What I want this morning, after seeing Stand Atlantic and ONE OK ROCK in Boston last night, is a sampler playlist of some of their older songs that I haven't played before.  

I can have this! Into Curio we go querying.  

First of all, I need to select these two artists. This could be as simple as  


or as compact as  


but these both assume I'm already following the artists, and I am, but I might want samplers of artists I'm not.  


In this version .matching artists looks up each name with a call to the Spotify /search API, .matches goes to the found artists, and #popularity/name.1 sorts them by popularity, groups them by name, and gets the first (most popular) artist out of each group, to avoid any spurious impostors.  

 

Now that we have the artists, we need their "older songs". E.g.:  


Here .artist catalogs goes back to the /artists API to get each artist's albums, .catalog goes to those albums, :release_date<2024 filters to just the ones released before 2024, and .tracks:-(:~live,~remix,~acoustic) gets all of those albums' tracks and drops the ones with "live", "remix" or "acoustic" in their titles. Which is messy filtering, since those words could be part of actual song titles, but the Spotify API doesn't give us any structured metadata to use to identify alternate versions, so we do what we can. We're going to be sampling a subset of songs, anyway, so if we miss a song we might not really have intended to exclude, it's fine.  

Administering "songs that I haven't played before" is a little tricker. I have my listening history, but many real-world songs exist as multiple technically-different tracks on different releases, and if I've already played a song as a single, I want to exclude the album copy of it, too. Here's a bit of DACTAL spellcasting to accomplish this:  


We toss the track pool and my listening history together, we merge (//) them by uri (in the listening history this field is called "spotify_track_uri", but on a track it's just "uri"), we group these merged track+listening items by main artist and normalized title, we drop any such group in which at least one of the tracks has a listening timestamp, and then take one representative track from each remaining unplayed group.  

Now we have a track pool. There are various ways to sample from it, but what I want today is a main-act/opening-act balance of 2 ONE OK ROCK songs to every Stand Atlantic song. And shuffled!  

We can shuffle a list in DACTAL by sorting it with a random-number key:  


In this case we need to shuffle twice: once before picking a set of random songs for each artist, and then again afterwards to randomize the combined playlist. Like this:  


Shuffle, group by main artist, get the first 20 tracks from the ONE OK ROCK group and the first 10 tracks from the Stand Atlantic group, and then shuffle those 30 together.  

All combined:  

 

 

 

 

Did I get what I wanted? Yap. And now I don't just have one OK rock playlist I wanted today, we have a sampler-making machine.  
 

PS: Sorting by a random number is fine, and allows the random sort to be combined with other sort-keys or rank-numbering. But when all we need is shuffling, ....shuffle is more efficient than #(....random).