Learning from the TI finalists: the smoke, the carry, and the other side of the map

The TI15 general smoke study answers four questions about the map photo: how many lanes are pushed, where the group goes, where the enemy is isolated, and if there is an objective in the window. This extra section asks a fifth question, which the general study didn't ask (and where was their carry?) and uses the answer to look at the matches of the three teams that reached the end of the tournament: TEAM VISION, Team Spirit , and Team Yandex.

The question is not academic. VISION farms with the Satanic scattered around the map, and the intuition of everyone who played against it was the same: find the Satanic and kill it. Measuring move by move, this intuition is the explanation of much of what went wrong.

One thing the reader needs to know about the base: who faced VISION in the tournament was Team Resilience, Team Falcons, BoomBoys, Team Spirit and (since the upper final, in the early hours of 08/22) Team Yandex. There are 19 maps, and almost all of them with the gold score against: the average advantage of those who pressed smoke against it was more than seven thousand gold negative. This makes every raw number of it optimistic for VISION and pessimistic for the others, and that's why every important cut appears here also discounting the gold advantage: comparison within equal score ranges, the same control as the general guide.

VISION is the toughest team to smoke in the tournament, and that's measured

Looking at each team as a defender (which happens to them when the opponent presses smoke), VISION is not only the best on the list. It is an anomaly.

Read her line slowly. Of the 111 tight smokes against VISION, 22% converted and 49% exploded: the opponent lost the exchange almost half the times he tried. Discounting the attacker's gold advantage, it still concedes 14.5 percentage points less conversion than the average defender in the tournament. It's not the score: it's defense.

And there's a second column that tells the story: VISION's pos1 died in 10% of the windows. On the other teams on the list, that number goes from 22% to 33%. The most hunted carry in the tournament is the one that dies the least.

The third number is what ties the two together: Satanic is the carry that looks the most alone. The average distance from his closest live teammate, the second the opponent presses the smoke, is 3,383 units (the highest among teams with a large sample) and in 48% of the moves it exceeds 3,000. He seems the easiest target on the list and is the hardest to kill.

Also notice the two lines of the other finalists, because they are the mirror: Spirit and Yandex suffer more smoke than average (+4.8 and +7.3 conversion points conceded, discounting the score). Among the three teams that decide the title, the difference in defense is greater than the difference in attack.

The new axis: how far from the carry the group has come

For each of the 954 smokes with the enemy pos1 alive, you can measure three things in the replay: where he was in the second of the squeeze, where the smoked group was 45 seconds later, and the distance between the two.

At first glance, the result is the opposite of the study's intuition: reaching the carry pays. Those who finished the smoke less than 2,000 from him converted 59% and exploded 25%; those who finished more than 7,000, 36% and 32%. But the short distance goes hand in hand with winning, and this is where separating the two things changes the conclusion.

The cut that separates it is the side of the map. Distance tells you if the group got to it; side says where half of the map it went hunting. The two things take off, and the middle way is the worst place in the study:

The reading is short: either you get to him, or you go to the other side of the map. What doesn't work is prowling around his side without getting there, and that's exactly what a team does when they "have their eye on the carry". Discounting the gold advantage, prowling around his side costs 7.4 percentage points more backfire and 0.33 net kill compared to going to the opposite side. Actually getting to him is worth +6.2 conversion points.

The explanation is geometric and is the same as the offlane error in the general guide. A smoke that goes to the half of the map where the carry is, but away from it, arrives at a place where: the carry sees its team on the minimap, the enemy rotation is short because everyone is on that side, and your escape lane is as long as possible. You paid the price of crossing without receiving the prize of catching someone.

The bait: the carry that seems alone and is not

The most useful number of this session isn't about distance: it's about what happens in the 45 seconds after the squeeze. Counting how many teammates were within 2,500 of the carry at the moment of the smoke and counting again at the end of it:

When the enemy carry is more alone during the smoke, it converts 51% and explodes 28%. When two or more teammates get close to it, it converts 34% and explodes 44%. And filtering only the moves in which the group actually went after him:

Going after the carry without reinforcement: 56% conversion. Going after the carry with two teammates appearing: 34%. It's the same decision with two opposite outcomes, and what changes isn't the carry: it's the one behind it.

