It was a raw evening at the end of October, and I had come in from my rounds to find Sherlock Holmes stretched in his armchair among a litter of printed tables, a pencil behind one ear and the fire very nearly out. He did not look up.
“You have been following the World Championship, Watson. Your newspaper is folded open at the League of Legends results, and the crease is soft from rereading.”
I confessed it. Sixteen teams had begun the Swiss stage on the twenty-third of October. By the thirty-first, eight of them had gone home, and very little had gone as I had been led to expect.
“Led by whom? By this?” He tapped the sheaf on his knee. “It is the StatChecked model: a rating for every professional team in the world, built from more than three thousand games of this season, and a forecast for every match before it is played. In this affair it is the nearest thing we have to a witness who was present before the crime. Let us see where its testimony held, and where it did not.”
| Match | Model’s pick | Model gave | Crowd’s picks | Winner |
|---|---|---|---|---|
| MKOI v BLG, Oct 23, one game | BLG | 77% | 0 MKOI, 25 BLG | MKOI |
| TLAW v CFO, Oct 23, one game | TLAW | 80% | 18 TLAW, 7 CFO | CFO |
| DK v T1, Oct 24, one game | T1 | 51% | 1 DK, 25 T1 | DK |
| IG v HLE, Oct 25, one game | HLE | 70% | 0 IG, 19 HLE | IG |
| BLG v TES, Oct 29, best of three | BLG | 78% | 16 BLG, 6 TES | TES 2–1 |
| LOS v TLAW, Oct 30, best of three | TLAW | 78% | 4 LOS, 21 TLAW | LOS 2–1 |
| BLG v HLE, Oct 31, best of three | HLE | 50% | 1 BLG, 23 HLE | BLG 2–0 |
The first evening
“Begin on the opening day,” said Holmes. “Bilibili Gaming, the third-rated team in the world by these figures, met Movistar KOI in a single game. The model gave Bilibili seventy-seven per cent. Every one of the twenty-five people in our pick’em who chose a side chose Bilibili. Movistar KOI won.”
“A fluke,” I said.
“A single game, which is much the same thing. Mark the format, Watson. The first rounds of the Swiss are played as one game apiece, and one game is a poor witness: it remembers only the last thing it saw. The same evening Team Liquid, given eighty per cent, lost to CTBC Flying Oyster. The next day T1 lost to Dplus KIA, and our crowd had backed T1 by twenty-five picks to one. The model, more modest, had called that one a coin toss: fifty-one per cent. When the evidence is evenly balanced, the crowd tends to hear only one side of it.”
The team nobody picked
He drew out another sheet. “On the twenty-fifth, Invictus Gaming met Hanwha Life Esports. By the model’s ratings Hanwha were the second-best team in the world, and it gave them seventy per cent. Nineteen of our players made a pick. Not one of them chose Invictus, and Invictus won.”
“Remember that name, Watson; we shall meet it again. Of the sixteen teams in this stage the model rated Invictus twelfth and Top Esports eleventh. Between them they beat Hanwha Life, Anyone’s Legend, G2 and, over three games, Bilibili. Both went through to the quarterfinals. Meanwhile four of the eight teams the model rated highest went home: Hanwha Life, second; G2, sixth; Team Liquid, seventh; and Anyone’s Legend, eighth.”
“Then the model is worthless,” I said.
Holmes looked at me with something like pity. “It called twenty-one of the thirty-three Swiss matches correctly: sixty-four per cent. Our crowd, picking the same matches, got fifty-five per cent of its picks right. A witness who is right two times in three in a tournament this wild is a witness worth hearing. The error is not in the witness, Watson, but in expecting any witness to be right every time.”
The last day
The Swiss stage ends in the cruellest round, among the teams with two wins and two losses: win and you go on, lose and you go home. LØS, the Brazilian team the model rated fifteenth of sixteen, had already beaten Team Secret Whales and Team Liquid over three games apiece, and fell at the last to Lyon. Movistar KOI, who had begun the stage by beating Bilibili, went out to Karmine Corp.
“And the last match of all,” said Holmes, “is the one I would have you study. Bilibili against Hanwha Life. The model, for the only time in the whole stage, could not separate them: fifty per cent exactly. Our crowd had no such difficulty. Twenty-three of its twenty-four picks went to Hanwha Life. Bilibili won two games to none.”
“Why?” I asked.
“Because a crowd votes for a reputation, and a model weighs a record. Hanwha Life had the reputation; the model’s ratings for the two teams were a single point apart, 1759 to 1758. When the facts are level, Watson, the man who admits they are level is wiser than the man who has already decided.”
Who went through
Gen.G and Dplus KIA came through unbeaten at three wins and no losses. Invictus Gaming, T1 and Top Esports finished three and one; Bilibili Gaming, Karmine Corp and Lyon three and two. Karmine Corp had needed to win all three of their Play-In series simply to take the sixteenth place in the stage. Hanwha Life, LØS and Movistar KOI went out at two and three; G2, CTBC Flying Oyster and Team Liquid at one and three; Anyone’s Legend and Team Secret Whales without a win.
One more fact went into Holmes’s commonplace book: the shortest game of the tournament came on the very first evening, Hanwha Life over Top Esports in twenty-three minutes and fifty-one seconds. He would not say why it interested him. The model’s ratings and its picks for the rest of the tournament are on our Worlds helper.
A Sherlock Holmes pastiche; the characters are Arthur Conan Doyle’s. Every result, rating, probability and pick count in it is real, taken from our Worlds helper and pick’em on the date above.