Before you press smoke: the four questions that decide the move

Almost every player treats the Smoke of Deceit as a play. It is not. The smoke is transport: it takes five heroes to a place on the map without them appearing on the opponent's minimap. What decides the move is the photo of the map the second someone presses, and this photo is made of three things that you can look at first.

This guide is the reading of this photo, in the order in which the decision happens. It is not an opinion: each number comes from 992 game smokes from 96 maps of The International 2026, with the full replay parsed: trail of the ten heroes at one point per second, every death with position, and each gold with the reason. The outcome of each smoke was measured in the next two-minute window.

Three terms appear in all tables:

• Converted: The team that smokou won the trade in the window (net kill in favor, with objective or gold).
• Backfire: the team that smokou lost the trade. It's not "nothing happened": it's the smoke turning against it.
• The isolated enemy: the second the smoke is pressed, it's the opposing hero farthest from the closest living teammate to him. It's the natural target of the move: whoever is more alone is the one who can take 5 on 1. The median distance in the tournament is 4,614 units (slightly more than the width of an entire lane), but the half of the map where he's alone changes everything, and that's what Rule 4 is all about.

To calibrate the expectation: in the entire tournament, 47% of the smokes yielded some kill while still under the smoke (35s), 74% yielded a kill in two minutes, and 59% cost at least one kill. The median until the first kill is 28 seconds; until the first kill, 43. In other words: the average smoke is a fight, not a clean ambush.

Four questions, in that order

The order is not aesthetic. Each question only makes sense if the previous one has passed: without a pushed lane there is no good destination, and with the target within your own map there is no destination at all.

The worst pair of answers of the tournament is "no lane pushed" with "go back to own field": 25% conversion and 55% backfire in 162 moves: the largest cell in the entire table. In this state the right decision is not to choose the destination better. It's not to smoke.

Rule 1: Two pushed lanes is the trigger

A pushed lane is coin in the air: 38% conversion vs. 34% backfire. Two saw 54% vs. 20%, and the three saw 69% vs. 17%. The jump is between one and two.

It makes sense for what the wave push is: five creeps walking in their field are five reasons for someone to show up, and they're what makes the point of question 4 exist. If you want to read more about controlling the wave itself, the guide to how to deal with waves is all about the gesture.

Here's the caution that separates this study from a silly correlation: lane pushed and winning go together. Every important cut in this guide has been recalculated within equal gold advantage ranges, comparing smoke with wave pushed versus smoke with wave back in the same leaderboard context. Discounting the score, having two lanes pushed is still worth +8.2 conversion points and −9.0 backfire.

The Team Yandex makes the entire gesture at 18:40 of game 2 against Team Liquid: safelane and offlane pushed, the group enters the opponent's jungle and the window closes at +7 kill without losing anyone. Notice that the mid is neutral: all three are not needed.

Team Yandex × Team Liquid

Game 2 · 18:40 · match 8955653541

4×

−45s

State of the lanes

safelanepushed +2.072

midNeutral −841

offlanepushed +6.003

The decision

pressed on

river, next to our safelane

Arrived in

Their jungle, next to our safelane

Bottom

2 of 4 alive

Isolated target

Their safelane

Isolation

1.729 un.

Result in 2 min

+7Liquid slaughter

+6.951Gold

converted

What happened

• +16sDark Seer
• +32sEmber Spirit
• +35sWinter Wyvern
• +104sWindranger
• +106sClockwerk
• +110sEmber Spirit
• +118sWinter Wyvern

The 2 min account closes in +120s

Who's who

Team Yandex
IoSlarkTidehunterHoodwinkBounty HunterTeam Liquid
Winter WyvernDark SeerWindrangerClockwerkEmber Spirit

Rule 2: the mid lane is the lane that counts

It's not just "how many": it's which ones. Within equal ranges of gold advantage, the mid push is the one that moves the result the most: +10.7 conversion points and −10.9 backfire. The safelane is worth half that, and the offlane is worth less than zero.

The reading is geometric. The wave of the pushed mid is the only one that threatens all three fronts at the same time: it opens the river, touches the middle T2 and forces the opponent to respond from anywhere on the map. The pushed safelane threatens a corner; the pushed offlane, the corner farthest from your base, and that's precisely why it costs instead of paying.

The eight settings show the same thing from another angle. Each row of the table opens: inside it are the moves that formed that number, with the game and the minute of each one, and one of them animated on the map.