Discounting the gold advantage, the pair of lines looks like this: getting to the carry with it isolated at 3,000 or more is worth +18.0 conversion points, −9.1 backfire, and +0.57 net kill. Getting to it accompanied is worth +3.2 conversion and +2.1 backfire, that is, practically nothing, with the backfire going up.

The clearest case of the tournament involves precisely Yandex, and on the bright side: Nigma Galaxy, 29:26 of game 2 against her. The Yandex's carry appears 7,276 units away from the nearest living teammate: the freest target a replay can show. The Nigma presses with a pushed lane (the offlane), returns to his own safelane behind him, and in the next 45 seconds three of his teammates come close. The window closes at 1 to 6 and −5,395 gold.

Nigma Galaxy × Team Yandex

Game 2 · 29:26 · match 8957749452

4×

−45s

State of the lanes

safelaneNeutral −829

midRetreated −1.973

offlanepushed +4.389

The decision

pressed on

Our mid

Arrived in

Our safelane

Bottom

2 of 5 alive

Isolated target

Their safelane

Isolation

7.276 un.

Result in 2 min

−5Liquid slaughter

−5.395Gold

Backfire

What happened

• +11sMirana (our)
• +28sHoodwink (our)
• +35sLargo (our)
• +38sEarth Spirit
• +40sClinkz (our)
• +44sVoid Spirit (our)
• +92sMirana (our)

The 2 min account closes in +120s

Who's who

Nigma Galaxy
ClinkzHoodwinkLargoMiranaVoid SpiritTeam Yandex
KezEarth SpiritDoomWinter WyvernRubick

Notice what Yandex is doing in this play: it lets its own carry look loose and keeps the team one rotation away. It is, by another name, what VISION does the entire game.

And this is where the defenders' table comes back with an uncomfortable piece of information for the two teams that can still face it: Spirit's average carry boost is 0.02 and Yandex 's is 0.11; VISION's is 0.34. When someone smoka in the direction of Spirit's carry or Yandex, help almost never appears. When someone smoka in the direction of Satanic, it appears.

Where Satanic farms

The question came with a hypothesis: if his farming lane is known, you can set up the smoke on it. The replay answers where his gold comes in: each drop of creep gold has one second, and the trail tells you where he was at that second.

The clear outline on each map surrounds the cells that concentrate half of his gold: until 20 minutes there are 25 cells, the safelane itself and the jungle on her side; after 20, there are 60, spread across the same side of the map. His lane doesn't change, it just gets fatter.

And the result is the opposite of fame. After ten minutes, Satanic takes 72% to 78% of his gold on his own field, and in the late game he is more homegrown than the average tournament carry (23% in the opponent's field after 45 minutes, against 29% of the other pos1). The two regions he uses the most are his own safelane (27% of the farm) and the jungle next to it (17%). He is not a carry that is always invading: he is a carry that occupies the most defensible corner of the map and forces you to go get him there.

This fits exactly with the previous two numbers. He seems isolated because he is far from the team, but he is far from the team inside, with his tower behind, his jungle around him and the team at a short rotation. It's the definition of good bait.

And the practical corollary is what the general study already said in another way: of the 111 moves against VISION, 51% ended up in the field of the one who pressed. Half of the smokes against it were not even an invasion: it was a team crossing its own map looking for someone.

Does the smoke delay his farm? Yes, but not the smoke that wins

The natural question after the farm map is if you can tie the two things together: use the smoke to deny his gold. The replay answers, counting only the creep gold that the opposing carry took in the 120 seconds before the squeeze and the 120 after.

In the entire tournament, a smoke costs an average of 189 farm gold to the opponent's carry. But the distribution of this cost is the opposite of the conversion table: getting to him takes away 352; going to the opposite side takes away 103. And when he dies, he takes away 582: which is the only way to really stop a carry.

Against VISION, the mismatch is explicit:

Going to the opposite side to Satanic didn't delay his farm at all (+10 gold, that is, he farmed the same or better). Circling his side took away 261. But it's the opposite side that converts, and prowling his side that explodes. In other words: the two things a coach wants (kill his economy and win the fight) are not in the same fate, and you have to choose.

