C3PA Explorer

Episodes & demos

• S01E01 Playhttps://jpeckenpaugh.github.io/c3pa-explorer/ ► Watch (opens in a new tab)

About

C3PA Explorer is a proof-of-concept web app for exploring the C3PA Dataset (400 expert-annotated privacy policies, CCPA/CPRA) at both the source level (verbatim annotator spans) and the derived level (deterministically re-extracted text units). It ships two runtimes sharing one SPA and SQLite schema: a FastAPI edition with a Python backend over a normalized SQLite database, and a browser edition running entirely in-browser on sql.js (SQLite compiled to WebAssembly) against a prebuilt, signed snapshot. The derived layer is produced by backend/extract.py, a fully deterministic rule-based pipeline (no trained models) that drops chrome, extracts leaf text blocks, merges lead-ins with lists, splits sentences with abbreviation/URL/email guards, classifies units as sentence or typed fragment, and performs 3-tier annotation re-alignment. A dashboard shows corpus stats and label distributions; documents can be browsed and searched, document detail has Source/Sentences/Fragments tabs with clickable alignment provenance, and a unit explorer supports filtering and exporting timestamped ZIP archives with partition files and meta.json. The browser edition auto-loads a signed snapshot (verified against envelope and db hashes in builds.json) and runs queries in a Web Worker with no server or network calls after download; a nav modal can switch to a FastAPI backend. Snapshots can be rebuilt on Google Colab via the included notebook, and tests cover backend API, frontend units, and FastAPI/ sql.js parity.

Summary

A proof-of-concept web app for exploring the C3PA Dataset (400 expert-annotated CCPA/CPRA privacy policies) at both the source level (verbatim annotator spans) and the deterministically re-extracted derived level.

Tech

• JavaScript
• Python
• FastAPI
• sql.js
• WebAssembly
• Bootstrap 5
• SQLite
• Jupyter

Tags

• annotation
• dataset-explorer
• fastapi
• nlp
• privacy-policies
• sql.js