How to Analyze Your Own Dota 2 Replay (6-Step Roadmap)

Almost everyone who "reviews the replay" does the same thing: they watch the match from their own point of view, from beginning to end, finding the same interesting moments that they found live. This is not analysis: it is rewatching.

Analysis is going after a specific question with replay as an instrument. This is the script, in the order in which a coach does it.

Step 0: Before opening the replay, look at four numbers

You don't need a paid tool. Open the game on OpenDota (or on any site that reads the parsy replay) and write down just this:

These four numbers tell you, in two minutes, where you should watch. Without them you will spend 40 minutes seeing what you already knew.

A technical detail that saves frustration: if the match appears without the detailed data (no neutral fields, no positions), it has not been parsed: only the scoreboard, draft and curves remain. The curves are enough for step 1.

Step 1: Find the turning point

Look at the gold advantage curve and mark the hand changes, not the final result. A loss is rarely linear.

The most reviewed match of The International 2026 (LGD × Nigma) has changed hands four times:

Every handshake is a place to watch. And the structural lesson holds true for your match, too: If the chart goes up and down multiple times, both sides are picking fights off favorable terms: the problem is one of posture, not skill.

Mark the minute the advantage started to fall and come back three minutes earlier. It is almost always there that the wrong decision is. That brings us to the most important step.

Step 2: Seek the decision, not the death

The instinct is to look at the lost fight. The mistake is rarely in it.

In the analysis of that match, the moment when the game turned was not the teamfight: it was a kill played before it:

"I realized that disaster starts earlier, with them trying to get a completely empty kill in the safe lane. (…) heroes holding a lane that was completely lost, and they completely losing the wave, taking damage in tower to get the kill."

The result: between minute 15 and 20, LGD went from +1,481 gold to −3,170. A swing of ~4.6k gold and ~9.1k experience in five minutes, with no brilliant play on the other side.

Then, in the replay, when you find the lost fight, ask the right question: *what was my team doing 90 seconds before this fight existed?* The answer is usually one of three

• we were after a hero instead of a piece of map;
• we were defending something that was going to go down anyway;
• We were together, stopped, without any wave advancing.

Step 3: Run the territory check

With the replay open, check four things: all objective, none require judgment:

1. When did T1 in your safe lane fall, and when did T2 fall on the same side? If the distance between the two is huge, you spent the match anchored in the corner of the map. In the aforementioned match it was 10:41 and 38:17: 28 minutes.
2. How many TPs your team spent to react, and what happened to the goal in the next three minutes. TP reaction is bad in 99% of cases: the opponent is walking, you pay for the displacement.
3. Did the invaders in the big jungle happen with the fields formed? Territory without resources is not territory.
4. Which side was the Roshan on in the big fights. Territory on the side of Rosh is worth more than that on the opposite side.

This is the part of the script that yields the most, because they are standard errors: when you find one, you find it in every game.

Step 4: Use the numbers as counter-evidence, not as proof

Here's the trap most fall into when using Dota stats: Entire game metrics don't prove anything about a moment.

The example is great precisely because it contradicts the coach's reading. The reading was that LGD invaded the big jungle in an empty cycle. Measured in the total of the match, LGD farmed more big jungle than Nigma: 263 fields against 234, about 10% more. And lost.

The aggregate does not belie the reading: it measures something else (volume, not timing). And used in reverse, it reinforces: it is not about how much you farm from the big jungle, it is about when.

Applied to you: if your GPM was good and you lost, the GPM does not absolve you. It only says that the problem was not farming.

Step 5: Be fair to yourself (it's method, not kindness)

Analysis that only looks for fault produces the wrong conclusion. Counterpoint is part of the method.

In the analysis of that game, the player who took the harshest criticism for positioning (died 6 times, 4 of them to the hero who ate the map) was also the best warder of the entire game: 16 observers and 21 sentries, more than any of the ten players. The correct conclusion changes with this data: it wasn't laziness, it was positioning. They are different problems, with different fixes.

Do the same with your replay. Before writing down "I played badly", separate: was it a decision, was it execution, was it positioning or was it information you didn't have?

Step 6: Leave with a sentence, not a list

An analysis that ends in ten items doesn't change anything in the next game. End with a sentence in the format:

"In the next match, when [situation], I'm going to [action] instead of [reflex]."

Real-world examples of this format:

• "When my safe lane is under pressure, I'll push another lane instead of TPar."
• "Before I invade the big jungle, I'll look at whether the field has formed instead of looking at whether they're there."
• "After winning a fight, I'll ask you what fields and waves there are instead of going from Rosh on auto."

One sentence per session. Three sessions, three corrected patterns: that's more than most get in a month.

What the replay doesn't show (and why it matters)

Being honest about the boundary of the dice avoids made-up conclusions. Not even the parsed replay of the public APIs answers everything:

• timing of each jungle camp: the list of dead neutrals does not come with a time, so the "farm cycle" is only verifiable in the video, on the match clock;
• use of Twin Gate: it is not exposed anywhere;
• Thin Distance from Hero to Wave: Position data is an accumulated histogram, too thick to judge positioning.

In other words: the numbers tell you where to look; the decision you only see by watching. The two things together are analysis. One without the other is guesswork or spreadsheet.

If you want the version with a coach

The script above is the same one we used in the DotaFlixreplay analysis, where your match is reviewed live by Astini or RdO and the recording stays with you. But it works alone, and it's worth running on your own first, because you arrive at the analysis with the question ready instead of with "see what I did wrong".

For the territory vocabulary that steps 3 and 4 use, start with the three macrogame rules.