The choice is arithmetic. A smoke that takes 261 of his farm and costs two of his heroes lost the exchange by a huge margin: two heroes in an TI late game are worth a few thousand gold in time and objective. A smoke that doesn't take any farm, but opens the tower and the Roshan on the opposite side, won more than 261, even with Satanic farming in peace during the play. The account is from the map, not the carry.

And there is a number that closes the argument on the other hand:

Satanic is the carry that loses the least farm when his team smokes: −121, against −262 for Spirit's carry and −253 for Yandex's. He is also the one who farms the most before the squeeze (1,380 creep in two minutes). Whoever tries to beat VISION by delaying Satanic is choosing the toughest target of the tournament for the hardest play to hit. The path left in the dice is the other: take the map faster than it turns farm into an item.

What actually happened: the 111 smokes against VISION

Drawing each one with the four dots that decide the move: where the smoke came from, where the group arrived 45 seconds later, where Satanic was in the second of the squeeze (the full dot) and where he was when the smoke ended (the ring, connected to the first by a dash).

His pair of marks is the new part, and it changes the reading. The median of what the Satanic walks in those 45 seconds is 3,499 units: almost the width of an entire lane. The squeeze photo, which is the information with which the team decides, ages in the middle of the play itself.

And the size of this movement separates the two ends: in the smokes that exploded, he walked 4,125 (median); in the ones that converted, 2,704. In other words, the smoke that goes wrong is often the smoke that chases a position that no longer exists: the group crosses the map behind where he was, and he has already rotated to the next jungle.

The two panels also don't have the same geography. The ones they converted point outwards, to the VISION field and to the river. The ones that exploded have a whole block of short arrows inside the field itself, and this is what the table by destination confirms:

Finishing in their own lane against VISION: 6% conversion and 60% backfire, in 35 moves. It's the worst region in the overall study appearing here multiplied, because against VISION the price of being out of position is higher.

And the distance to the Satanic, in her cut:

Here the study's intuition is confirmed and gains a number. Against VISION, getting close to Satanic is the worst decision on the list: 14.6 percentage points less conversion and 20.8 more backfire, discounting the advantage. The n is small (11 moves against 95) and the direction is the same in all neighboring cuts: at 2,000 to 4,000 from him, 18% conversion and 55% backfire; going to the side of his map costs 9.1 conversion points.

The other side of the same table is revenue:

Smoke with objective in the window, against VISION: 63% conversion and 17% backfire. Without objective: 10% and 57%. Discounting the gold advantage, the objective is worth +48.4 conversion points and −41.2 backfire: the biggest effect of any cut in this guide, and more than double what the same cut is worth in the entire tournament. Against a team that doesn't make a mistake, smoke has to exist to take something from the map, not to look for a hero.

And when Satanic was inside the field of the smokou (the situation that seems like a gift) the result was 15% conversion and 73% backfire in 26 shots. He was there because his team was behind.

It's worth reading the first line carefully: in the eleven windows where Satanic died, the account was 36% conversion against 55% backfire. Killing Satanic, against VISION, was no guarantee of anything, because the price of getting to him was almost always another hero killed on this side.

Who presses with the map stopped, and who does it by winning

Rule 1 of the general study is about pushed lane, with the 1,500 unit ruler inside the opponent's field. Lowering the bar to "the wave has passed the river, at least a little" (with a floor of 400 units, so as not to count the wave stopped in the river as if it had passed) you can ask something else: who presses the smoke with the map completely stopped, and in what situation?

The step in the second table is the same as in the general study, and it is clean: no wave on their side converts 32% and explodes 51%; one wave, 37% and 36%; two, 48% and 24%; the three, 65% and 16%.

The table per team has two readings. The first is the frequency: VISION is the team that least presses with the map stopped (17%, tied with Iron Wing and Resilience), and Vici Gaming, Falcons and LGD do so in more than 30% of smokes. Spirit 21%, Yandex 19%: the two better than the tournament average, which makes this *not* their weak point.

