On this page: https://www.postgresql.org/docs/15/sql-merge.html the first and second examples seems to be contrasted (by "this would be exactly equivalent to the following statement"), however the difference does not seem to related to the stated reason ("the MATCHED result does not change"). It seems like the difference should involve the order of WHEN clauses? Of course, it might be that I don't understand the point, in which case maybe the point could be stated more clearly?
Yeah, that is a pretty poor pair of examples. Given that a given customer can reasonably be assumed to have more than one recent transaction the MERGE has a good chance of failing.
The only difference between the two is the second one uses an explicit subquery as the source while the first simply names a table. If the subquery had a GROUP BY customer_id that would be a good change explaining that the second query is different because it is resilient in the face of duplicate customer recent transactions.
While here...source_alias (...completely hides...the fact that a query was issued). What? Probably it should read (not verified) that it is actually required when the source is a query (maybe tweaking the syntax to match).
