After the tower: force the T3 or go hunting?

This guide was born from a ranked game. At 25:31, Radiant took the Dire's second T2 (top and mid) with the gold even and an Aegis on the carry. Two heroes went to the top T3, the carry stayed farming their jungle, and the other two went back to their own side of the map. The Dawnbreaker died alone at 26:21, the team grouped at the T3 at 27:15, lost the fight 3 to 1 plus the Aegis, and the game ended right there. In the post-game, two opinions: "we had to force the T3 to make them TP and lock them in their base" against "we had to go hunt heroes around the map".

Instead of arguing, we measured. We took the 240 maps from The International 2026 and the Esports World Cup 2026 that already had parsed replays for the smoke study, found every time a team took the enemy's second T2 before 30 minutes with no T3 down, and looked at the next three minutes: where the five went, how many enemies TPed to defend, who died, how much gold changed and whether the T3 fell. Then we asked the smaller question, which comes up far more often: you took ONE T2, do you hit the T3 right after or not?

The data

There are 122 situations of a second T2 (71 on the Radiant side) and 379 T2 kills before 30:00, all with the trail of the ten heroes at one point per second. Two definitions the reader needs to keep in mind, because they are the ones Astini asked for:

• Pressuring the T3 is not bringing five. It is any hero of ours within 1,800 units of an enemy T3. How many were there comes in as a variable, further down, but not in the definition.
• Defensive TP is an enemy who jumped more than 2,500 units in one second and landed at their T3 or in their base. It is the "make them TP" of the conversation, measured on the trail.

One caveat that applies to the whole study: pros almost only reach the second T2 when they are already winning. In 80 of the 122 situations the team was 6 thousand gold or more ahead, and in 83 it had just killed Roshan. That is why every important comparison also appears within equal ranges of score, and the "gold even" case, which is the ranked one, has only 11 plays.

What pros do after the second T2

In 93 of the 122 situations (76%), the team went for the T3 in the next three minutes, and won 92% of those games. Going hero hunting happened 9 times (7%), and three of those are the same team, BoomBoys. Those who hunted won 56%, ended the five minutes with a negative gold median and took no T3 afterwards. Those who got hunted (the team spread out and someone died far from the tower) won 40%.

The clip above is the model. Yandex takes Inner Circle's mid T2 at 26:18 with the gold practically even (+1,955), all five go straight to the top T3, and the tower falls in 23 seconds. Inner Circle does not even try the TP: it goes for the fight at the mid T3, loses 7 to 1, and the game ends at 27:43.

Forcing TPs is not the objective

The "force it to make them TP" thesis has a right part and a wrong part. The right one: forcing TPs works almost always. In 74 of the 122 situations all five enemies were pulled to the T3 or the base within three minutes, and the median is two defensive TPs per play. The wrong one: how many TPs you force does not separate winners from losers.

Read the win column: 87% with no TP forced, 86% with one, 79% with two, 85% with three or more. It is flat. The TP is a side effect of any pressure, and a side effect is not an objective. What separates is what happens after the enemy arrives: if there are three or more of ours at the tower to receive whoever landed from the TP, the tower falls or the fight is won; if there are one or two, the forced TP turned into an ambush against us.

One or two heroes at the T3 forcing two or more TPs: 69% wins and the T3 falls within five minutes in 8% of cases. Three or more at the T3 forcing the same TPs: 91% and 64%. And the group that went to the T3 without forcing any TP, because the enemy was not in position to defend, won 27 out of 27.

The clip is Team Nemesis against Spirit at the EWC. One hero at the T3 forces five TPs. The team is not there to receive them, Spirit reorganizes and wins the trade, and Nemesis loses the game. A TP forced with one hero is a pretext, not a plan.

How many go is what decides

Within the range of teams with less than 6 thousand of advantage, which is the ranked range: one or two heroes at the T3 give 61% wins, 27 kills for 30 deaths and the T3 falling within five minutes in 6% of plays. Three or more give 89%, 54 kills for 19 deaths and 56%. With six thousand or more ahead the difference shrinks (88% against 96%), but the tower still falls almost five times more often when three or more go.

