It’s that time of the year where everyone digs out their Ouija boards, Crystal Balls 90-sided dices, tea leaves and Tarot cards. I’m mostly partial to throwing darts at a dart-board. Yeah, you guessed it - it’s Compendium Prediction time.
In the build up to both TI7 and TI8 there were frequent balance patches, with Icefrog acting as a greenskeeper attempting to create ideal conditions for the upcoming event. One of the problem this brings is that it means that for many predictions you need to look at an event smaller (and more recent) slice of data. Winstrike, Serenity, IG, and OG for example haven’t played pro LAN matches in 7.16+. As a compromise, some predictions require looking moderately far back (general playstyle, overarching distributions) for the teams involved, some require looking very recently at trends across all teams (including teams not present at TI8), and some are just crapshoots anyways so it’s almost random.
We’ll focus on the last section because it’s the easiest one to use stats to get an advantage (they’re all hard, but this is the easier) category to predict. The other sections we’ll mostly skim through since there’s often multiple predictions affected by one single outlier of a game (very long game with loads of kills, deaths and assists).
Total Games Played At Main Event
The Supermajor (which actually featured a superior format to The International) saw 21 bo3s and a bo5 final. The bo3s averaged 2.43 games per series, compared to TI7’s 2.29 games per bo3 series. If we pin in the finals as a 4-game series, this puts a bound of 46.9 to 49.3 overall matches. If you think the games will be slightly closer, or the finals will go the full five games then go for 50-54.
Total Heroes Picked & Banned
Given that we predict ~47 main event games and there are ~144 games (could possibly be tiebreakers) in the group stage, this takes the total games to ~191. If we look at a similar number of the most recent matches involving at least one of the TI8 teams, we see 113 unique heroes picked and 102 heroes banned. Since TI only involves the exact 18 teams, and they each come in with a limited number of strategies - these values are probably slightly over-exaggerated (the bans more so than the picks, since teams sometimes ban off-meta signature heroes of their opposition).
Most Combined Total Kills In A Game
For the next few predictions I’m just going to defer to the most recent ~190ish LAN games. I’m filtering on LAN games here because it generally means balanced matchups which is more reasonable than often imbalanced online qualifiers (even the TI qualifiers).
Longest & Shortest Games
Since TI7 we’ve only seen 6 matches go beyond the 90 minute mark, and only one of those was on LAN (Fnatic vs Vici at ESL One Genting). We’ve also seen no games sub-10 minutes. Although The International is a land of mystery with cosmically rare events occuring - I think it’s unlikely we see these get broken. If we look at the last ~190ish LAN matches we see only a few occurrences come close to the extrema I described.
Looking back at TI7 paints a similar picture for shortest game, but does feature two slugfests that exceeded the 90 minute mark: Liquid/VP and Empire/IG.V. This is probably just because TI is so important to everyone - teams are not willing to resign and will fight until the bitter end.
Most Kills In A Game
I’m just going to ignore options below 16 - last year we had nearly 50 performances with more kills at TI. Let’s look back to our ~190ish game recent LAN sample.
I’m somewhat tempted by 26+, but think I’ll play it a bit safer and go with 23-25.
Most Deaths By A Hero In A Game
Our trusty sample suggests 18-20 is the cutoff, last year Notail’s Pugna held the record for most deaths with 20 (in a loss vs VP). I’m going to play it a bit conservatively however, and take 18-20, but 15-17 might be safer.
Most Assists By A Hero In A Game
Three players last year got 36+ assists: Solo, Q and super. Specific heroes like Bounty Hunter, Ancient Apparition, Spectre and Zeus are way more likely to get many assists - and at least in the case of Zeus and Spectre we should see them at least 12-15 times each. We’re going big here.
Highest GPM In A Game
Recently, Anti-Mage and Alchemist have been very rarely picked. These are the most notorious GPM cheese-gods, and the ones most likely to feature in any record-books. Beyond them, Morphling is probably the most likely candidate (and he has been picked relatively frequently in recent LAN matches). That said, only one Morphling ever has > 900 GPM in a LAN match (Sccc, vs OG at TI7) although many have come close. To me, this suggests 800-899 GPM is certainly the lowest I’ll go. That said, all it’ll take is one decent match from an AM or Alchemist to bust this, so I’m mostly tempted to go 900-999.
