# 分配与切换的冻结模型对照

已完成 96/96 个城市模型任务；已运行 1056/1536 条计划上限轨迹；无机会环节 96。未完成和无机会分别保留。

A原提案后正常重选；B同提案后阶段保持；C局部替换提案后正常重选；D同替换后阶段保持；E撤销当次改派并保持原任务。规则历史的替换来自新版HATT。

每城每模型每环节只取100至699步中首个合格的拟改派状态。不同历史选到的状态不同，仅同一状态内部形成因果对照。保持上限按原轨迹阶段持续时间95分位预先计算；存在策略依赖和终局截断。区间仅反映城市差异，未覆盖训练seed。单成员替换同时改变任务与联盟成员配置。

|模型|环节|对照|城市数|送达变化pp|城市95%区间|胜/平/负|平均损失变化|
|---|---|---|---:|---:|---|---|---:|
|new|approach|B-A|32|-0.31|-1.05,+0.46|4/16/12|-0.16|
|new|approach|D-C|32|+0.30|-0.69,+1.65|8/15/9|-2.12|
|new|approach|D-B|32|+0.66|-0.62,+1.80|15/10/7|-3.22|
|new|approach|C-A|32|+0.04|-0.86,+0.90|10/15/7|-1.25|
|new|approach|E-A|32|+0.32|-0.53,+1.19|12/6/14|-1.72|
|new|wait|B-A|32|+0.05|-0.19,+0.31|2/27/3|-0.75|
|new|wait|D-C|32|-0.12|-0.53,+0.15|4/25/3|+0.66|
|new|wait|D-B|32|+0.46|-0.39,+1.32|16/8/8|-0.75|
|new|wait|C-A|32|+0.64|-0.29,+1.52|14/11/7|-2.16|
|new|wait|E-A|32|-0.66|-1.68,+0.37|10/7/15|+0.09|
|previous|approach|B-A|32|-0.79|-1.44,-0.20|4/19/9|+1.22|
|previous|approach|D-C|32|+0.15|-0.99,+1.18|11/13/8|-0.09|
|previous|approach|D-B|32|-0.13|-1.27,+0.88|11/9/12|+1.00|
|previous|approach|C-A|32|-1.07|-1.81,-0.42|2/17/13|+2.31|
|previous|approach|E-A|32|+0.08|-1.29,+1.56|15/6/11|-0.41|
|previous|wait|B-A|32|-0.47|-1.07,-0.02|2/24/6|+1.00|
|previous|wait|D-C|32|-0.25|-0.72,+0.14|3/25/4|+0.59|
|previous|wait|D-B|32|+0.60|-0.39,+1.61|12/10/10|-1.44|
|previous|wait|C-A|32|+0.38|-0.57,+1.27|10/12/10|-1.03|
|previous|wait|E-A|32|-0.35|-1.39,+0.64|12/6/14|+1.78|
|rule|approach|B-A|32|+0.14|-0.49,+0.83|7/21/4|-0.06|
|rule|approach|D-C|32|-0.12|-0.71,+0.40|7/19/6|-0.12|
|rule|approach|D-B|32|+0.18|-0.61,+0.91|16/10/6|-0.59|
|rule|approach|C-A|32|+0.43|-0.31,+1.11|11/19/2|-0.53|
|rule|approach|E-A|32|-0.70|-1.50,+0.12|9/8/15|+0.72|
|rule|wait|B-A|32|+0.16|+0.00,+0.41|2/30/0|-0.06|
|rule|wait|D-C|32|+0.10|+0.00,+0.25|3/29/0|-0.25|
|rule|wait|D-B|32|-0.43|-0.98,+0.09|8/15/9|-0.12|
|rule|wait|C-A|32|-0.37|-1.00,+0.18|7/18/7|+0.06|
|rule|wait|E-A|32|+0.45|-0.23,+1.18|10/15/7|-0.94|