The Final (TI8) Countdown

Ben Steenhuisen
Apr 25, 2018 · 4 min read

Four months from today The International 2018 Grand Finals will be played with millions of viewers tuning in around the world. It could be the first year we see a two-time winner of Dota’s de facto World Cup, or an underdog could come out of nowhere to snatch the Aegis. Eight teams will qualify directly for the event based on their performance this year in the Dota Pro Circuit (DPC), and it’s likely (but not guaranteed) that eight to ten more will qualify in various regional qualifiers. The window for those direct invites is closing, with only 5 more DPC events left this season ….

There’s ~2.2 * 10²⁰ ending top 4 permutations these final events could create, but obviously some of those outcomes are way more likely than others (and most outcomes don’t actually affect which teams will qualify for a direct spot). As a result, we can effectively model some of these remaining brackets using the team’s current Glicko ratings and Monte Carlo methods. This provides a way to reasonably compute expected (average) outcomes, but in a way that’s flexible enough to be updated live (every time a match is played it can impact the final outcome). This lets us measure some key metrics:

  • How likely is a team to earn enough points be qualify for TI?
  • What are their expected points and placement?
  • What’s the average cutoff point?

There are three assumptions made in this model:

  • A team doesn’t undo it’s post-roster-lock changes (i.e. OG doesn’t get Resolution back) and attempt to qualify with that roster (provided they didn’t kick the player from the in-game locked team).
  • None of the currently eligible top 10 teams make a roster change that’d make them ineligible for a direct spot.
  • VGJ. Storm win the Supermajor NA Qualifier, the final qualifier spot for main-season DPC. I could’ve added this qualifier as part of the model, or waited until tomorrow for the qualifier to end — but for now I’m going to be a bit lazy and assume they win the qualifier (and I’ll adjust it if it’s wrong). Technically 5 of the 6 teams could still win it — but it’s very likely VGJ. Storm win against isGG (who are already out of the running) and get at least 1 game off Immortals (who need to win all remaining matches to qualify). I’ll update the model tomorrow anyways.

So let’s look at the model output as of now.

10⁵ iteration Monte Carlo simulation of remaining DPC events

The average 8th place value is 2968 points — so as of now that’s the expected cutoff. It’s a bit lower than what I predicted at the start of the season (3.2k), but I also expected the top-performing teams to take bigger breaks than they have done (VP want that 3rd Mercedes!).

Some immediate observations:

  • VP / Liquid / Secret are basically locks. In none of the model iterations did any fail to qualify. This isn’t to say it’s impossible that one might fail to qualify (VP are absolute locks though), it would just require a monumentally rare and specific set of circumstances to occur — basically collusion in all the remaining events.
  • Mineski would be very unfortunate to not make it. After a win at DAC and some valuable points catapulting them into top 4, they’re actually only modeled to earn 1346 points more (and 400 less than Newbee), but with their recent form it’s possible they earn a few more points to absolutely secure them a spot.
  • Newbee are in a precarious position. They’ve placed consistently towards the top at every event but fell short at the last Major (9th-12th, DAC). They are currently 5th and projected to reach a relatively safe 4k mark, but if they don’t keep the points rolling in they could find themselves overtaken by fellow compatriots.
  • LGD and Vici need to keep their feet on the gas if they want to keep their future in their own hands. What’s most important for both of these teams is that the remaining points are all picked up by a single on-form team (let’s say EG) or some quasi-locks (VP, Liquid, Secret) but aren’t evenly spread among the middle-of-the-table teams.
  • EG vs the world, it’s close to 50:50 on if EG make it. They too are projected to double-up, but only are attending 3 events (2 majors + 1 minor). Whilst they are one of the favourites to win GESC, there’s effectively only 450 points up for grabs there (150 * 3). They’ll need as well as another top 4 finish if they want to displace VGJ. Thunder who are currently at 1935 (but with only the Supermajor left for them to earn points in). Fnatic and TNC have both put up some honourable performances but will need to go deep in playoffs.
  • Optic, NaVi, FTM & Vega all have qualified in at least 0.04% of the 10⁵ simulations, and basically for them this means performing well in one or more majors: winning a normal major is 2250 teampoints, winning the Supermajor is 3375 team points (which is on average enough to make top 8).
  • OG and VGJ. Storm could win some more points, but this is effectively removing points from the system (which teams like Newbee and Mineski would love to happen).

All in all, we’ll see the window for teams slowly shut as the remaining ~362 games of the DPC main-season happen. This is the chance for the teams on the precipice to make a difference, or fall short and have to do regional qualifiers. In either case it’s an exciting time of the year, so naturally I’ll keep the model updated as results start streaming in.


Professional Dota 2 statistics

Ben Steenhuisen

Written by

Dota 2 statsman and occasional caster | runs @datdota



Professional Dota 2 statistics