152 lines
9.8 KiB
HTML
152 lines
9.8 KiB
HTML
<!doctype html>
|
||
<html lang="zh-CN">
|
||
<head>
|
||
<meta charset="utf-8">
|
||
<meta name="viewport" content="width=device-width, initial-scale=1">
|
||
<title>FishDice 随机逻辑解释</title>
|
||
<link rel="stylesheet" href="style.css">
|
||
</head>
|
||
<body>
|
||
<main class="shell">
|
||
<div class="topbar"><a href="index.html">← 返回总览</a><span class="brand">随机逻辑解释</span></div>
|
||
|
||
<section class="hero">
|
||
<h1>FishDice 随机逻辑：从权重候选到分层池漏斗</h1>
|
||
<p class="subtitle">这页解释当前仓库里两套随机决策：一套服务 Lucky Dice 候选选择，另一套是更通用的 Pool Funnel 分层随机池。它们共同的设计目标是：随机可注入、权重可解释、结果可回放、失败可观测。</p>
|
||
</section>
|
||
|
||
<section>
|
||
<h2>一句话模型</h2>
|
||
<div class="note good">随机并不是“从列表里随手挑一个”。框架会先筛候选、算权重、只在最高优先级集合里抽取，再把随机值、选中项、状态变化和后续决策写入结果对象或 Trace。</div>
|
||
<div class="grid-3">
|
||
<div class="metric"><div class="value">IRandomSource</div><div class="label">随机源由外部注入，测试可以用固定序列复现结果。</div></div>
|
||
<div class="metric"><div class="value">Weight</div><div class="label">候选通过权重参与抽取，权重为 0 的项不会进入抽奖池。</div></div>
|
||
<div class="metric"><div class="value">Trace</div><div class="label">Pool Funnel 会记录每层权重、随机值、命中结果、状态前后差异。</div></div>
|
||
</div>
|
||
</section>
|
||
|
||
<section>
|
||
<h2>流程图：Pool Funnel 随机决策</h2>
|
||
<div class="diagram-panel">
|
||
<div class="logic-flow" aria-label="Pool Funnel 随机流程图">
|
||
<div class="logic-node">
|
||
<strong>1. 输入请求</strong>
|
||
<span><code>RandomPoolDefinition</code> + <code>PoolFunnelRequest</code> 进入服务。</span>
|
||
</div>
|
||
<div class="logic-node">
|
||
<strong>2. 解析入口层</strong>
|
||
<span>检查 <code>EntryLayerIds</code>，过滤不可用层，按 priority 选入口。</span>
|
||
</div>
|
||
<div class="logic-node">
|
||
<strong>3. 读取状态</strong>
|
||
<span>从 <code>IRandomPoolStateStore</code> 取当前层 luck value 和 roll count。</span>
|
||
</div>
|
||
<div class="logic-node">
|
||
<strong>4. 计算权重</strong>
|
||
<span>默认公式：<code>BaseWeight + Factor * LuckValue</code>，小于 0 归零。</span>
|
||
</div>
|
||
<div class="logic-node break">
|
||
<strong>5. 筛候选池</strong>
|
||
<span>只保留权重大于 0 的结果，再取最高 priority 的结果组。</span>
|
||
</div>
|
||
<div class="logic-node alt">
|
||
<strong>6. 权重抽取</strong>
|
||
<span><code>Next(0, totalWeight)</code> 得到随机游标，累加权重命中结果。</span>
|
||
</div>
|
||
<div class="logic-node alt">
|
||
<strong>7. 执行动作</strong>
|
||
<span>把选中结果声明的 actions 交给 dispatcher，默认只排队成功结果。</span>
|
||
</div>
|
||
<div class="logic-node alt">
|
||
<strong>8. 沉淀状态</strong>
|
||
<span>默认将 <code>LuckValue += LuckValueDelta</code>，并让 roll count +1。</span>
|
||
</div>
|
||
<div class="logic-node alt">
|
||
<strong>9. 判断去向</strong>
|
||
<span>结果配置决定 Stop 或 NextLayer；NextLayer 会进入下一层循环。</span>
|
||
</div>
|
||
<div class="logic-node stop">
|
||
<strong>10. 输出结果</strong>
|
||
<span>返回 outcome、每层详情和 <code>PoolFunnelTrace</code>，便于复盘。</span>
|
||
</div>
|
||
</div>
|
||
</div>
|
||
<p>这个流程的关键不是“随机算法多复杂”，而是把随机前后的上下文都保留下来：入口层为什么可用、候选为什么被过滤、某个结果为什么权重更高、抽中的随机游标落在哪个区间。</p>
|
||
</section>
|
||
|
||
<section>
|
||
<h2>框架图：随机池组件关系</h2>
|
||
<div class="diagram-panel">
|
||
<div class="framework-map" aria-label="随机池框架图">
|
||
<div class="framework-column">
|
||
<div class="framework-title">配置与输入</div>
|
||
<div class="framework-box"><strong>RandomPoolDefinition</strong><p>定义 pool id、入口层列表、所有随机层。</p></div>
|
||
<div class="framework-box"><strong>RandomLayerDefinition</strong><p>定义 layer id、优先级、可用性 key、结果集合。</p></div>
|
||
<div class="framework-box"><strong>RandomResultDefinition</strong><p>定义 result key、priority、基础权重、幸运值因子、状态增量和下一步决策。</p></div>
|
||
</div>
|
||
|
||
