General Cube draft with bots

[japahn]

Cube Cobra's draft bots are a bit too consistent, basically following a pick order based on their Elo ranking. That means I almost never see some cards, and always see others. For some reason Siege-Gang Commander has a lowish rank, and I pick it in every red deck, for example.

Cube Tutors bots are good (both castle and regular), but they don't have Ikoria yet and it's unclear for how long the site will stay up.

Tappedout's bots are horrible at sticking to colors and there are basically no color signals.

What site do you use for trying out drafts with bots? What biases do they have that you need to take into account?

 

[TrainmasterGT]

Cube Cobra's draft bots are a bit too consistent, basically following a pick order based on their Elo ranking. That means I almost never see some cards, and always see others. For some reason Siege-Gang Commander has a lowish rank, and I pick it in every red deck, for example.

Cube Tutors bots are good (both castle and regular), but they don't have Ikoria yet and it's unclear for how long the site will stay up.

Tappedout's bots are horrible at sticking to colors and there are basically no color signals.

What site do you use for trying out drafts with bots? What biases do they have that you need to take into account?

I think cubecobra bots are pretty good since they draft somewhat similarly to a skilled human drafter. They're not perfect, but they usually come up with something fairly playable. I don't expect any bot to be able to draft a good synergy deck outside of picking good cards that fit into a particular dominant strategy.

The issue with cubetutor bots is that since you train them, your biases will be reflected by the bots. If you severely overrate something, the bots will as well, warping the data you would receive in a normal draft. A CC bot picking every 1-mana mana dork it sees is more likely to happen in an actual draft than a CT bot leaving one for 15th pick.

Tappedout is not a good cube management tool in any case. I don't think it would be worth using even if the bots had pro-tour tier skills.

For some reason Siege-Gang Commander has a lowish rank

I think it's because Siege-Gang isn't very good in high power environments, which CC bots are really designed to work with. I'm not going to sit here and call Siege-Gang Commander a bad magic card, frankly it isn't. However, when the power ceiling of a given format is Thundermaw Hellkite or some similar efficient threat, a 2/2 for 5 which makes some tokens and requires an additional mana payment to do anything else is lackluster. It's not an early pick in those environments.

 

[japahn]

Cube cobras feel like 7 robots drafting religiously by Frank Karsten's pick order before he actually played the set.l

True about Cube tutor's regular bots, but castle AI bots aren't trainable and you won't bias them. I like to mix them as it half biased one way and half biased another way actually adds up no not letting either bias transpire.

Regarding tappedout or any other tool not for cube management, the idea would be to export the list into it to test it.

 

[Onderzeeboot]

Castle AI bots are awesome, but by this point CC is so much ahead of CT in terms of support that I'm just drafting on CC. There's an issue on their backlog to make better bots, so it's a matter of time before CC also introduces better bots, I think.

 

[Erik Twice]

Cube Cobra's bots are indeed weak. It's very noticiable with fixing, you can end with way more duals than you are supposed to. I indeed think Cube Tutor's bots are much smarter once trained and lead to more normal drafting experiences.

Right now, I use both, though I like the interface of CC better so I do my changes there now and then port them over.

 

[shamizy]

I think it's only a matter of time before CC catches up as far as draft AI is concerned. Early on my cube was way too easy to draft, but I think the bots are start to catch on as far as fixing is concerned as I can't really wheel a dual land all that often anymore. I'd give it a few months before they can get around to overhauling that, it's mostly been aesthetic QOL stuff thus far on their docket.

 

[Rowan_CB]

I only just brought my cube to CC, due to CT not having Ikoria, and haven't done a draft there yet. That said, the fact that Ryan Saxe seems to be working with them (they're applying his recommender algorithm) is a great sign. His recommender (like the bot he developed to draft retail limited sets) is really good at learning synergy; if he ever develops a draft bot for CC, it will be amazing.