TSQL count vs sum distinct values - sql-server

Question

I have a slightly confusing conundrum and I have been stuck all day.
I have the following types of data ...
For each customer record I have order numbers and for each order, I have a series of package numbers and for each package number, I have a possibility of zones... Normally the math would be relatively simple if there was 1 package with 1 or more zones we just select distinct amount of seats for example.
+-----------+-------+-----+------+-------+
| customer | order | pkg | zone | seats |
+-----------+-------+-----+------+-------+
| 1 | 1 | 11 | 7 | 2 |
| 1 | 1 | 12 | 7 | 2 |
+-----------+-------+-----+------+-------+
We know customer 1 has 2 seats per package.
Here is where it gets tricky
+----------+-------+-----+------+-------+
| customer | order | pkg | zone | seats |
+----------+-------+-----+------+-------+
| 2 | 3 | 8 | 5 | 2 |
| 2 | 3 | 9 | 5 | 2 |
| 2 | 3 | 10 | 5 | 2 |
-- In the above case we know a given customer has one order #3, with three packages in the same zone each package has two seats.
| 2 | 3 | 9 | 6 | 1 |
| 2 | 3 | 9 | 8 | 1 |
| 2 | 3 | 10 | 7 | 2 |
+----------+-------+-----+------+-------+
-- Here things are confusing because the same customer, has a single order #3 (and its possible
-- both scenarios occur in one single order) with two packages 9 and 10, package 9 has two zones
-- 1 and 1 and package 10 has one zones with two seats. how do we distinguish when we simply count
-- the seats like in the first/second occurrence or when we sum the seats like in the last example.
To reiterate a single customer would have a single order each order can have many packages in it with distinct package numbers each package can have 1 or more zones and each zone can have 1 or more seats.
When the zones are the same for a single package we simply count distinct. when a single order+package has more than one zone we sum we don't count.
I can't figure out how to code the logic. Please help.
My columns are customer_no, order_no, pkg_no, zone_no and pkg_seats.
Here is a real example
+----------+-------+-----+-------+------+
| customer | order | pkg | seats | zone |
+----------+-------+-----+-------+------+
| 374 | 876 | 68 | 2 | 26 |
| 374 | 876 | 68 | 1 | 32 |
| 374 | 876 | 68 | 1 | 56 |
| 374 | 876 | 71 | 2 | 56 |
| 374 | 876 | 71 | 2 | 79 |
| 862 | 538 | 71 | 2 | 33 |
| 862 | 538 | 71 | 1 | 81 |
| 862 | 538 | 71 | 1 | 82 |
-- In the below case we simply count 2. in the above we sum.
| 575 | 994 | 68 | 2 | 34 |
| 575 | 994 | 68 | 2 | 79 |
+----------+-------+-----+-------+------+
I should add one super confusing piece. We have a series of packages that are part of other packages. For example package 68, 70 and 71 are all together and the parent package is 68.
I can't figure out the grouping.

Answer 1

with data as (
select *,
min(zone_no) over
(partition by customer_no, order_no, pkg_no) as min_zone_no1,
min(zone_no) over
(partition by customer_no, order_no, pkg_no, pkg_seats) as min_zone_no2
from T
)
select
customer_no, order_no,
sum(case when zone_no = min_zone_no1 then pkg_seats end) as seat_total1,
sum(case when zone_no = min_zone_no2 then pkg_seats end) as seat_total2
from data
group by customer_no, order_no
order by customer_no, order_no;
I've poured over your description a few times and I'm still not certain I'm on the right track. You seem to have a problem of double-counting: essentially you want a sum, but some of the rows shouldn't be included. (To "count distinct seats" is likely the wrong nomenclature here.)
My approach above is to try and identify sets of rows that involve "duplicates" and some data that will assist in counting only one of them. I'm not sure what to make of order 876 which has different numbers of seats across the three zones.