Project Overview

Project Overview

PolicyWatcher is an open-source monitoring platform for observing how major digital companies update privacy policies, terms of service, AI terms, acceptable-use rules and related governance documents.

The project is intended for:

• citizens and digital-rights observers;
• researchers;
• legal and compliance teams;
• data protection officers;
• civic-tech communities;
• journalists and public-interest analysts.

Core Problem

Online policies change frequently, but those changes are often difficult to notice, compare and understand. The documents are long, fragmented across jurisdictions, sometimes dynamically rendered, and often distributed across legal hubs.

PolicyWatcher addresses this by combining:

• source retrieval;
• change detection;
• structured AI-assisted analysis;
• public evidence gating;
• operational QA;
• transparent documentation.

What Makes the Platform Different

PolicyWatcher is not only a scraper and not only an AI summarizer. The 3.5 Confidence track introduced a stricter evidence discipline:

• every public signal should trace back to source evidence;
• uncertain sources are suspended instead of silently published;
• seeded records are blocked from public views in production;
• first verified retrievals can become baselines without creating false change events;
• admin actions and Dataset QA decisions are logged.

Main Product Areas

• Public workspace: dashboard, timeline, policy cards, matrix, compare views, share pages and reports.
• Signals board: evidence-oriented ranking of source coverage, public baselines and source movement.
• Admin console: Cron Manager, Dataset QA, company registry, KPI audit, access logs and VPS service monitoring.
• Ingestion layer: five-strategy retrieval cascade with diagnostics and public evidence gates.
• VPS companion services: optional Playwright renderer and operations agent for recovery, backup and controlled updates.

Non-Certification Boundary

PolicyWatcher maps evidence. It does not certify companies and does not determine legal compliance. Automated outputs should be treated as structured signals that require human interpretation and source verification.