2025–2026 Delaware wrestlers. Rating estimate, not a lock.
versus
is projected at . This is a rating estimate, not a lock.
This tool estimates P(A beats B) from Glicko-2 ratings built on Delaware bout files. It is a model, not a forecast of what must happen on the mat.
Boys and girls are rated in separate pools. The search list is wrestlers with a scored 2025–2026 bout. Earlier seasons and off-season results still feed the rating.
Each wrestler has a rating (skill estimate, starts at 1500) and an RD (rating deviation). RD is uncertainty: lots of scored bouts pull RD down; long gaps without matches let it rise again.
After every week of results, winners move up and losers move down. The size of the move depends on the opponent. Beating a higher-rated wrestler is a bigger gain than beating a much lower-rated one. Beating someone with a high RD (noisy rating) counts for less than beating someone whose rating is well established.
Not every bout moves the rating the same amount. Better wrestlers tend to compete in the off-season, so those results — and quality out-of-state wins — count more.
Ratings are converted to Glicko’s internal scale (divide by 173.7178, centered on 1500). The chance A beats B is the Glicko expected score, using B’s RD:
g(φB) = 1 / √(1 + 3 φB² / π²)
P(A beats B) = 1 / (1 + exp( − g(φB) × (μA − μB) ))
μ is the scaled rating and φ is the scaled RD. If A is clearly stronger, μA − μB is large and the probability is high. If B’s RD is large, g(φB) shrinks, which pulls the probability back toward 50% — the model refuses to be overconfident about an uncertain opponent.
Confidence (high / medium / low) is based on the worse of the two RDs, not on the size of the percentage.
