About PennNews

PennNews is an automated local news site that publishes crime-related articles based on official Pennsylvania State Police Public Information Release Reports (PIRRs). Our goal is to provide timely, accurate coverage of crime news while maintaining the highest standards of journalistic integrity.

How It Works

Our system operates through a fully automated pipeline:

1. Ingestion: We monitor the official PSP website for new Public Information Release Reports (PIRRs) published by each station across Pennsylvania.
2. Parsing: Each PDF release is automatically parsed to extract key information: date, location, incident type, individuals involved, and charges.
3. Classification: Releases are classified as either "person-named" (mentioning specific individuals) or "non-person-specific" (general incidents, crashes, safety notices).
4. Draft Generation: An AI assistant generates article drafts using only facts present in the original police release, preserving allegation framing language.
5. Review: Every article that names a specific individual is reviewed by a human editor before publication. Non-person-specific items may be auto-published.
6. Publication: Approved articles are published with clear attribution to the original PSP release.

Editorial Standards

Allegation Framing

Every article preserves the "allegation" nature of police releases. You'll see language like:

• "Police say..."
• "According to the release..."
• "Charged with..." (not "guilty of...")
• "Allegedly..."

Charges are allegations, not convictions. All individuals are presumed innocent until proven guilty in court.

Source Transparency

Every article includes:

• A link to the original PSP press release
• Clear identification of the issuing station
• Disclosure that the article was AI-assisted and human-reviewed

Corrections Policy

If charges are later dropped, if the PSP updates a release, or if a court outcome becomes known, we append a correction or update to the original article rather than silently editing it. This maintains a transparent record of what was reported and when.

AI Assistance Disclosure

This site uses artificial intelligence to assist in generating article drafts from police press releases. Here's what that means in practice:

• The AI writes first drafts based solely on facts in the official police release. It does not invent details, fill in gaps, or make assumptions.
• Human editors review every article that names a specific individual. Editors verify facts against the source, correct errors, and ensure proper framing.
• The system is designed with defense in depth: even if the AI classification misses a person-named article, an independent NER (Named Entity Recognition) check catches it and diverts it to human review.
• All AI output is logged for reproducibility. We can always show exactly what the AI generated and what the source text was.

Data Preservation

PIRRs are only kept on the PSP website for approximately 30 days before being removed. We archive the original PDFs alongside our generated articles, ensuring that the permanent record of these public documents is preserved.

What We Don't Do

• We do not publish booking photos (PSP doesn't release them, and we don't scrape mugshot sites)
• We do not request or attempt to access exempt investigative records
• We do not publish information beyond what's in the official public releases
• We do not silently edit or remove articles — corrections are appended transparently

Technical Details

For those interested in the technical implementation:

• Ingest: Python-based scraper monitoring PSP's PIRR pages, with URL allowlisting and respectful rate limiting
• Parsing: PDF parsing using pdfplumber in an isolated worker with strict resource limits
• Classification: Rule-based classifier with defense-in-depth NER verification using regex-based named-entity patterns (spaCy is not currently deployed)
• LLM: Claude models via OpenRouter API for draft generation
• Database: PostgreSQL with full audit logging and versioning
• Storage: S3-compatible object storage for PDFs with versioning enabled

Contact

For questions, corrections, or takedown requests, please contact us. We take our responsibility as a crime news publisher seriously and are committed to accuracy, fairness, and transparency.

Last updated: July 2026