A record of 1,653 matches, and a close read of the 6 played since coming back. Every finding traces to one feedback loop in the model of Dota as a resource-throughput system. The detectors describe; recurrence across matches does the ranking, because one game is noise.
The account opened in April 2015. Almost all of it was played before the break, so the averages below describe a player who no longer exists. They are a baseline, not a target.
2,289 days of silence between 17 May 2020 and 24 August 2026. Everything before that gap is history; everything after it is the sample this report is actually about.
The eight heroes with the most games on the account, and how they did.








Aggregate shape of the games played since 24 August. Farm rate is the one number that beats the old career average; the percentile column says how it reads against every player on the same hero.
Every detector against every game, ordered by how many matches each pattern appears in. A block is a game the detector fired on, and the darker it is the further past its own threshold that game went. Open a row for what it means and the plan attached to it. The top row is the binding constraint, and it is the only one worth acting on until it stops appearing.
It appears in 5 of the 6 matches, which makes it the binding constraint. Give it ten to twenty-five games before opening another.
24 Aug W
24 Aug W
24 Aug W
25 Aug W
25 Aug W
26 Aug LBeing seen, read directly. A death outside every recorded fight is one they found rather than won, which means the wave told them where you were before you got there.
Look at where you were standing in the ten seconds before each one.
The cheapest buffer in the game, and the only thing that turns a wave on the far side of the map into a wave you can take. Without one, the engine runs at walking speed.
The detector counts scrolls in the purchase log, not visits to base, so it reads one number: scrolls bought per ten minutes of the match.
Being seen. A concentrated killer means they are predicting where you clear waves. This is a habit, not a fight problem.
Change where your next appearance happens, not how you fight.
The engine lost its gain. Clear rate feeds gold, gold feeds strength, strength feeds clear rate, and the loop ran backwards.
Check the minute your first wave-clearing item finished.
One map, two farms. Two engines share it, so a lane you never contest is a lane that pays them full price. Their percentile is the measure of how little the map cost them.
Check whether you ever took anything from their half of the river.
The tilt spiral. Every second dead is a second the engine is off, and the respawn timer grows as the game goes.
Treat the buyback and the walk back as part of the cost of the fight.
The line under each match is your gold advantage across the game, drawn from your side. Flat is even, below the rule is behind. Open a match for its build, its bracket percentiles, the minutes after every fight, and the plans that follow from it.
loss
against Mars at 617 gpm
assumed role: OfflaneWave access is the moment your hero can kill a whole creep wave from outside its reach, without trading health for it. Before that moment every wave costs you time and position; after it, waves become gold at a rate that compounds. The clock below is a proxy for that moment: the first of six items to finish — Manta Style, Aghanim’s Scepter, Mjollnir, Radiance, Battle Fury, Maelstrom. Those six are the ones the corpus raises most often in wave-clearing context, which makes them a measure of what gets said rather than of what clears a wave. A hero whose spells already clear one crosses the threshold without buying anything on the list, and the proxy will miss it.
no item none of the six landed. Read it as a question rather than a verdict: either the waves were walked into all game, or this hero already cleared them with a spell
8:51Arcane Boots
18:30Pipe of Insight
18:30Ring of Tarrasque
23:35Mekansm
28:44Guardian Greaves
31:14Mithril Hammer
40:26Black King BarEvery item over 1400 gold, stamped with the clock time it finished. The ones ringed in teal are the six above.
The comparison against public games comes from OpenDota’s scenarios endpoint. It carries no rank filter, so it is every bracket rolled together rather than the players above you, and it counts games in seven coarse buckets that stop at minute 30. Read it as roughly where the timing sits, not as a rank.
Each bar is where this game sits among games on this hero in your own bracket, not across every bracket, so it reads straighter than the all-bracket number on the card. The notch is the middle of that bracket. Nothing here is coloured for good or bad; a row marked ↓ is one where the small number is the good end. Lane efficiency 69%, position 3, teamfight participation 73%, 2 buybacks.