The second reading is the scoreboard column, and it's the reason the table exists: all teams do it by losing, except VISION. Her average gold advantage in these smokes is positive (+692), against −5,000 to −11,500 for everyone else. That is: when the map is still, the other nine smok because they're behind and need a miracle; she smoka because she chose. And yet it's her weakest cut (30% conversion vs. 53% on the others).

Two caveats. The first: most of the raw difference is the score, not the wave. Discounting the gold advantage, pressing without any wave passing the river is worth −0.3 conversion points (i.e., nothing), but still +7.3 backfire points. The standing wave does not prevent the smoke from working; it increases the chance of going very wrong, because no one on the other side needs to be anywhere. The second: the 400 unit floor is conservative and the result does not depend on it: from 0 to 1,500 units the cut is between 31% and 33% conversion and 46% and 52% backfire.

No one waits for the enemy to die to smoke

The general study already showed that smoke with the opponent missing converts much more. Counting how many enemies were alive in the second of each squeeze, team by team, the size of the missed opportunity appears:

The average is not the die: she sticks to 5 on all teams (from 4.69 to 4.96). The die is the pair of columns on the right. VISION pressed 91% of her smokes with all five enemies alive, and converted 46%; in the 11 moves where there was someone dead on the other side, they converted 82%. The Yandex: 84% of smokes with five alive, 41% conversion: against 80% in the 20 shots with someone dead. Spirit: 89% with five alive, 40%: against 67% in 12.

No team in the tournament exceeds 26% of tight smokes with the opponent missing, and most are below 15%. It's not that anyone knows: it's that the smoke is usually pressed by an internal clock ("it's been five minutes since we've smoked") instead of by a map trigger. The cheapest trigger in the study is the most ignored: the opponent coming back from death.

Two honest caveats. First: the "4 or less" group is small on all teams (11 to 22 moves), so its rate indicates direction, not fine measure. Second: part of the effect is the score itself: whoever just killed someone is usually winning the match. But the difference here is too big to be just that: in the entire base, smokar with the opponent missing converts 58% against 41%, and discounted the gold advantage is still worth +12.7 conversion points and −8.1 backfire (131 moves against 861).

The Eight Maps: Spirit and Yandex vs. VISION

VISION played five maps against Spirit (two in the group stage, three in the upper semifinal) and three against Yandex (the upper final, in the early hours of 08/22). The smoke count is brutal, and has a pattern that doesn't fail in any of the eight.

On all eight maps, whoever won the smoke war won the map. No exception: in the six that VISION won, their conversion rate was higher; in the two that they lost (Game 2 against Spirit and Game 2 against Yandex), the bars switch sides. It's not proof of cause, and the map result also pushes the smoke rate up. But eight out of eight, with the golden score against in almost every attempt on this side, is the most button-like thing this study found.

Adding up: Spirit and Yandex made 43 smokes against her and converted 9, with 26 backfires; VISION made 41, converted 26 and exploded 11.

What Spirit did on the map she won

The first move is the entire recipe in twenty seconds: 21:04, safelane pushed, smoke to the pit on the safe side, turret on the window, 2-0. The Satanic was 7,922 units from the destination, in his offlane. Spirit didn't go after him: it went after something on the map, on the opposite side.

Team Spirit × TEAM VISION

Game 2 · 21:04 · match 8957272720

4×

−45s

State of the lanes

safelanepushed +1.866

midRetreated −1.605

offlaneRetreated −4.726

The decision

pressed on

Jungle wow, next to our offlane

Arrived in

Pit next to our safe lane

Bottom

2 of 5 alive

Isolated target

Their offlane

Isolation

2.249 un.

Result in 2 min

+2Liquid slaughter

+1.589Gold

converted

What happened

• +64sDark Willow
• +68sUndying

The 2 min account closes in +120s

Who's who

Team Spirit
Winter WyvernPhantom LancerNight StalkerClockwerkDoomTEAM VISION
Drow RangerTimbersawPuckDark WillowUndying

The second is the switch that decided the map. At 27:53, both teams press smoke in the same second. Spirit with the safelane and mid pushed; VISION with the three lanes back, because lane is relative, and what was "two pushes" to one side is "none" to the other. In the next two minutes, six heroes fall, all from VISION: +6 and +4,811 gold for Spirit, −6 and −4,811 for VISION, the same six kills on both accounts.

