How does the autobuild for Sealed work? Does it just take the strongest 2 colors and then fill the deck with the top 23 cards in those colors? Does it do anything to try to control for curve or creature count?
Your program is good at picking colors, but makes questionable choices about deck construction. I think it's something that could be improved upon with some simple programming logic.
Here's a good example. The pool and deck the Autobuild gave me...
Only 12 creatures and only 1 2-drop. That's low for any deck, but even worse when you have 2 creatures that require sacrifices and 3 equipment. Also Double Act of Treason? Death Baron with 0 zombies? Gearsmith Guardian with 0 blue creatures?
I don't expect it to decide it's worth splashing green for the Elder Dragon and Goreclaw, or how that relates to the Dragon synergy with Spit Fire, the 4-power count, and the presence of Rupture Spire. Limiting it to 2 color decks is reasonable.
But there are some other things it could do better, like the number of creatures. 12 is too few in most cases (some archetypes in some formats are an exception). As a general rule, you could force a minimum number of creatures (e.g. 15) and have it pick creatures until it satisfies that number.
You could also try to control for creature curve. 1 2-drop is too few for most formats. The pool has more on-color 2-drops. They're filler, but filler is usually better than not having bodies at the right time. You could have it pick the best X 2-drops to ensure they're enough. You could also put caps on the number of 5s and 6+s to ensure the curve doesn't get too high.
You could control for the number of equipment and vehicles. 3 copies of Marauder's Axe is excessive, especially with 12 creatures and essentially no bears to equip. You could force a maximum of 2 equipment. That won't always be the right choice, but as a general rule of thumb I think it would lead to smoother auto-builds. In the rare exceptions where you want more equipment, users can tweak the deck.
I also think your algorithm has either overrated Act of Treason or needs to otherwise cap the number of spells that are card disadvantage. This pool has plenty of other options for the spell slots. Maybe if you forced the minimum 15 creatures Act of Treason naturally wouldn't make the cut and you'd be OK.
Here's what I built. It may not be the best build, especially since we don't know the format yet, but it seems better.
Accounting for all synergies is hard. But your algorithm could account for some basic synergies (especially in core set), like color-matters, artifacts matter, and tribal. You could build this into both the draftsim and the autobuild feature (if it builds the deck by iteratively selecting cards down the list).
This would involve each card having a certain number of attributes, not just the color and rating. I don't know how complex it would be to attach those features (e.g. color, type, tribe, equipment, mana cost), but I think that would help both the draftsim and the autobuilds by allowing the ratings to be modified by past choices.
First of all, thanks a ton for the detailed response. I definitely am actively looking to improve the drafting algorithm and this is good feedback. You can get some background for some of your questions on the FAQ.
Yes, the autobuild (and draft algo, for that matter because they have the same basis) assumes for simplicity's sake that you're building a 2 color deck. There are so many permuatations of different decks you could have that are 2+ splash, 3 color; 4/5 color with a ton of fixing that it hasn't been a priority to focus on this yet. Cards aren't labeled things like "fixing" for it to take that into account so that is something that would need to be assigned for every single set.
Same goes for synergy - that differs vastly set to set. Sometimes there are obvious ones (eg tribal) and sometimes not so obvious (eg cards that are "good in a control deck"). And a card can have one of these attributes or many. There's no way that could be evaluated by me within half a day of the set being spoiled and made live for everyone to use
However, that said I am looking into using machine learning to try and suss out synergies, so stay tuned on that front. Mana curve is a bit less open ended so that's definitely in the plans to include.
Its a great tool. I think the draft simulator is good. Not too worried about the Autobuild but I recognize it takes the top 23 by pick value. Its a quick start to edit a final list. Does the AI update its card pick-list values as other people draft the set or does it keep the static values you input at the beginning? It would be great to add a few basic chart tools for the built deck like a mana curve and cards totals by type (creature, spell, artifact, lands). However, I can cut and paste the export list easy to a few sites that provide this (just usually too much effort for test drafts).
I update the ratings periodically as my understanding of the format changes. It's currently recording picks and I'm working with some people on trying to do stuff with the data, but it's a big project.