Between 32:00 and 37:00, Mars earned 3,204 more gold than you — 917 a minute against your 276. This is the window your own earnings fell furthest behind the enemy in your position, which is the only gold comparison in this report that is about you rather than about all ten players.
| Minute | You earned | Mars earned | Difference |
|---|---|---|---|
| 32:00 – 33:00 | 391 | 414 | -23 |
| 33:00 – 34:00 | 155 | 935 | -780 |
| 34:00 – 35:00 | 463 | 707 | -244 |
| 35:00 – 36:00 | 275 | 1,551 | -1,276 |
| 36:00 – 37:00 | 100 | 981 | -881 |
Explained: you were holding a base. 5 of your own buildings fell inside these five minutes. You were defending, not farming, and the gap is what defence costs. Nothing here is a clear-rate error, and the question worth asking is about the twenty minutes before it.
Where the five minutes went, from the replay position track: 80s dead; 82s in your base; 64s in your jungle; 52s in mid lane, your half, standing still for 52s of it and walking for 168s.
Teal is your team, ember is theirs, and your own name is marked. Places are approximate: the map is cut into bases, lanes, river and the two jungles, and the diagonal decides whose half you were on.
Last hits taken by you on Razor and by Mars in the 3 minutes after each fight ends. This is loop R1 measured at the moment it usually leaks.
The verdict column reads last hits, not gold, on purpose. Gold in this window includes kills, assists and buildings, so a team that won the fight and kept fighting can show a fat gold rate having touched no wave at all — which is the exact leak this table exists to catch. The gold rate sits beside it because it says whether the map paid you, and the gap between the two numbers is the interesting part.
Mars is not their carry. It is the enemy OpenDota put in the same position as you (position 3), because the honest comparison is against whoever contests the same ground.
| The fight | Window scored | You, Razor | Mars | |
|---|---|---|---|---|
| 6:30 – 7:16 | 7:16 – 10:16 | 15 lh 429 gpm | 14 lh 521 gpm | held |
| 12:29 – 13:07 | 13:07 – 16:07 | 11 lh 268 gpm | 16 lh 446 gpm | held |
| 14:52 – 15:31 | 15:31 – 18:31 | 15 lh 362 gpm | 28 lh 755 gpm | lost |
| 20:04 – 20:52 | 20:52 – 23:52 | 19 lh 640 gpm | 22 lh 479 gpm | held |
| 22:02 – 22:35 | 22:35 – 25:35 | 23 lh 545 gpm | 24 lh 573 gpm | held |
| 24:38 – 25:29 | 25:29 – 28:29 | 6 lh 267 gpm | 26 lh 761 gpm | lost |
| 28:21 – 28:59 | 28:59 – 31:59 | 16 lh 599 gpm | 16 lh 526 gpm | held |
| 30:53 – 31:32 | 31:32 – 34:32 | 12 lh 377 gpm | 32 lh 624 gpm | lost |
| 34:50 – 35:28 | 35:28 – 38:28 | 8 lh 516 gpm | 27 lh 1,067 gpm | lost |
| 37:19 – 38:06 | 38:06 – 41:06 | 25 lh 500 gpm | 21 lh 647 gpm | held |
| 42:51 – 44:09 | 44:09 – 47:09 | 0 lh 0 gpm | 0 lh 0 gpm | held |
7Spectre
3Invoker
1Mars
1Shadow ShamanOne trigger you can perceive, one action nobody else can move, one rate a detector can check. Run one at a time.
win
against Juggernaut at 521 gpm
assumed role: Safelane carryWave access is the moment your hero can kill a whole creep wave from outside its reach, without trading health for it. Before that moment every wave costs you time and position; after it, waves become gold at a rate that compounds. The clock below is a proxy for that moment: the first of six items to finish — Manta Style, Aghanim’s Scepter, Mjollnir, Radiance, Battle Fury, Maelstrom. Those six are the ones the corpus raises most often in wave-clearing context, which makes them a measure of what gets said rather than of what clears a wave. A hero whose spells already clear one crosses the threshold without buying anything on the list, and the proxy will miss it.
