Using LAG to get the previous non-identical value while not considering certain columns

This may be an extension to the question
Using LAG function to obtain value that is not necessarily in the previous row

I have a table my_table that I would like to create a new column previous_order_date is based on the episode_id, file_id, and order_date. There is another column offer_id that I don't want to consider for the previous_order_date calculation

Query:

create table my_table
(episode_id number,
  file_id number,
  offer_id number,
  order_date date
);

insert all
 into my_table values (1234,3000,700, to_date('2020/10/10','YYYY/MM/DD'))
 into my_table values (1234,3000,800, to_date('2020/10/10','YYYY/MM/DD'))
 into my_table values (1234,3000,900, to_date('2020/10/10','YYYY/MM/DD'))
 into my_table values (1234,3000,800, to_date('2020/10/11','YYYY/MM/DD'))
 into my_table values (1234,3000,900, to_date('2020/10/11','YYYY/MM/DD'))
 into my_table values (1234,4000,700, to_date('2020/10/10','YYYY/MM/DD'))
 into my_table values (1234,4000,800, to_date('2020/10/10','YYYY/MM/DD'))
 into my_table values (1234,4000,700, to_date('2020/10/11','YYYY/MM/DD'))
 into my_table values (4321,3000,700, to_date('2020/10/10','YYYY/MM/DD'))
 into my_table values (4321,3000,800, to_date('2020/10/10','YYYY/MM/DD'))
 into my_table values (4321,3000,900, to_date('2020/10/10','YYYY/MM/DD'))
 into my_table values (4321,3000,800, to_date('2020/10/11','YYYY/MM/DD'))
 into my_table values (4321,3000,900, to_date('2020/10/11','YYYY/MM/DD'))
 into my_table values (4321,4000,700, to_date('2020/10/10','YYYY/MM/DD'))
 into my_table values (4321,4000,800, to_date('2020/10/10','YYYY/MM/DD'))
 into my_table values (4321,5000,700, to_date('2020/10/10','YYYY/MM/DD'))
 into my_table values (4321,5000,700, to_date('2020/10/11','YYYY/MM/DD'))
 into my_table values (4321,4000,700, to_date('2020/10/11','YYYY/MM/DD'))
 into my_table values (4321,3000,700, to_date('2020/10/12','YYYY/MM/DD'))

I want to use LAG to get the previous order_date based on partitioned by episode_id and file_id, that is I don't want to get the previous row order_date if it's the same with a different offer_id.

I'm trying to get the output to be like this:

but I've tried running this query

select episode_id
       ,file_id
       ,offer_id
      ,order_date
      ,lag(order_date) over(partition by episode_id, file_id order by order_date) as previous_order_date
from my_table

however that produces this table

which takes the previous row order_date regardless of offer_id. It is quite a big query to get to this point, and I need the offer_id column for further calculations since I'm joining other tables based on that column. So I was wondering since the real-world application has order_date down to the minute if there was a way to have LAG look at previous rows that were at least 1 minute older, similar to the INTERVAL function.

I have tried selecting the episode_id, file_id, and order_date in a CTE and using LAG there but this duplicated the rows, one row with the order_date as the previous_order_date, and one row with a null previous_order_date. The same occurred when I tried to use DENSE_RANK for some reason.

My DB version is 19c

You appear to want to PARTITION BY episode_id, file_id, offer_id within the LAG:

select episode_id,
       file_id,
       offer_id,
       order_date,
       LAG(order_date) OVER(
         PARTITION BY episode_id, file_id, offer_id
         ORDER BY order_date
       ) AS previous_order_date
FROM   my_table

Or, maybe, you can use MATCH_RECOGNIZE:

select *
FROM   my_table
MATCH_RECOGNIZE(
  PARTITION BY episode_id, file_id
  ORDER BY order_date DESC, offer_id DESC
  MEASURES
    current_row.offer_id AS offer_id,
    current_row.order_date AS order_date,
    prev_day.order_date AS previous_order_date
  AFTER MATCH SKIP TO NEXT ROW
  PATTERN ( current_row {- same_day* -} prev_day? )
  DEFINE
    same_day AS TRUNC(current_row.order_date) = TRUNC(same_day.order_date)
)
ORDER BY episode_id, file_id, order_date, offer_id

Which, for your sample data, output:

fiddle