If that was the case, the graph would never have gotten to the heights it did.
What happened is that as the corpus of useful info increased, the need to pose new Qs decreased. AI much accelerated that decline by making available an 'oracle' trained on that corpus.
The orig wants a mono output with one of the original channels as signal source. This involves downmixing i.e. rematrixing the audio.
The dupe want to just mute one of the channels, not repan it. One can't apply map_channel to do what the dupe wants.
One can use a couple of methods to achieve the dupe, including pan. But the syntax of pan needed for the dupe case is not the same as the orig, or deducible from it. They need to consult the docs (fortuitously, the dupe case is an illustrated example) or get a direct answer. The 'technique' shown in the orig is not intuitively adaptable to the dupe - one needs to know about the implicit muting that pan applies, which is not documented or evident in the orig answer. So it's not a duplicate of the source Q.
You are positing that only questions with cosmetic or extraneous differences are marked as duplicates.
That's not the case. As a maintainer of a popular project who has engaged with thousands of Qs on SO related to that project, I've seen many Qs marked as duplicate where the actual answer would be different in a non-trivial manner. When I look at who all moderated on those Qs, they are usually users who haven't contributed to that topic at SO.