Radiance at 17:02 the first of the six to finish. Of 1,673 public games that bought it on this hero, 26% bought it earlier than you
5:19Phase Boots
13:24Sacred Relic
17:02Radiance
19:36Sange
25:13Sange and Yasha
25:13Yasha
29:58Aghanim's Scepter
32:27Mithril Hammer
32:50Skull Basher
36:51Abyssal Blade
40:10Hyperstone
41:28Platemail
41:50Assault CuirassEvery item over 1400 gold, stamped with the clock time it finished. The ones ringed in teal are the six above.
The comparison against public games comes from OpenDota’s scenarios endpoint. It carries no rank filter, so it is every bracket rolled together rather than the players above you, and it counts games in seven coarse buckets that stop at minute 30. Read it as roughly where the timing sits, not as a rank.
Each bar is where this game sits among games on this hero in your own bracket, not across every bracket, so it reads straighter than the all-bracket number on the card. The notch is the middle of that bracket. Nothing here is coloured for good or bad; a row marked ↓ is one where the small number is the good end. Lane efficiency 69%, position 1, teamfight participation 55%, 0 buybacks.
Between 20:00 and 25:00, Juggernaut earned 2,138 more gold than you — 939 a minute against your 511. This is the window your own earnings fell furthest behind the enemy in your position, which is the only gold comparison in this report that is about you rather than about all ten players.
| Minute | You earned | Juggernaut earned | Difference |
|---|---|---|---|
| 20:00 – 21:00 | 468 | 1,564 | -1,096 |
| 21:00 – 22:00 | 701 | 444 | +257 |
| 22:00 – 23:00 | 263 | 822 | -559 |
| 23:00 – 24:00 | 391 | 1,064 | -673 |
| 24:00 – 25:00 | 735 | 802 | -67 |
Explained: you were in a fight. 68 of the 300 seconds sat inside a recorded teamfight. Read this against the table above instead: the question is not what you earned during the fight, it is what you took in the minutes after it.
Where the five minutes went, from the replay position track: 60s dead; 84s in your jungle; 52s in bottom lane, your half; 39s in their jungle, standing still for 63s of it and walking for 159s.
Teal is your team, ember is theirs, and your own name is marked. Places are approximate: the map is cut into bases, lanes, river and the two jungles, and the diagonal decides whose half you were on.
Last hits taken by you on Lifestealer and by Juggernaut in the 3 minutes after each fight ends. This is loop R1 measured at the moment it usually leaks.
The verdict column reads last hits, not gold, on purpose. Gold in this window includes kills, assists and buildings, so a team that won the fight and kept fighting can show a fat gold rate having touched no wave at all — which is the exact leak this table exists to catch. The gold rate sits beside it because it says whether the map paid you, and the gap between the two numbers is the interesting part.
Juggernaut is not their carry. It is the enemy OpenDota put in the same position as you (position 1), because the honest comparison is against whoever contests the same ground.
| The fight | Window scored | You, Lifestealer | Juggernaut | |
|---|---|---|---|---|
| 3:42 – 4:20 | 4:20 – 7:20 | 14 lh 295 gpm | 7 lh 234 gpm | held |
| 10:51 – 11:29 | 11:29 – 14:29 | 19 lh 558 gpm | 10 lh 199 gpm | held |
| 11:46 – 12:38 | 12:38 – 15:38 | 25 lh 619 gpm | 13 lh 252 gpm | held |
| 12:56 – 13:29 | 13:29 – 16:29 | 21 lh 470 gpm | 14 lh 286 gpm | held |
| 16:50 – 17:27 | 17:27 – 20:27 | 21 lh 594 gpm | 16 lh 460 gpm | held |
| 18:33 – 19:16 | 19:16 – 22:16 | 26 lh 638 gpm | 23 lh 810 gpm | held |
| 19:40 – 20:26 | 20:26 – 23:26 | 20 lh 477 gpm | 32 lh 943 gpm | lost |
| 22:45 – 23:27 | 23:27 – 26:27 | 26 lh 583 gpm | 37 lh 764 gpm | held |
| 28:43 – 29:40 | 29:40 – 32:40 | 17 lh 1,190 gpm | 32 lh 565 gpm | lost |
| 31:30 – 32:36 | 32:36 – 35:36 | 56 lh 1,256 gpm | 1 lh 245 gpm | held |
| 34:47 – 35:41 | 35:41 – 38:41 | 18 lh 779 gpm | 49 lh 1,262 gpm | lost |
| 38:35 – 39:28 | 39:28 – 42:28 | 13 lh 1,638 gpm | 10 lh 395 gpm | held |
| 40:33 – 41:11 | 41:11 – 44:11 | 3 lh 520 gpm | 9 lh 164 gpm | lost |
1Riki
1Sniper
1LionOne trigger you can perceive, one action nobody else can move, one rate a detector can check. Run one at a time.