Team Spirit × TEAM VISION

Game 2 · 27:53 · match 8957272720

4×

−45s

State of the lanes

safelanepushed +3.385

midpushed +2.637

offlaneNeutral −349

The decision

pressed on

Pit next to our safe lane

Arrived in

—

Bottom

3 of 5 alive

Isolated target

Their mid

Isolation

1.169 un.

Result in 2 min

+6Liquid slaughter

+4.811Gold

converted

27:53 · The payback

Team

TEAM VISION

Lanes

none

2 min

−6 Slaughter · −4.811 Gold

What happened

• +15sTimbersaw
• +16sUndying
• +28sDrow Ranger
• +30sDark Willow
• +30sPuck
• +34sUndying

The 2 min account closes in +120s

Who's who

Team Spirit
Winter WyvernPhantom LancerNight StalkerClockwerkDoomTEAM VISION
Drow RangerTimbersawPuckDark WillowUndying

And on the same map, fifteen minutes earlier, VISION makes the mistake that this entire guide describes, in the opposite direction. At 17:28 she presses with the offlane pushed and goes to 852 units of Yatoro, who was glued to his team (974 of the closest teammate). Two of his teammates arrive in the 45 seconds. The window closes at 0 to 6. The rule has no side: whoever goes on top of a carry accompanied pays the same price, no matter what shirt he plays.

The Upper Final: What Yandex Did, and What Went Wrong

Yandex tightened 14 smokes on the three maps: 29% converted and 57% exploded: better than Spirit (14% and 64%) and still far from paying. Two moves explain the difference between her maps.

The mistake is in game 1, at 32:24, with no lane pushed, and the group going to 1,319 units of the Satanic, who was in his jungle, to 1,079 of the closest teammate, that is, accompanied. The window closes at 0 to 3 and −5,009 gold. It is the forbidden recipe of this guide in one move: no lane pushed, behind the carry, with the carry close to his team.

Team Yandex × TEAM VISION

Game 1 · 32:24 · match 8958842370

4×

−45s

State of the lanes

safelaneRetreated −6.157

midNeutral +456

offlaneRetreated −6.964

The decision

pressed on

Our safelane

Arrived in

Their mid

Bottom

1 of 5 alive

Isolated target

Our safelane

Isolation

6.329 un.

Result in 2 min

−3Liquid slaughter

−5.009Gold

Backfire

What happened

• +45sCentaur Warrunner (our)
• +51sLina (our)
• +54sUndying (our)

The 2 min account closes in +120s

Who's who

Team Yandex
TerrorbladeLinaCentaur WarrunnerUndyingHoodwinkTEAM VISION
Lone DruidSlardarPangolierKeeper of the LightTechies

Four minutes later, on the same map, came the change that closed the game. At 36:14 VISION presses with the offlane pushed and goes to the river: 5 to 1 and +7.631. Nineteen seconds later Yandex responds with a smoke of its own, to the pit on the safe side, and the answer is what it costs: 1 to 5 and −9.144. It's the same design as the "counter-smoke" of the general study: the two invisible groups meet and whoever had the worst map loses.

TEAM VISION × Team Yandex

Game 1 · 36:14 · match 8958842370

4×

−45s

State of the lanes

safelaneNeutral −141

midRetreated −2.383

offlanepushed +6.753

The decision

pressed on

Our mid

Arrived in

Rio, next to our offlane

Bottom

4 of 5 alive

Isolated target

Their jungle, next to our safelane

Isolation

3.686 un.

Result in 2 min

+4Liquid slaughter

+7.631Gold

converted

36:33 · The payback

Team

Team Yandex

Lanes

Offlane

Destination

Pit next to our safe lane

2 min

−4 Slaughter · −9.144 Gold

What happened

• +77sCentaur Warrunner
• +86sLina
• +87sUndying
• +88sTechies (our)
• +92sTerrorblade
• +97sHoodwink

The 2 min account closes in +120s

Who's who

TEAM VISION
Lone DruidSlardarPangolierKeeper of the LightTechiesTeam Yandex
TerrorbladeLinaCentaur WarrunnerUndyingHoodwink

And the map that Yandex won (game 2) is the most interesting of the three for what it doesn't have: it pressed only two smokes in the entire map (the fewest of any map in this series) while VISION pressed six and blew up four of them, all destined inside the field itself (the base, the offlane, the jungle and the mid of it). In other words: Yandex won the map where they stopped smoking and let VISION spend smoke defending. Against a team that suffers little from smoke, saving the play is a decision, not an omission.

