Why the Data Pipeline Matters More Than the Model

Every wrestling rating site lives or dies on its data, not its math. The fanciest model in the world is useless if it doesn't know that the Jake Smith who placed at a folkstyle tournament in November is the same Jake Smith who wrestled freestyle in May. MatMetrics spends a disproportionate amount of engineering effort on data ingestion and reconciliation for exactly that reason.

We pull from three independent sources: FloWrestling for high school and college matches with verified bout IDs, USA Bracketing for youth and freestyle/Greco events, and Boutboard for live high school tournament coverage primarily across Texas, Oklahoma, and Arkansas. Each source has its own identifiers, naming conventions, and edge cases. A reconciliation pass stitches them together into a canonical wrestler, a canonical match, and a canonical event so the rating system sees one career, not three.

It is messy work. Names come in with inconsistent spellings, weight classes get reported in different units, and team affiliations change mid-season. We maintain an ever-growing set of rules — nickname patterns, school name normalizers, weight-class boys/girls inference — to keep duplicates from polluting the database. When we get it wrong, we hear about it, and we fix it.

If you ever notice a wrestler who appears twice or a result that looks miscredited, the contact form goes straight to the engineering team. The model on top of that data only matters because the data underneath is clean.