Last year multiple Alchemist’s were picked, but only Ana’s Alchemist (vs Execration) got over 1k GPM. On LAN in the last few months, only 3 heroes have broken the 1k mark.
Most Picked Hero
So many options - mostly likely versatile supports. I’m going with Windranger because I think a few teams will practice with her and have her in their toolbox. Will also be a common opener, and one that won’t get too many bans. I think Crystal Maiden and Winter Wyvern are definitely considerations, but they also should get a fair amount of bans.
Most Banned Hero
You want to pick some annoying and dodgy last-pick hero that’s only going to be picked like 8 times and only by good teams. Initial considerations include Ember Spirit, Tinker, Huskar, Slark, Anti-Mage, Meepo, Templar Assassin, Broodmother or Queen of Pain. I think that Ember and Brood will be picked by teams in suboptimal positions (Optic sometimes go Brood when it’s an okay pick, but not a great pick). Meepo will mostly be banned by teams against the good Meepo teams. Queen of Pain might also be a 2nd-phase staple for some teams, so the winrate could suffer as a result.
Highest Kill Average
Some kill-focused heroes will be good, including PA, Ember Spirit, Tinker, Clinkz, Meepo, Ursa, Morphling, Storm Spirit. The biggest concern for PA, Clinkz and possibly even Tinker & Ursa is they might not be picked enough. Huskar games might end very quickly though.
Highest Assist Average
A few key heroes are head and shoulders above others in this category: Global Heroes (Zeus, Spectre, Ancient Apparition), Constant Fighters (Spirit Breaker, Bounty Hunter) and Teamfight Gods (Undying, Disruptor, Dark Seer).
I think Zeus will be picked a huge amount of times, and consistently get loads of assists. As a core he might also get a lot more kills than assists - but due to his aura will continue to score highly in this category. Spirit Breaker, Bounty and Dark Seer are unlikely to feature enough.
Lowest Death Average
Difficult to catch annoying heroes.
Highest Last Hit Average
Highest Highest XPM
Will Meepo be picked 5 times? If you think so pick Meepo. Otherwise you’re looking for cores that constantly are looking for kills (more efficient than timing-based hypercarries, or ones that rely on stacks).
Hero with Most Kills In A Game
You need a hero that will get out of control, but the enemy will believe they can fight on. Multiple good teams at TI8 have very proficient TA players, so I’m going to pick.
Hero with Most Last Hits In A Game
Slightly different from average last hits - Anti-Mage aims to win the game once he’s 6-slotted; whilst Medusa aims to turtle even longer.
Coming into the event it seems that VP, LGD and Liquid are very close in skill at the top. Most people focus on the VP/Liquid rivalry, but LGD have shown to hold their own against both - and more recently pushed VP all the way to the 5th and final game at the Supermajor.
It’s hardly surprising that I think these 3 teams will dominate the positive aspects of the team predictions, but it might be worth looking in which group Pain and Invictus Gaming end up because I view them as the weakest and if a good team is in a group with both of them it might mean two easier series.
To dig a bit more into the data, check out here.
Like the team predictions, a lot of the categories just seemed like a Miracle/Maybe/RAMZES/No[o]ne showdown.
S4 & Yapzor are perhaps the most unusual picks - I think that Secret will have relatively long matches (will hold their own against the top 3, but also will play it patiently against teams worse than them). This means Yapzor will be playing more Dota than GH (who averages the most assists per minute over the sample I looked at). EG have shown they’re willing to make interesting laning setups to favour their cores, and this might call for Gustav to play unconventional heroes. MidOne and Resolution have also played a huge number of unique heroes recently. I think that EG are a team that will be undervalued coming into the event - and if they coming top 6 or top 8 its means they play enough games to still qualify for this category.
To dig a bit more into the data, check out here. I mostly did normalization for the average skill of the opponents that the players played against.