Two lines say almost everything. "Safelane + mid" converts 64% and explodes 14%: the best possible pair, almost as good as having all three. And "only the offlane" converts 27% and explodes 38%, the worst of the eight: the wave is there, in their field, the whole team sees it, but it's alone, it doesn't pull rotation, it doesn't generate a goal, and whoever crosses behind it is the only one out of position. Notice that there the average advantage before the move is practically zero: it's not the effect of losing, it's the configuration.

Rule 3: the fate that always costs is your own field

With two or more lanes pushed, almost every destination pays: their lane converts 63%, their jungle 66%. With one lane, everything drops about ten points. And with none, only the river and the pit still hold the line, because there you are not crossing anything.

What the matrix shows in all three rows is the last column: going back to one's own field is the only answer that gets worse in any map state: 25% with no lane, 20% with one, and even with two or three only 42%. It's not fate that's bad per se: it's just that almost every smoke for one's own field is a defensive smoke, made by those who are already behind.

Two corners of the map deserve their own name. Your own mid converts 21% and explodes 52% in 85 moves; and your own base is the worst region in the entire study: 10% conversion vs. 68% backfire in 31 moves. Before you draw the wrong rule, look at what these moves are: strongly negative average advantage, clock on late, enemy inside your map. The number doesn't say "never smoke back": it says that when you need it, smoke isn't the remedy.

The Team Spirit at 19:28 of Game 1 against Iron Wing shows the opposite path with little material: one lane pushed, the safelane, with the other two retreated. Instead of crossing the lane, the group enters the jungle and exits 5-1 in the window.

Team Spirit × Iron Wing

Game 1 · 19:28 · match 8955247801

4×

−45s

State of the lanes

safelanepushed +2.670

midRetreated −2.392

offlaneRetreated −6.358

The decision

pressed on

Our safelane

Arrived in

Their jungle, next to our safelane

Bottom

1 of 5 alive

Isolated target

Their safelane

Isolation

7.199 un.

Result in 2 min

+4Liquid slaughter

+267Gold

converted

What happened

• +12sDoom
• +16sShadow Fiend (our)
• +17sEmber Spirit
• +21sHoodwink
• +102sLifestealer
• +118sUndying
• +122sDoom

The 2 min account closes in +120s

Who's who

Team Spirit
ClockwerkShadow FiendLinaDark WillowUnderlordIron Wing
DoomHoodwinkUndyingEmber SpiritLifestealer

Rule 4: Where is the enemy isolated

The second the smoke is pressed, the position of the five living enemies is in the replay. Measuring the most isolated of them (the distance from the nearest living companion) and the region they are in:

The rule is short: if the "isolated" hero is inside your map, he is not isolated: he is in front of his team, which comes behind. You are hunting the tip of a formation, and it is the worst cut on the list (33% / 49% in 189 moves). Breaking down by region, the winger is the target stopped in your mid: 22% conversion and 73% backfire in 37 shots.

There's also a finer effect, and it's on the side of the map, not terrain: in the opponent's jungle, the target on the side of your safelane converts 46% and explodes 21%; the same hero, also alone, also in their jungle, but on the side of your offlane, converts 37% and explodes 37%. His escape lane is all to their side, their tower is close and yours is far away. It's the difference between hunting with your tower behind you and hunting with their tower.

The most useful counterexample in the study is this, because the team gets the first three questions right and loses the move anyway: Team Yandex, 43:33 of game 1 against Team Liquid. The three lanes pushed, the five under the smoke, destination in the opponent's jungle. The target is a hero almost 10,000 units away from his team: he seems like the most free target of the tournament. But he's on our offlane's side. First killed in 33 seconds, and the window closes at −2 kill and −7,898 gold.

Team Yandex × Team Liquid

Game 1 · 43:33 · match 8955549026

4×

−45s

State of the lanes

safelanepushed +6.668

midpushed +7.156

offlanepushed +10.799

The decision

pressed on

Our safelane

Arrived in

Their jungle, next to our safelane

Bottom

5 of 5 alive

Isolated target

Their jungle, next to our offlane

Isolation

9.885 un.

Result in 2 min

−2Liquid slaughter

−7.898Gold

Backfire

What happened

• +33sWindranger (our)
• +39sTusk
• +40sIo (our)
• +46sHoodwink (our)

The 2 min account closes in +120s

Who's who

Team Yandex
WindrangerIoLargoHoodwinkEnchantressTeam Liquid
MiranaDawnbreakerNature's ProphetTuskTiny

The Strongest Discriminator: Goal in the Window

If you can only save one number from this guide, save this one:

Smoke that exists to get a kill is currency in the air. Smoke that exists to open an objective (clear pit vision, secure the Roshan, take away the tower that the pushed wave is already hitting) converts 80% against 23%, more than three times. And notice how Rule 1 feeds this one: pushed wave is precisely what makes the objective exist.

