Episode 1: Launch

Fig. 1 - A sample table containing one curling game of scores

Statistics

With our extensive database at our disposal, we delved into the analysis. Our initial focus was understanding the dynamics of a curling game. Common wisdom suggested that having the hammer increased the likelihood of scoring two points, while scoring two without the hammer posed a significant challenge. To explore this, we set out to calculate the percentage breakdown of how often specific situations occurred in a game.

Using Python scripts, we fetched data from the database, filtered it to match predefined situations, and then determined the frequency of each event. Rigorous statistical measures, including confidence intervals and margins of error, were applied to ensure the accuracy of our findings. The outcome of this research yielded CurlData’s "Typical End Result Percentages," offering insights into the frequency of blanks, single points, and 2-point scores with the hammer—providing a comprehensive picture of scoring patterns in curling games.

Fig. 2 - Typical End Result Percentages

Moving forward, our focus shifted to understanding how teams secure victories in a game. Curling is structured in "ends," each presenting an opportunity for teams to score points. Teams face decision points before each end, determining strategic goals such as "We aim for a two-point score" or "We aim to limit the opponent to one point." We envisioned providing curlers with real-time odds at these decision points, illustrating how subsequent actions could influence their chances of winning.

To compute these statistics, we designed a program to explore all possible scenarios within a curling game based on the remaining "ends" and calculate how often teams won games in each situation. The y-axis represented the remaining ends, while the x-axis covered score differentials and hammer possession. For instance, with four ends remaining, if a team possessed the hammer and led by one point, their chance of winning was determined to be 81% (Fig. 3). Rigorous confidence statistics were again applied to ensure the precision of our results, and these statistics evolved into CurlData’s "Situational Win Rates."

Fig. 3 - Situational Win Rates

Dashboard

With our initial research complete, we had not only calculated key stats but also unearthed a wealth of insights (though we're just scratching the surface in this article!). The time had come to present our findings. Originally, we envisioned creating simple graphs highlighting intriguing situations. However, recognizing the serious demand for analytics in the curling world, we pondered the idea of developing something interactive—a tool to empower curlers to work with and learn analytics for game improvement. And that's precisely what we crafted!

We designed a custom dashboard that accepts user inputs in a table format mirroring a real curling scoreboard. The concept was simple: as a game unfolds, users input scores into our scoreboard to receive instant analytics. The system generates a live graph, depicting the current and historical win rates throughout the game. Simultaneously, as users input their data, dotted lines emerge, offering future predictions. For instance, the current win rate might be 55%, but if the team scores 2 points in the upcoming end, it could rise to 77%, or if the opponent scores, it might drop to 30%. This feature became a cornerstone in training curlers to understand how win rates evolve based on end outcomes.

Fig. 4 - The CurlData Dashboard