S15 世界赛结果 / S15 Worlds Results

请先完整解压，再双击“index.html”，选择中文或 English。无需安装 Python，也无需联网。单独把 HTML 拖出文件夹会丢失头像。

Extract the entire ZIP, then open “index.html” and choose Chinese or English. No Python installation or internet connection is required. Keep the HTML files, assets and v2 folders together.

页面操作：切换模型和指标；输入中文姓名、称号或英文名搜索；选取率柱子分为上单、打野、中单、下路、辅助五段；点击英雄、柱状图或表格可查看分路明细。参数下载按钮导出当前模式参数。

Controls: select a model and metric, search by English or Chinese name/title, and click a champion/bar/table row for role details. Pick-rate bars stack Top, Jungle, Mid, Bot and Support. The download button exports the current mode’s parameters.

全部统计以整个世界赛84局为分母（入围赛4、瑞士轮52、淘汰赛28）。选取率=选取局数/总局数；综合BP热度=(选取+主动禁用)/总局数；平均选取手数仅对被选取的情况求均值，使用1–10手或完整BP的1–20步。禁用没有可确认的最终分路。

All aggregate rates use the entire 84-game tournament (4 Play-in, 52 Swiss, 28 Knockout). Pick rate = picks/games; draft presence = (picks + active bans)/games. Mean pick position is conditional on selection, using pick rank 1–10 or full draft step 1–20. Bans have no observed final role.

阅读误差时请区分：V2赛前选取率RMSE约4.236个百分点；84局全量后验约0.815个百分点。全量后验已使用所有实际结果，不能把这个低误差当成赛前预测能力。56局更新模式也使用了部分世界赛结果。模型说明中提供时间验证、训练误差与复用淘汰赛回测的完整说明。

V2 pre-event pick-rate RMSE is about 4.236 percentage points; the all-84 posterior is about 0.815 pp. The posterior uses all actual outcomes and is not a pre-event accuracy score. The after-56 update also uses Worlds outcomes. Reports distinguish chronological validation, training error and the reused Knockout evaluation.

文件：中英文交互页面；中英文模型说明；全英雄对照表；v2下完整预测、误差及核验JSON；assets下171个本地头像与官方名称清单。此包用于查看S15结果；S16预测尚未运行。

Files: bilingual dashboards and reports, all-champion tables, detailed prediction/error/validation JSON in v2, and 171 local portraits with name/source manifests in assets. This package shows S15 results. S16 prediction has not been run.