BudgetSWU: my latest pile of code

I made a little web app called BudgetSWU. It recommends decklists for a trading card game called Star Wars: Unlimited. I started playing SWU almost a year and a half ago because I had a dream I was in a mall opening a pack of Yugioh cards--another trading card game that I played as a kid--and it gave me enough nostalgia to do some googlin' and discover SWU.

As an adult I now have lots more money to buy cards than I did as a kid (yay) but much less time to play with them (boo). So I don't spend much time building my own decks; I just find decklists on the internet and buy the cards for those.

BudgetSWU is now one of those places on the internet. It gets decklists from melee.gg which is used in tournaments (large and small), including each decklist's (player's) placement in the tournament. It also gets pricing data for all the cards in those decklists from tcgcsv.com, which in turn aggregates pricing data from TCGPlayer, the ebay of trading cards (it is owned by ebay, in fact).

As a final bit of data, I compute "popularity" by taking each deck's "archetype" (the leader + base, for those in the know--except that I'm actually using just base color) and seeing how many decklists in the entire corpus have that same archetype. It's cooler to play decks that not everyone else is already playing, know what I mean?

Then BudgetSWU filters all the decklists down to a handful of recommendations by computing a Pareto Front. The idea here is that we have multiple factors with which we can rank decks (cost, performance, popularity/nicheness) and we don't necessarily know which factor is most important (if deck A costs 50% more than deck B, how much better/niche does deck A have to be for it to be a better recommendation than deck B?). But we do know that if deck A beats deck B on all three of those factors, then deck A is probably the better recommendation. So we filter out all the decks that have a clearly better alternative, and then we tell the user "here's what's left; you decide." As of writing, that ends up being less than 20 decks, which is a very manageable number for a user to scroll through.

The next cool bit is that if you make an account, you can enter what cards you already own, and then the recommendations will be computed based on the cost of only the cards you don't have yet. Entering in all your cards would be a pain since players old enough to have jobs usually have thousands (of cards, not jobs). So BudgetSWU sorts all the cards by cost, most expensive first. (Right now the most expensive cards go for about $30-ish a piece). You can start from the top and enter quantities until you got to cards that cost, say, a dollar each or whatever you feel like, which doesn't take long.

Finally: whenever a new decklist makes it into your personalized recommendations (tournaments happen every week), you get a notification email. Booyah.

I generated the code in Clojure. It uses my web framework Biff. I've recently been working on some LLM skills that provide guidance for writing Biff code that doesn't suck. And I have to say, I'm fairly pleased with the results. The code still isn't anything I would print out and frame on my wall, but the overall structure is generally inline with how you're supposed to write Biff apps. If this was an app I planned to develop further (it's not), I would be happy with the state the code is in as a starting point.

If you're into specifications, you can read the one for BudgetSWU over here. i.e. that page is the thing I actually typed, mostly.