race-recap
Henley 2026, Day 2: Kings College sends the last American home, and Radley and St. Paul's pull clear
Sixteen crews went afloat for the Round of 16 on Wednesday, and the story of the day was the boat that didn't come back. Kings College beat Deerfield by two and a quarter lengths, and with that the…
Jul 3, 2026·Read full analysis →
race-recap
Henley 2026, Day 1: The favourites hold the Princess Elizabeth, and Deerfield is the last American standing
rankings
The British rankings, and a preview of National Schools'
rankings
The Power 10: Women's V8+ and 2V8+, after Stotesbury
rankings
The Power 10: Men's V8+ and 2V8+, after Stotesbury
analysis
Stotesbury 2026: St. Joseph's Prep owns the Schuylkill, Montclair owns the weekend's story
Recent ResultsAll regattas →
M V8+Jul 5
Henley Royal Regatta
Finals
| 1 | 6:20.0 |
| 2 | 6:22.05 |
M V8+Jul 4
Henley Royal Regatta
Semifinals
| 1 | 6:18.0 |
| 2 | 6:21.04 |
M V8+Jul 4
Henley Royal Regatta
Semifinals
| 1 | 6:29.0 |
| 2 | 6:36.05 |
M V8+Jul 3
Henley Royal Regatta
Quarterfinals
| 1 | 6:32.0 |
| 2 | 6:40.68 |
M V8+Jul 3
Henley Royal Regatta
Quarterfinals
| 1 | 6:32.0 |
| 2 | 6:36.73 |
M V8+Jul 3
Henley Royal Regatta
Quarterfinals
| 1 | 6:18.0 |
| 2 | 6:20.04 |
M V8+Jul 3
Henley Royal Regatta
Quarterfinals
| 1 | 6:17.0 |
| 2 | 6:18.00 |
M V8+Jul 1
Henley Royal Regatta
Round of 16
| 1 | 6:29.0 |
| 2 | 6:33.70 |
How RPI Works
Every rating is a full probability distribution, not a single number — and every race updates all of them at once.
The Algorithm
1
Every rating is a best guess with an uncertainty band
Each crew has a most-likely speed plus a credible interval— the range we're 90% confident covers the truth. Crews with many recent races have tight intervals; new crews start wide and tighten as data accumulates.
St. Pauls 2083 · 90% CI 2048–2118
2
Win probability comes from comparing two distributions
When two crews race, we compare their full posterior distributions — not just the means. A wide gap between curves means a confident pick; overlapping curves mean a coin-flip race.
St. Pauls 2083 vs St. Edwards 1998
Predicted: 100% chance for St. Pauls
Margin: +3.2 to +11.2 sec (90% interval at 2112m)
Posterior comparison
3
All races inform every rating at once
Rather than nudging one rating after each race, the model fits every team's skill jointly using hierarchical pooling. Beating a strong-league crew counts more — automatically — because each league's overall strength is itself learned from the races within it.
One race result → every team's posterior shifts
League means recalibrate jointly with team ratings
No K-factor — points per race emerge from the data
Accuracy & Calibration
90%
Calls correct
1,600+
H2H pairs tested
PredictedActual win rate
50–55%67%
55–60%65%
60–65%82%
65–70%78%
70–80%86%
80–90%89%
90–100%95%
When the model predicts X%, teams win ~X% of the time.
Measured on the last ~3 months of real races against the live Bayesian engine's current posterior.
Measured on the last ~3 months of real races against the live Bayesian engine's current posterior.