Conditionally finding groups of duplicate column pairs, if one of the pairs contains a specific flag. Then updating all to that txId - sql

Question

I am new to SQL. Bad data got inserted into a table. The table represents items that a user has purchased. txId is an id generated upon purchase. The assumption is each item/user combination all have the same txId.
id | item | user | txId | date (ms)
-----------------------------------------------
1 | cup | bob | 10 | 1000000
2 | cup | bob | -1 | 1000000
3 | cup | bob | 10 | 1000000
4 | cup | jim | -1 | 2000000
5 | hat | bob | 10 | 1000000
6 | pen | tom | -1 | 3000000
7 | pen | tom | -1 | 3000000
8 | pen | tom | 13 | 3000000
9 | shoe | bob | 10 | 1000000
10 | hat | dan | -1 | 4000000
11 | hat | dan | -1 | 4000000
I am trying to find all item/sku groups that have a txId of -1 AND another valid txId (that is not -1). I don't care about line 4, since cup/jim is only 1 row of item/user. Line 5/9 also only have 1 row for the item/user group. Line 10/11 do not have a valid transaction id, so I don't want them either.
I want my results to be:
id | item | user | txId | date (ms)
-----------------------------------------------
1 | cup | bob | 10 | 1000000
2 | cup | bob | -1 | 1000000
3 | cup | bob | 10 | 1000000
6 | pen | tom | -1 | 3000000
7 | pen | tom | -1 | 3000000
8 | pen | tom | 13 | 3000000
Here is what I have tried
select groups.id, groups.item, groups.user, groups.txId, groups.date
from (
SELECT
*
FROM (
SELECT
*
, COUNT(*) OVER (PARTITION BY item, user) AS occurrences
FROM myTable tbl
) t
WHERE t.occ > 1
) groups
INNER JOIN
(
SELECT * FROM myTable WHERE txId = -1
) violators
ON violators.user = groups.user AND violators.item = groups.item
ORDER BY groups.user, groups.item DESC
I am getting some false positives and can't figure out why.

Answer 1

You can use exists:
select t.*
from t
where (t.txid > 0 and
exists (select 1
from t t2
where t2.item = t.item and t2.user = t.usr and
t2.txid < 0
)
) or
(t.txid < 0 and
exists (select 1
from t t2
where t2.item = t.item and t2.user = t.usr and
t2.txid > 0
)
);
You can also use window functions:
select t.*
from (select t.*,
min(txid) over (partition by item, user) as min_txid,
max(txid) over (partition by item, user) as max_txid
from t
) t
where min_txid = -1 and max_txid > 0;