Recipes, ordered by what they deliver

All combinations of state of the lanes × target with at least eight moves.

The first four lines have one thing in common, and it's the only thing they have in common: two pushed lanes, or the pushed safelane, and the group going to their field. The last three too: the group returning to their own field.

The move that sums it all up

Eight seconds separate two smokes at minute 45 of game 2 of BoomBoys × Iron Wing. The BoomBoys tightens with the three lanes pushed and goes to the opponent's mid. The Iron Wing responds in the sequence with a defensive smoke, into the jungle itself, and as lane is relative, what was "three pushes" to one side is "no push" to the other.

In the next two minutes, eight heroes fall, all from Iron Wing. They are the same eight kills: +7 and +12,516 gold on the BoomBoysaccount, −8 and −13,172 on the Iron Wingaccount. Same map, same ten heroes, same second: the whole difference is in who had the map stretched out and who was trying to put out the fire under the smoke.

BoomBoys × Iron Wing

Game 2 · 45:32 · match 8956821666

4×

−45s

State of the lanes

safelanepushed +3.902

midpushed +1.674

offlanepushed +9.354

The decision

pressed on

Pit next to our safe lane

Arrived in

Their mid

Bottom

3 of 5 alive

Isolated target

Their mid

Isolation

2.595 un.

Result in 2 min

+7Liquid slaughter

+12.516Gold

converted

45:40 · The payback

Team

Iron Wing

Lanes

none

Destination

Jungle wow, next to our offlane

2 min

−8 Slaughter · −13.172 Gold

What happened

• +35sDoom
• +42sLion
• +50sLina
• +52sLion
• +59sDark Seer
• +63sBounty Hunter
• +116sDark Seer
• +123sBounty Hunter

The 2 min account closes in +120s

Who's who

BoomBoys
IoPuckCentaur WarrunnerHoodwinkWinter WyvernIron Wing
Dark SeerBounty HunterLionLinaDoom

That's why "no lane pushed → our own field" is the worst cell in the study. It's almost always this play: the team that's already behind using the smoke to try to defend.

And on the other side, the right gesture with ease: TEAM VISION, 29:02 of game 3 against BoomBoys. The three pushes, entry through their jungle, 6 to 0 and +6.471. The isolated enemy was in the jungle on the side of the safelane: the right side of the map to hunt.

TEAM VISION × BoomBoys

Game 3 · 29:02 · match 8955456108

4×

−45s

State of the lanes

safelanepushed +9.349

midpushed +8.061

offlanepushed +7.975

The decision

pressed on

Our offlane

Arrived in

Their jungle, next to our offlane

Bottom

2 of 5 alive

Isolated target

Their jungle, next to our safelane

Isolation

5.688 un.

Result in 2 min

+6Liquid slaughter

+6.471Gold

converted

What happened

• +19sLina
• +24sMirana
• +106sMirana
• +107sLina
• +108sBane
• +110sUnderlord

The 2 min account closes in +120s

Who's who

TEAM VISION
Nature's ProphetSlardarEmber SpiritBounty HunterWinter WyvernBoomBoys
Lone DruidLinaUnderlordMiranaBane

The four mistakes that cost the most

Each of these is a real slice of the tournament, not a rule of common sense. The number is the backfire rate: the proportion of shots in which the team that smoked lost the exchange.

1. Smokar without a pushed lane and go back to one's own field: 55% backfire (25% converts, n=162). The biggest cell in the entire table, and the worst. If the map is like this, the play is to push a lane, not disappear.
2. Hunt down the isolated enemy inside your map: 49% backfire (33% converts, n=189). He's not alone: he's in front of his team. In his own mid, the backfire goes to 73%.
3. Using the pushed offlane as a trigger: 38% backfire (27% converts, n=122). On its own, it's the worst of the eight settings. The wave is in their court and doesn't pull anyone. Discounting the score, the pushed offlane is worth −4.2 conversion points.
4. Smokar with the offlane pushed and back to the field: 52% backfire (14% converts, n=50). The combination of the two mistakes above, and the recipe with the lowest conversion of the entire study.

