为什么多对一合并联接会更改数据集的排序顺序？

我有一个计算的查询row_Number()。该表在与行号的分区和排序相同的列（和顺序）上有一个聚集索引。

使用合并连接（多对一）时，需要进行排序，即使聚集索引的顺序正确。删除连接也会删除排序操作。

应该支持 row_number 计算的聚集索引：

create clustered index [ClusteredIndex_e060df3fbf464a8eb9b6ea5d46a9a5f5] on [dbo].[log1]
(
    [client] asc,
    [orderId] asc,
    [campaign] asc,
    [id] asc,
    [DateStamp] asc
)

create clustered index [ClusteredIndex_dd0ee53e050d436cba2cab7c678a39e5] on [dbo].[LiveReference]
(
    [client] asc,
    [orderId] asc,
    [campaign] asc
)

查询：

    with cr as 
(
    select distinct client, orderId,campaign
    from LiveReference
)

select e.[DateStamp]
  ,e.[campaign]
  ,e.[client]
  ,e.[orderId]
  ,e.[ad]
  ,e.[id]
  ,e.[source]
,row_number() over (partition by e.[client] ,
                        e.[orderId] ,
                        e.[campaign] ,
                        e.[id] 
                    order by e.[DateStamp]) as num
from [dbo].[log1] e 
inner join cr on 
                        e.client = cr.client
                        and e.campaign = cr.campaign
                        and e.orderId =  cr.orderId

它给出了以下计划：

删除连接也会删除排序：

select e.[DateStamp]
  ,e.[campaign]
  ,e.[client]
  ,e.[orderId]
  ,e.[ad]
  ,e.[id]
  ,e.[source]
,row_number() over (partition by e.[client] ,
                        e.[orderId] ,
                        e.[campaign] ,
                        e.[id] 
                    order by e.[DateStamp]) as num
from [dbo].[log1] e 

（我知道这也会删除由连接执行的过滤，但这并不能解释为什么排除这些行会更改顺序）

为什么排序连接的结果会不按正确的顺序排列？