The time someone of ours spent at the T3 tells the same story: up to 29 seconds is passing by (76% wins, T3 in 15%); 30 to 89 seconds is pressure (95%, 54%); 90 seconds or more is the tower falling (94%, 89%).

This one is the mirror of the game that opened the guide. Team Nemesis takes Spirit's second T2 at 28:06 with the gold even (−1,147) and Aegis. Two heroes go poke the T3, three stay spread out. Spirit arrives, kills the three spread out, and Nemesis ends the five minutes 9,570 gold down. The T3 never falls.

What about the score?

The obvious objection is that whoever pressures the T3 is whoever is already winning. That is true, and that is why the next comparison is within the same range of score.

• Two thousand or more behind: nobody pressured the T3. The six teams that reached this situation losing won one.
• Even (±2k): those who pressured won 3 of 4, with 17 kills for 3 deaths; those who did not won 2 of 7, with 12 for 16. It is few plays, but the direction is the same as in the other ranges.
• Two to six thousand ahead: 93% against 82% wins, and the T3 falls within five minutes in 57% against 9%.
• Six thousand or more: 96% against 88%, T3 in 64% against 21%.

In every range where someone pressured, pressuring won more and took the tower sooner. The difference in wins is small when the score is already settled; the difference in when the tower falls is not.

Aegis is the context of two thirds of the situations: 83 of the 122 second T2s came within five minutes of a Roshan. The sequence Roshan → two T2s → T3 is the pattern of the professional game, and those with Aegis won 88% against 77% of those without. The game that opened the guide had Aegis, and spent its first two minutes with two heroes at the tower.

Which two T2s fell also matters. Their safelane + mid (the Dire's top and mid T2, when you are Radiant) is the weakest configuration: 74% wins against 94% for mid + offlane. It is the side of the map where their carry farms and defends, and the Dire's top T3 is the farthest from its T2 (3,300 units of walking, against 2,400 in the other two lanes).

Going hunting

The nine plays in which the team went hunting instead of going to the tower have five wins, four losses and a negative gold median at five minutes. The clip is BoomBoys against TEAM VISION at TI15, gold even, second T2 at 29:11: two kills on the hunt, no extra tower, and a 77-minute game that VISION wins.

And the worst outcome, ten times in 122: the team does not go to the tower, spreads out and gets hunted on its own side. Spirit against Yandex at the EWC, gold even (−1,822), second T2 at 28:46: nobody steps on the T3, Yandex comes into Spirit's side, and Yandex's T3 never falls.

The list above is the 11 plays with the gold even, the ranked case. Each team name opens the game on OpenDota; the clock is the one of the second T2. It is worth watching the two Nemesis plays side by side with the Yandex one.

The smaller question: you took a T2, do you hit the T3 next?

The second T2 is a rare moment. The first T2 happens in almost every game, and the decision is the same: do you hit the T3 of that lane right after or not? We measured the 379 T2 kills before 30:00, looking at where the enemy was in the second the tower fell (the distance from the nearest living enemy to the T3 of the same lane) and whether someone of ours reached that T3 in the next two minutes.

Three readings:

1. "Enemy far away" is rare and does not yield towers. In only 27 of the 379 kills was there no enemy within 7 thousand of the T3. Those who hit the T3 in those cases took the tower within three minutes in 6% of cases, and within ten minutes in 38%. The reason is the game, not the team: the T3 has backdoor protection. Without our creeps near it, the tower takes far less damage and regenerates. Hitting the T3 with the enemy far away and without the wave forces TPs, but takes nothing down.
2. When the enemy is on the way (3.5 to 7 thousand from the T3), hitting is what keeps you from dying. Those who hit the T3: 91% wins, 155 kills for 40 deaths, T3 within three minutes in 27%. Those who did not: 81%, 103 kills for 76 deaths, and the T3 did not fall a single time within three minutes. The team that takes the T2 and stays around it, or spreads out, is the one caught by the enemy that was arriving.
3. When the enemy is already at the T3, hitting is still worth it. It is the most common case (220 of the 379): 90% wins and 21% T3 for those who hit, against 73% and 1% for those who did not.