<div class="framework-column">
|
||
<div class="framework-title">核心编排</div>
|
||
<div class="framework-box"><strong>PoolFunnelRandomService</strong><p>负责入口层解析、循环 RollLayer、处理空池、处理 NextLayer、生成最终结果。</p></div>
|
||
<div class="framework-box"><strong>RollLayer</strong><p>读取状态 → 算全部权重 → 取最高优先级候选 → 权重抽取 → action → settlement → decision。</p></div>
|
||
<div class="framework-box"><strong>PoolFunnelRunResult</strong><p>输出 session、pool、每层结果、最终 outcome 和可回放 Trace。</p></div>
|
||
</div>
|
||
|
||
<div class="framework-column">
|
||
<div class="framework-title">可替换策略</div>
|
||
<div class="framework-box"><strong>IRandomSource</strong><p>控制随机值来源，线上用真实随机，测试用脚本随机。</p></div>
|
||
<div class="framework-box"><strong>ILayerAvailabilityPolicy</strong><p>判断层是否开放，默认 AlwaysAvailable。</p></div>
|
||
<div class="framework-box"><strong>IResultWeightCalculator</strong><p>计算结果权重，默认线性幸运值公式。</p></div>
|
||
<div class="framework-box"><strong>IPoolSettlementStrategy</strong><p>处理抽中后的状态变化，默认按 delta 修改幸运值。</p></div>
|
||
<div class="framework-box"><strong>IFunnelDecisionStrategy</strong><p>决定停在当前结果，还是进入下一层。</p></div>
|
||
</div>
|
||
</div>
|
||
</div>
|
||
</section>
|
||
|
||
<section>
|
||
<h2>核心公式</h2>
|
||
<p>默认权重计算来自 <code>LinearLuckValueWeightCalculator</code>：</p>
|
||
<div class="formula">weight = max(0, BaseWeight + LuckValueWeightFactor * LuckValue)</div>
|
||
<p>抽取时先求最高优先级候选的总权重，然后生成 <code>[0, totalWeight)</code> 的随机数。遍历候选并累加权重，随机数落入哪个区间，就选中哪个结果。</p>
|
||
<div class="note warning">priority 比 weight 更先发生作用：低优先级结果即使权重很高，也不会和最高优先级结果同池竞争。</div>
|
||
</section>
|
||
|
||
<section>
|
||
<h2>Lucky Dice 候选随机</h2>
|
||
<div class="step-list">
|
||
<div class="step-item"><strong>先过滤。</strong>只保留 enabled、满足 player level、满足 trigger source 的候选。</div>
|
||
<div class="step-item"><strong>再按权重抽取。</strong>权重大于 0 的候选进入池子，使用同样的累加区间方式命中结果。</div>
|
||
<div class="step-item"><strong>最后兜底。</strong>如果没有可抽候选，优先找 default candidate；仍没有时返回 <code>NormalRewardFallback</code>。</div>
|
||
<div class="step-item"><strong>补展示槽。</strong>选中候选后决定 result slot count，候选未配置时使用 Lucky Dice 骰子集默认槽数。</div>
|
||
</div>
|
||
<p>因此 Lucky Dice 的随机逻辑更轻：它没有多层状态沉淀，也不循环跳层；它主要服务“从当前可用特殊结果里选一个入口”。</p>
|
||
</section>
|
||
|
||
<section>
|
||
<h2>为什么这样设计</h2>
|
||
<table>
|
||
<tr><th>问题</th><th>设计回答</th><th>收益</th></tr>
|
||
<tr><td>测试里怎么复现随机？</td><td>随机源走 <code>IRandomSource</code> 注入。</td><td>固定随机序列即可断言选中项。</td></tr>
|
||
<tr><td>概率为什么变化？</td><td>权重计算器显式依赖基础权重和幸运值。</td><td>策划参数、保底逻辑、状态变化都能解释。</td></tr>
|
||
<tr><td>结果池为空怎么办？</td><td>Pool Funnel 返回 <code>NoWeightedResult</code>；Lucky Dice 走 default 或普通奖励兜底。</td><td>失败路径不会静默吞掉。</td></tr>
|
||
<tr><td>以后要换概率算法怎么办？</td><td>可替换 weight calculator、settlement、decision strategy。</td><td>核心编排稳定，策略可以独立演进。</td></tr>
|
||
</table>
|
||
</section>
|
||
|
||
<section class="evidence">
|
||
<h2>代码证据</h2>
|
||
<ul>
|
||
<li><span class="path">PoolFunnelRandomService.cs:52-88</span>：入口层解析、循环滚层和 NextLayer 处理。</li>
|
||
<li><span class="path">PoolFunnelRandomService.cs:117-169</span>：权重计算、最高优先级候选、随机游标、状态沉淀和决策。</li>
|
||
<li><span class="path">PoolFunnelStrategies.cs:35-74</span>：默认可用性、线性权重和 delta settlement。</li>
|
||
<li><span class="path">LuckyDiceFlowService.cs:16-49</span>：Lucky Dice 候选过滤、权重选择、默认结果兜底。</li>
|
||
<li><span class="path">LuckyDiceRollFlowService.cs:65-90</span>：只有 action 产生 <code>LuckyDiceRequestedEffect</code> 时才进入 Lucky Dice 随机选择。</li>
|
||
</ul>
|
||
</section>
|
||
|
||
<div class="footer">生成目录：<span class="path">analysis/FishDice/random-logic.html</span>。这是离线静态页面，可直接用浏览器打开。</div>
|
||
</main>
|
||
</body>
|
||
</html>
|