you what’s inside a bundle without opening every file.
2. Witness interview transcripts — often reveal timelines and corroborating details.
3. Financial records and travel logs — these are structured and easy to parse with AI.
4. Media (images/videos) — visual evidence can confirm locations and timelines; catalog them separately.
5. State‑level supplements (New Mexico, Florida) — these can fill geographic gaps in the federal archive.

🔹 How to use AI in your research — practical, ethical, and reproducible workflows

1. Triage and prioritization (fast)

• What to do: feed metadata indexes and exhibit lists into an AI assistant to rank files by likely relevance (e.g., contains names, dates, or media).
• Why it helps: reduces the initial review load from millions of pages to a manageable prioritized queue.

2. OCR cleanup and searchable text

• What to do: run OCR on scanned PDFs with a high‑quality engine, then use AI to correct common OCR errors and normalize dates and names.
• Why it helps: improves search accuracy and entity extraction.

3. Entity extraction and linking

• What to do: use AI to extract names, organizations, locations, phone numbers, and dates; then link entities across documents to build a network graph.
• Why it helps: reveals clusters and repeated associations that human readers might miss.

4. Redaction mapping

• What to do: use AI to detect redaction patterns (e.g., which pages and which redaction types appear most often) and produce a redaction heatmap.
• Why it helps: shows where blind spots are concentrated and where FOIA or court pressure might yield the most new information.

5. Media analysis

• What to do: run image‑forensics and reverse‑image checks; use AI to transcribe audio and to extract timestamps and metadata from video files.
• Why it helps: confirms provenance and links media to events.

6. Timeline normalization

• What to do: extract dates and times from all documents, normalize to a single timezone, and use AI to flag conflicting or ambiguous entries for human review.
• Why it helps: builds a coherent chronology across jurisdictions.

7. Summarization and human‑in‑the‑loop verification

• What to do: generate concise summaries of prioritized documents, then have a human reviewer verify and annotate. Keep the human reviewer as the final gate for publication.
• Why it helps: speeds reporting while preserving accuracy and ethical oversight.

🔹 Example AI workflow (step‑by‑step)

1. Download metadata index CSV from DOJ release.
2. Run a script to extract filenames and Bates ranges.
3. Use an AI model to score each file for relevance (0–100) based on keywords and metadata).
4. Pull the top 1% of files for manual review.
5. Run OCR and entity extraction on those files.
6. Build a network graph of entities and run community detection to find clusters.
7. Draft summaries for each cluster and assign human reviewers.

🔹 Message from Steven Smith

🔹 What the House pause means for researchers

• Short term: less immediate congressional pressure may slow high‑profile hearings, but it does not stop judicial or administrative releases. Expect DOJ and court‑ordered supplements to continue.
• Medium term: the pause gives researchers time to build robust, reproducible pipelines and to pressure state repositories (New Mexico, Florida) for complementary records.
• Long term: when the House resumes, new legislative requests could force additional targeted releases; having a clean, auditable research trail will make your reporting more credible.

🔹 Quick checklist for readers (actionable)

• Download the latest DOJ metadata index.
• Sort by upload date and Bates range.
• Run OCR on scanned PDFs before searching.
• Use AI to prioritize and extract entities, but verify everything manually.
• Catalog media separately and run forensic checks.
• Reach out to state repositories for complementary records (New Mexico, Florida).
• Keep a public research log with filenames, dates, and notes.

🔹 Closing: Numbers, Tools, and Responsibility