In competitive gaming and sports, "lopsided" matches—where one team is significantly stronger than the other—often fail to engage audiences because the outcome feels predetermined. To combat this, designers introduce "Game Changers": high-reward objectives that allow an underdog to make a sudden comeback. However, if the reward is too low, the underdog remains hopeless; if it is too high, the rest of the game's mechanics become irrelevant. Researchers from Peking University have developed a mathematical model to find the "Goldilocks zone" for these mechanics, maximizing audience excitement through the lens of "surprise."
Quantifying Audience Engagement: The Surprise Metric
The authors define "Surprise" based on the work of Ely et al. (2015), where surprise is the absolute change in the probability of a team winning from one state to the next. The goal of a game designer is to find a reward value ($x$) that maximizes the sum of these changes over the course of a match.
As shown in the belief curves above, a high-surprise match maintains uncertainty and dramatic shifts in win probability, whereas a low-surprise match sees the win probability flatline early, leading to a predictable and unengaging conclusion.
The Quidditch Model: A Markovian Approach
To derive a closed-form expression for expected surprise, the researchers first modeled the fictional sport of Quidditch as a time-homogeneous Markov chain. In this model, each round presents probabilities for standard goals ($p$) and the game-ending Golden Snitch ($q$).
By using eigendecomposition of transition matrices, the authors derived a simple estimator for the optimal Snitch score: $\tilde{x} \approx \frac{1}{2q}(\frac{1-p}{p} - 1)$. This formula suggests that the optimal reward should be worth exactly enough to offset the lead the stronger team is expected to build through standard play.
From Magic to MOBAs: The Snowball Effect
Moving beyond the static probabilities of Quidditch, the authors applied their framework to Multiplayer Online Battle Arenas (MOBAs) like League of Legends and DOTA 2. Unlike Quidditch, MOBAs are dynamic; winning a fight grants wealth (gold/items), which increases the probability of winning the next fight—a phenomenon known as the "snowball effect."
Calibrating their model with data from over 14,000 professional matches, the researchers found:
- The Zero-Reward Paradox: For perfectly balanced teams in a static game, the optimal reward for a "Game Changer" is actually zero. Adding high-variance items in a fair fight can actually reduce total surprise.
- Snowballing Requires Rewards: In MOBAs, even balanced teams benefit from a "Game Changer" reward (approx. 2800 gold in LOL and 5200 in DOTA 2). This is because early leads tend to snowball, and a mid-to-late game objective (like Baron Nashor or Roshan) is necessary to prevent the win probability from reaching 100% too early.
Conclusion
The study demonstrates that the optimal "comeback" reward scales linearly with the strength gap between competing teams. While the current model assumes static player strategies, it provides a foundational framework for eSports designers to tune high-stakes mechanics. Future work may explore "dynamic" rewards that scale automatically based on a team's deficit, ensuring that no matter the skill gap, the next "Game Changer" is always within reach.
Summary updated