The wave is the key to the backdoor. Those who hit the T3 with the wave (three or more lane creeps killed by our heroes near the T3 within two minutes) took the tower within three minutes in 23% of cases; without the wave, 14%; and those who did not hit, 1%. Took the T2? The next question is "where is the next wave of this lane", not "where is the enemy".

And again the number of heroes: one or two at the T3 take the tower within three minutes in 1% of cases; three or more, in 40%. The lone hero at the T3 is the same story as the second T2, the tower does not fall and the forced TP is the enemy's, onto him.

Their safelane T2 is the one that least turns into a T3 in sequence (8% within three minutes, 26% within ten), and mid is the one that most does (14% and 61%). And the tower that falls in sequence is far more often the T3 after the second T2 of the game (19%) than after the first (6%): at the first one the map is still closed, and the enemy has two other lanes to defend.

The case of Radiant at the Dire's top T2

It is the most common ranked case and the one of the game that opened the guide: you are Radiant, you took the Dire's top T2 (their safelane T2), and the top T3 is 3,300 units away.

There are 74 kills. The Dire's top T3 almost never falls in sequence: 14% within three minutes in the best group, and zero within ten minutes in the ten cases in which the enemy was far away, whether Radiant hit or not. When the enemy was already at the T3, hitting still yielded more (79% wins and 14% T3 against 65% and 5%). When it was on the way, the 20 teams that did not hit won 90% and the 6 that did, 83%, with the tower falling in one case.

The honest reading: the Dire's top T2 is the tower where "hit the T3 right after" is worth the least, because it is the farthest and the most defended, and what decides there is the wave and the group, not the rush. The ten plays with the enemy far away are in the list below, for whoever wants to watch.

What to take to ranked

1. Took the T2? Look at the enemy's five and at the wave, in that order. If this lane's wave is arriving at the T3 and you are three or more, go. If there is no wave, the T3 has backdoor and the TP you force is against you.
2. Forcing TPs with one hero is not a plan. It is the play with the worst outcome of the study when the rest of the team is not there: 61% wins with less than 6k ahead, and the tower falls 6% of the time.
3. Three or more at the T3 is the threshold. It turns 61% into 89% wins with a similar score, and makes the tower fall nine times more often.
4. Do not spread out to farm after the T2. The 76 dead of those who did not hit with the enemy on the way are exactly that: the team that took the tower and turned into five targets.
5. Going hunting is 7% of the professional game, and yields little. Five wins out of nine, negative gold at five minutes, no tower. When the score is even, pros do not do it.
6. With Aegis, the tower is the use of the Aegis. Two thirds of professional second T2s come from a Roshan. Aegis and five minutes ahead on the clock is the moment to go, not to farm.
7. Behind on the score, nobody forces. The six teams 2k behind did not go to the T3, and five lost. Then the conversation is another one (defend and wait for the mistake), and this guide is not about it.

Method and caveats

The replay parse is the same as in the smoke study: trail of the ten heroes at one point per second, deaths with position and gold by reason. The time of each tower comes from OpenDota. The position of the towers is not in the combat log: it is the median position of the killing team, in the second of the kill, measured over 87 to 117 kills per T3. The defensive TP is inferred from the position jump on the trail, so it counts portal and scroll together and is approximate. The gold advantage is the one from the OpenDota chart, per minute.

Three limits the reader needs to know. The "gold even" case has 11 plays, and its rates are direction, not measurement. Every comparison between "pressured" and "did not pressure" has selection: teams do not decide at random, and what the table by score range does is reduce the bias, not eliminate it. And the professional game is played with voice communication and five people who train together: what it shows is what the decision yields when it is executed, not what it yields when half the team did not hear.

The full list, with the clock and the link of each game, is above. The sister guide on what to do when the enemy has the Aegis is Play as if the enemy had Aegis, and the high ground one is High ground: how to attack and defend.