AI-Assisted Protocol Scraper & Database
Figure skating produces some of the most granular athletic performance data in all of sport. Under the ISU Judging System, every jump, step sequence, and every other element is scored individually by a panel of five to nine judges as well as a technical panel. In addition, each judge assesses the overall performance with respect to artistic "Component" scores. A single competition can generate thousands of discrete data points, but the catch is almost none of it is practically accessible.
The ISU publishes official competition results as individual, static PDF files — one per event, spanning decades of competitions. Each document is dense with nested tables, shifting layouts, inconsistent column structures and names that defeat virtually every conventional data parser. Resources like Skatingscores.com do an exceptional job providing breakdowns for individual performances, but deep cross-competition analysis — the kind that asks questions across thousands of events and millions of data points — requires something different entirely.
That's what drove us to build a custom extraction pipeline combining machine learning and natural language processing. The system works through thousands of historical ISU PDFs, reliably identifying and extracting every score and infraction from every judge for every element and component regardless of document formatting inconsistencies. The result is a single, unified database of over 25,000 performances, structured and ready for rigorous data science.
Interested in a particular dataset?
Whether you need historical scoring records from a specific discipline, full panels from past Grand Prix events, or customized data exports for statistical research—let us know. We can compile and deliver custom datasets tailored to your specific analysis goals.