What Spirit and Yandex should do differently

Before the rules, the portrait of the three. The smokes of each one, against all opponents:

The three choose similar: about 60% of the smokes end in the opponent's camp, 64% to 73% are squeezed with an isolated enemy at 3,000 or more, and the interval between smokes is five minutes in all of them. The difference is not in the trigger: it's in what each one does with it.

Yandex is the team that smoks the most of the three (123 moves) and the one that explodes the most (37%): it smoks in a risky context. And it has in its own repertoire the gesture that this guide recommends.Team Yandex, 38:44 against Team Liquid, with the game tied (+784 advantage): mid and offlane pushed, and Liquid's carry appearing at 6,258 from his team, in the river on the opposite side. Yandex ignores the free target, goes to the opponent's safelane, takes out the tower and closes the window at 4-0 and +7.703.

Team Yandex × Team Liquid

Game 1 · 38:44 · match 8944475950

4×

−45s

State of the lanes

safelaneRetreated −2.692

midpushed +5.402

offlanepushed +5.725

The decision

pressed on

Their jungle, next to our offlane

Arrived in

Their safelane

Bottom

3 of 4 alive

Isolated target

Their mid

Isolation

9.901 un.

Result in 2 min

+4Liquid slaughter

+7.703Gold

converted

What happened

• +20sEarth Spirit
• +45sUnderlord
• +49sKeeper of the Light
• +51sHoodwink

The 2 min account closes in +120s

Who's who

Team Yandex
KezInvokerDoomWinter WyvernRubickTeam Liquid
HoodwinkUnderlordEarth SpiritKeeper of the LightWindranger

On the other side, VISION's manual in one move. 13:47 of game 1 against Spirit, game practically tied (+2,764): the three lanes pushed, entry through the opponent's jungle on the safelane side (the side of the map where hunting is cheaper, according to the general study) with Spirit's carry 6,674 away. She doesn't look for a hero: she occupies the map.

TEAM VISION × Team Spirit

Game 1 · 13:47 · match 8946351114

4×

−45s

State of the lanes

safelanepushed +3.527

midpushed +5.033

offlanepushed +5.979

The decision

pressed on

Their offlane

Arrived in

Their jungle, next to our safelane

Bottom

2 of 4 alive

Isolated target

their base

Isolation

5.076 un.

Result in 2 min

+2Liquid slaughter

+1.549Gold

converted

What happened

• +30sRingmaster
• +32sSlardar
• +32sDark Willow (our)
• +54sClockwerk
• +116sDark Willow (our)
• +116sSlardar

The 2 min account closes in +120s

Who's who

TEAM VISION
Nature's ProphetTimbersawEarth SpiritDark WillowWinter WyvernTeam Spirit
ClockwerkDrow RangerSlardarRingmasterDoom

And Spirit's best smoke in the entire series is precisely the one that follows this rule. 13:36 of Game 1 of the Upper Semifinal, with no lane pushed and with the score against (−2.701): the worst possible context. But Satanic was in his offlane, and Spirit went to the jungle on the other side: 4 kills to 2, +1.571 gold. No lane pushed is the state where the general guide says not to smoke; when you smoke anyway, the side of the map is what's left to decide.

Team Spirit × TEAM VISION

Game 1 · 13:36 · match 8957180942

4×

−45s

State of the lanes

safelaneRetreated −5.573

midRetreated −4.282

offlaneNeutral +869

The decision

pressed on

Our offlane

Arrived in

Their jungle, next to our safelane

Bottom

1 of 4 alive

Isolated target

Their offlane

Isolation

5.610 un.

