The database currently contains information about 439421 Ability Draft games, played between May 28, 2020 and Jul 11, 2020.
Calculating accurate win rates for all 100,000+ pairs of abilities requires a lot of data. With that many pairs, it is likely there are some that have gotten very "lucky" or "unlucky". So it is best not to trust those rates completely, at least until more games are played.
Especially when looking at pairs with the highest or lowest win rates, pairs with few games played are more likely to come up, not because they actually have a very high or low rate, but because small sample size means higher variance.
Somewhat arbitrarily. It is the difference between the combined win rate and the geometric mean of the individual win rates. It is intended to measure the strength of interactions, independent of their components. For example, Sticky Napalm and Ion Shell both have very average win rates individually, but they have high synergy, leading to a higher winrate when picked together.
No, unfortunately Valve does not make this information available in their API.
Yes, click the column header.
Yes, hover over any win rate (or tap on mobile) to see wins, losses, and 95% confidence interval.
I use the Jeffreys Interval. The win rate is modeled as ~ Beta(1/2 + wins, 1/2 + losses), and I give the locations at which the CDF of that distribution is 2.5% and at which it's 97.5%.
My name is Joe Kelley. I've been programming for fun since 2006 and professionally since 2010. I've been playing DotA for fun since 2004, and still haven't gone professional. You can find links to all of the rest of my online presence on my website: www.joefkelley.com. Feel free to email me at firstname.lastname@example.org.