lead计算留存率

lead 计算次日留存率，一般没有问题，但是如果统计第3日，第4日及以后，会有一点点的小问题，会存在逻辑不严谨；

正确代码

with tmp_da_uid_project as (
    select 
    ds
    ,to_date(ds,'yyyymmdd') as dt 
    ,project_id
    ,uid 
    ,count(distinct case when event_id = 2001 then uid end) as uv 
    ,count( case when event_id = 2001 then uid end) as pv
    ,count(distinct uid ) as uv_all
    ,count( uid ) as pv_all
    from tmp_zj_project_utrack_detail_quickbi 
    where event_id = 2001
    group by 
    ds
    ,project_id
    ,uid 
)

, tmp_da_uid_project_ext as (
  select 
 *
,datediff(t1.ds_login_lead,t1.dt,'dd') as diff_login_lead_project
,datediff(nvl(t1.ds_login_lead2,t1.ds_login_lead),t1.dt,'dd') as diff_login_lead2_project
  from(
    select 
    ds 
    ,dt 
    ,project_id
    ,uid 
    ,uv 
    ,pv 
    ,uv_all 
    ,pv_all 
    ,lag(dt,1)over(partition by project_id,uid order by dt asc  ) as ds_login_lag
    ,lead(dt,1)over(partition by project_id,uid order by dt asc  ) as ds_login_lead
    ,lead(dt,2)over(partition by project_id,uid order by dt asc  ) as ds_login_lead2
    from tmp_da_uid_project
   )t1

)

select 
ds
,project_id
,count(distinct uid ) as uv 
,count(distinct case when diff_login_lead_project = 1 then uid end) as 次日
,count(distinct case when diff_login_lead2_project = 2 then uid end) as 第3日
from tmp_da_uid_project_ext 
where project_id = '2178'
and uid = '1833905061400054'
group by 
ds
,project_id

错误代码示例截图：
明细数据信息

正确统计

错误统计