洗牌 – 作为实验引擎的游戏模式
1 分•作者: gok2•7 个月前
你好！
我是 Colonist 的联合创始人，Colonist 是一款手机和桌面端在线卡坦岛风格游戏。随着时间的推移，我们的玩家群体和社区已经增长到数百万，这仍然让人觉得不可思议。
我们一直在构建新的活动、地图和模式，但这需要大量的时间和精力。
重点是 Shuffle。
Shuffle 是一个简单的匹配队列，每周轮换一组地图、模式和规则集。我们用它来：
* 了解玩家真正喜欢哪些地图/模式
* 每周提供不同的配置
* 在全新的模式进入主要的休闲或排位队列之前进行测试
我们希望使用 Shuffle 来测试诸如：
* 非标准资源分配/交易比例
* 强盗/发展卡牌变体
* 替代获胜条件
* 全新的游戏模式
在底层，它基本上是一个位于匹配系统之上的小型实验引擎：
* 每个 Shuffle 选项都被建模为一个子配置（地图、规则、玩家数量）并存储在数据库中
* 一个周期性任务使用基于近期偏好的加权随机算法 + 冷却时间来选择下一个配置
* 比赛会记录以该配置为键的指标，以便我们之后可以比较不同的变体
我喜欢这种测试平台兼作“实验引擎”的想法，并认为从内部分享它可能会很有趣。
我也很好奇：
* 你认为这种系统还有哪些意想不到的用途？
* 你希望在这样的系统中测试哪些机制？
很乐意回答技术或产品问题，并分享更多细节。
查看原文
Hello hello,<p>Cofounder of Colonist here, an online Catan-style game on mobile and desktop. Over time the player base and community have grown into the millions, which still feels pretty wild.<p>We’re constantly building new events, maps, and modes, but that takes a lot of time and effort.<p>The point is Shuffle.<p>Shuffle is a simple matchmaking queue that rotates a set of maps, modes, and rule sets every week. We use it to:
- Find out which maps/modes players actually like
- Serve different configurations week by week
- Test brand-new modes before they go into the main casual or ranked queues<p>We want to use Shuffle to test things like:
- Non-standard resource distributions / trade ratios
- Robber / dev-card variants
- Alternative win conditions
- Completely new game modes<p>Under the hood, it’s basically a tiny experiment engine on top of matchmaking:
- Each Shuffle option is modeled as a sub-config (map, rules, player-count) and stored in a DB
- A periodic job picks the next config using recency-biased weighted random + cooldowns
- Matches log metrics keyed by that config so we can compare variants afterwards<p>I like this idea of a test harness doubling as an “experiment engine” and thought it might be interesting to share from the inside.<p>I’m also curious:
- What other out-of-the-box uses do you see for a system like this?
- What are some mechanics you’d want tested in a system like this?<p>Happy to answer technical or product questions and to share more details.