Result in 2 min

+2Liquid slaughter

+1.571Gold

converted

What happened

• +21sSlark
• +38sOgre Magi
• +69sDoom
• +70sTerrorblade (our)
• +78sHoodwink (our)
• +105sRubick

The 2 min account closes in +120s

Who's who

Team Spirit
UndyingTerrorbladeEmber SpiritHoodwinkNight StalkerTEAM VISION
Nature's ProphetDoomSlarkRubickOgre Magi

The five rules, in the order in which the decision happens

1. Against VISION, smoke without a goal is smoke that should not exist. 10% conversion vs. 57% backfire when there was no Roshan or turret in the window; 63% vs. 17% when there was. If the play doesn't have a map target, it's a 1 in 10 bet.
2. Don't walk in the half of the map where Satanic is without reaching him. Prowling his side is the worst of the three possible destinations: +7.4 backfire points in the entire tournament, and −9.1 conversion in the cutout against VISION.
3. Before you go on him, count who can arrive, and count on him to walk. The criterion is not "he's alone": it's "he'll continue alone for the next 45 seconds". Going after the carry without reinforcement converts 56%; with two arriving, 34%. And Satanic walks 3,499 units (median) during smoke: aiming at his position is aiming at the past.
4. If the answer to rule 3 is "people will show up", go the other way. It's the same gesture as Yandex at 38:44 and Spirit at 13:36: their carry away from the team is information about where their team isn't.
5. Never finish the smoke inside your own field. Against VISION this gave 6% conversion in your own lane. If the smoke you want to press ends on your side of the river, what the map is asking for is not smoke: it's wave.

And the defensive corollary, which applies to both sides of the final: VISION will smoke on top of your carry, and your carry is the least rescued of the tournament (0.02 reinforcement on Spirit, 0.11 on Yandex, against 0.34 for it). The defensive gesture is not for the carry to farm closer: it's for the team to stay one rotation away from it, which is what turns "isolated carry" into bait. The teams that close the formation more when they smoke suffer less: VISION closes 658 dispersion units in the 45 seconds it suffers smoke, and responds with its own smoke in 35% of the moves. Spirit is the one that closes the formation the least in the entire table: 347 units.

Method, and where it has a limit

• Base. The same 992 game smokes of 96 TI15 maps from the general study, with full replay parseado. Of this total, 954 have the enemy pos1 alive at the moment of the squeeze: it is the base of every cut of the carry axis. The cut of the three finalists has 351 moves of 55 maps; the cut against VISION, 111 moves of 19 maps.
• Who is the post1 comes from the field of estimated position of the OpenDota, conferred in the public lineup of the two teams in the final of the upper. Where the field is null, the reserve criterion is the highest net worth on the side.
• The ruler is the same. "Converted" and "backfire" are the definitions of the general study, measured in the 120-second window. Nothing has been reclassified for this session.
• The base exit smoke was left out of the dataset, and not just the cuts: pressed before minute 1, it exists to hide where the team comes from. While it was marked in the dataset, it continued to pollute the interval between smokes and the counter-smoke account.
• Two instants, and they are not the same. The carry's position is measured in the second of the squeeze (which is the information the team had) and the destination, 45 seconds later. The "reinforcement" compares the two instants on purpose: it is the difference between what was seen and what was found.
• The interrupted farm counts only creep gold (lane and neutral) from the opponent's pos1, in the 120s before and 120s after the squeeze. Kill and bounty are out of place: they are the result of the fight itself, and what you want to measure is the farm that the smoke took from him.
• Every major cut went through gold advantage control, in bands of 5,000, with a weighted average of the differences. Without this, "getting close to the carry" measures only "team winning".
• Where the number is thin. The "arrived at Satanic" cut has 11 moves, and is written in the text. It indicates direction, not fine measure: the neighboring cuts (2,000 to 4,000, and the side of the map) point to the same side, and that's what supports the reading. Any line with less than 30 moves here should be read like this, and the cut of Yandex against VISION (14 smokes in 3 maps) is exactly one of those.
• The tournament is not over. This guide has been closed with the maps until the upper final of 08/22; the grand final is not yet in the account.
• What this study doesn't see: ward and vision, hero and item composition. A Force Staff, a Glimmer or a ward in the right place changes any of these moves, and none of that counts: smoke here is measured as a map decision, not as a teamfight.