When smoke is not the medicine

Three contexts in which the number says to keep the smoke:

• Behind over 5,000 gold: 21% converts, 58% explodes. Smokar doesn't recover lost match: it only accelerates. In the opposite lane, ahead of 5k or more, it's 65% / 15%.
• Right after the opponent smokars: when the other team smokes in the 30 seconds around its use, its smoke converts 37% and explodes 40% (n=171). The two invisible groups meet and whoever had the worst map loses.
• Between 30 and 45 minutes: is the highest backfire range of the tournament: 43% converts, but 40% explodes (n=321). Teleport and vision items are ready on both sides, and the position error costs the match instead of costing a kill.

What changed when the sample tripled

This guide was born with 254 moves from the 23 maps of the main event. With the group stage of the six teams entering, the base became 936 moves from 89 maps, and two conclusions changed owners. The 992 moves of 96 maps today (the main event to the final of the upper) did not change any of them. They are recorded, because a study that only publishes what survives does not let anyone learn:

• The lane that counts became the mid. With 23 maps, the pushed safelane was worth +14.3 conversion points and the mid +9.1. With 96, the mid is worth +10.7 and the safelane drops to +5.3. What hasn't changed in either of the two rounds: the pushed offlane is worth less than zero.
• The jungle trap on the offlane side emptied in half. With 30 moves it was 20% / 53% against 54% / 13% on the other side: a reversal. With 117 moves it is 37% / 37% against 46% / 21%: the effect remains in the same direction, but it is a difference in quality, not a reversal.

What stood in every round: goal in the window is the strongest discriminator, two pushed lanes is the threshold, the offlane alone is the worst setup, and going back to one's own field is the fate that always costs.

The extra session: the three finalists, and the carry as the axis

This guide measures the map photo. He didn't look at something that decides a lot of moves: where was the opposing carry in the second of the squeeze. This question became a separate session, with the three teams that reached the end of TI15: learning from the TI finalists.

Its summary in one line: either you get to their carry, or you go to the other side of the map; Hanging around the half of the map he's on, without getting to him, is the worst of the three destinations (+7.6 backfire points, discounting the score). And against TEAM VISION, which is the toughest smokar team in the tournament, not even getting to him worked: it was like that on her eight maps against Spirit and Yandex.

The third chapter: who is inside the smoke

This guide and the finalists' guide measure the play. Neither of them looks at the squad, and the difference it makes is big: area control heroes add up to 12 conversion points when they are under smoke, and cover heroes on the opposing team take up to 14 points from theirs. This is in the hero inside the smoke.

Method, and where it has a limit

• Base. 992 game smoke from 96 TI15 maps, with the full replay parsed: trail of the ten heroes at one point per second, every hero kill with position, and every drop of gold with the reason. The outcome is measured within 120 seconds of use.
• The base is not uniform, and that matters. There are the 30 maps of the main event (all eight teams) plus the 66 group stage maps of the six closely followed teams: BoomBoys, Team Yandex, Team Spirit, TEAM VISION, Team Liquid and Nigma Galaxy. In other words: the main event is in one piece, the group stage is by the cut of these six. Only matches with replay parse enter.
• Relative lane, not geographic. Every cut uses the team's safelane/mid/offlane: the same piece of map is safelane to one side and offlane to the other, and that's the variable that decides. "Pushed" is the front of the wave to more than 1,500 units inside the enemy field.
• Position of the waves. No parser event has a creep coordinate: checked on the 58 types it emits; only ward brings x/y. The front of each lane is rebuilt by the kill of lane creep: the combat log gives the gold with the motif and the lane gives where the hero was at that second, in a 60-second window before the smoke.
• The two pits of the Roshan were measured, not guessed: by the median position of the heroes of the team that killed, in the deaths of Roshan of these maps.
• Every major cut went through the gold advantage control: comparison within four scoring ranges, weighted average of the differences. Without this, "lane pushed" measures only "team winning".
• Out of the account: the base exit smoke (pressed before minute 1) is discarded from the dataset, not just the cuts: it exists to hide where the team comes from, not to hunt, and while it was marked in the dataset it continued to contaminate the interval between smokes and the counter-smoke account. Also besides a map without replay salt in the collection.
• Where the number is thin. The regions shown have at least 8 bids and the recipes at least 8; the n is written next to each line. Clipping less than 30 bids indicates direction, not fine measurement: that's exactly what the section above showed happening.

If you want to apply this to your own game, the path is the same as in the guide on how to analyze your own replay: pause the second the smoke was pressed and answer the four questions with what's on the screen. You don't need to parser any to see how many lanes are pushed.