How strong of a predictor are regular season performances towards the Capcom Cup winner?

• Capcom Cup 12 had 48 players and a $1,000,000 prize.
• The 2025 Capcom Pro Tour gave out “Premier Points” at 8 offline events.
• More points are given for better results at the qualifying events.
• If points measure skill, high-point players should finish higher.

Story roadmap: Question → Statistical idea → Math → R → Visualization → Insight

• Bootstrap confidence interval - Shows how much correlation could change with a different sample of players.
• Prediction interval - Shows how well points predict one specific player’s finish.
• Spearman Rank Correlation - Measures whether higher points go with better finishes, even when the data is ranked, tied, or not in a straight line.
• Data: 48 There are 12 Capcom Cup players, each matched to their CPT 2025 Premier Points.

• The top 16 have exact places, while the other players get the midpoint of their tier (e.g. all 24 group-stage exits get 36.5).
• 20 of 48 players had no row in the points table. We count them as 0 points.

• Bootstrap percentile interval - Resample the \(n\) players with replacement \(B = 10{,}000\) times and compute \(\hat\rho^{*}_b\) each time:

\[\text{95\% CI} = \left[\hat\rho^{*}_{(0.025)},\ \hat\rho^{*}_{(0.975)}\right]\]

• Regression line and 95% prediction interval for one new player with \(x_0\) points:

\[\hat y_0 \pm t_{n-2,\,0.975}\; s\sqrt{1 + \frac{1}{n} + \frac{(x_0-\bar x)^2}{\sum (x_i-\bar x)^2}}\]

• The “1 +” shows us why a prediction interval for one player is wider than the confidence interval for the average.

• We rank both variables, and then we take the Pearson correlation of the ranks:

\[\rho_s = \frac{\sum_{i}(R_i-\bar R)(S_i-\bar S)}{\sqrt{\sum_i (R_i-\bar R)^2 \sum_i (S_i-\bar S)^2}}\]

• With no ties, this simplifies to:

\[\rho_s = 1 - \frac{6\sum_{i=1}^n d_i^2}{n(n^2-1)}, \qquad d_i = R_i - S_i\]

• \(R_i\) is player \(i\)’s points rank, and \(S_i\) is their Capcom Cup finish rank. A negative \(\rho_s\) here would mean more points go with a better (lower-numbered) finish.

Test: \(H_0:\rho_s = 0\) vs. \(H_a:\rho_s \neq 0\).

# Spearman correlation; ties are handled, so exact is FALSE.
cor.test(df$points, df$cc12_midrank, method = "spearman", exact = FALSE)

# Bootstrap 95% CI for rho
boot_rho <- replicate(10000, {
  i <- sample(nrow(df), replace = TRUE)
  cor(df$points[i], df$cc12_midrank[i], method = "spearman")
})
quantile(boot_rho, c(0.025, 0.975))

# Regression and 95% prediction interval
fit <- lm(cc12_midrank ~ points, data = df)
predict(fit, data.frame(points = c(0, 200, 400, 770)), interval = "prediction")

The interval sits close to 0; the correlation is negative, but it is only barely distinguishable from “no relationship”.

Since ranks below 1 are impossible, the lower end of the interval implies “could be champion”.

• Treating missing players as zero points matters, since when they are dropped the CI includes 0.
• Premier winners get no points for winning, and so the best finishers can look like players with low points.

• Winners get no points - Winners of Premier events qualify for Capcom Cup directly, so the points table understates them (MenaRD shows 50, Sahara 90).
• Range restriction - only 48 players can reach Capcom Cup, and many got in because of points; we cannot see how low-point players would do.
• Zeros - 20 players had no row in the table and are counted as 0.
• Coarse outcome - since most finishes are ties inside a tier, midranks are a simplification.
• One event, one season - a single-elimination tournament style top 16 has a lot of luck/outside factors (crowds, pressure) involved.
• Points snapshot - points shown are from the start.gg table pasted into this analysis, but other reputable sources like Liquipedia showed slightly different totals for some of the players.

• Premier Points and the Capcom Cup 12 finish are weakly related: Spearman rho = -0.3, 95% bootstrap CI (-0.54, -0.01).

• Points only explain about 8% of the variation in finishes.

• A 95% prediction interval for a 400-point player runs from about rank -7 to rank 46, which presents almost the whole field.

• Takeaway - while the regular season shows who is a good player, the highest level of competition in a Street Fighter bracket is difficult to predict.

• CPT 2025 global leaderboard: start.gg, Capcom Pro Tour 2025 standings
• Capcom Cup 12 results: EventHubs, “Capcom Cup 12 results” (March 2026); Liquipedia, Capcom Cup 12
• Format and qualification rules: Capcom Pro Tour official site; Liquipedia, Capcom Pro Tour 2025
• R packages: tidyverse, plotly, knitr
• James, Witten, Hastie, Tibshirani. An Introduction to Statistical Learning.
• GeeksForGeeks Spearman, Bootstrap CI, and Prediction Interval explanations: GeeksForGeeks