win
against Riki at 438 gpm
assumed role: OfflaneWave access is the moment your hero can kill a whole creep wave from outside its reach, without trading health for it. Before that moment every wave costs you time and position; after it, waves become gold at a rate that compounds. The clock below is a proxy for that moment: the first of six items to finish — Manta Style, Aghanim’s Scepter, Mjollnir, Radiance, Battle Fury, Maelstrom. Those six are the ones the corpus raises most often in wave-clearing context, which makes them a measure of what gets said rather than of what clears a wave. A hero whose spells already clear one crosses the threshold without buying anything on the list, and the proxy will miss it.
Manta Style at 20:00 the first of the six to finish. Of 334 public games that bought it on this hero, 11% bought it earlier than you
8:55Power Treads
15:19Yasha
20:00Manta Style
24:58Mithril Hammer
28:35Black King Bar
33:48Platemail
41:17Shiva's Guard
45:39Aghanim's ScepterEvery item over 1400 gold, stamped with the clock time it finished. The ones ringed in teal are the six above.
The comparison against public games comes from OpenDota’s scenarios endpoint. It carries no rank filter, so it is every bracket rolled together rather than the players above you, and it counts games in seven coarse buckets that stop at minute 30. Read it as roughly where the timing sits, not as a rank.
Each bar is where this game sits among games on this hero in your own bracket, not across every bracket, so it reads straighter than the all-bracket number on the card. The notch is the middle of that bracket. Nothing here is coloured for good or bad; a row marked ↓ is one where the small number is the good end. Lane efficiency 69%, position 3, teamfight participation 52%, 0 buybacks.
Between 35:00 and 40:00, Riki earned 2,639 more gold than you — 836 a minute against your 308. This is the window your own earnings fell furthest behind the enemy in your position, which is the only gold comparison in this report that is about you rather than about all ten players.
| Minute | You earned | Riki earned | Difference |
|---|---|---|---|
| 35:00 – 36:00 | 365 | 722 | -357 |
| 36:00 – 37:00 | 580 | 347 | +233 |
| 37:00 – 38:00 | 341 | 1,434 | -1,093 |
| 38:00 – 39:00 | 157 | 1,279 | -1,122 |
| 39:00 – 40:00 | 100 | 400 | -300 |
Explained: you were dead for much of it. The replay track goes out for 74 of the 300 seconds, which is death. That is the gap’s cause. It belongs to the detectors about dying, not to this one.
Where the five minutes went, from the replay position track: 74s dead; 78s in your jungle; 51s in their jungle; 44s in the river, standing still for 88s of it and walking for 138s.
Teal is your team, ember is theirs, and your own name is marked. Places are approximate: the map is cut into bases, lanes, river and the two jungles, and the diagonal decides whose half you were on.
Last hits taken by you on Razor and by Riki in the 3 minutes after each fight ends. This is loop R1 measured at the moment it usually leaks.
The verdict column reads last hits, not gold, on purpose. Gold in this window includes kills, assists and buildings, so a team that won the fight and kept fighting can show a fat gold rate having touched no wave at all — which is the exact leak this table exists to catch. The gold rate sits beside it because it says whether the map paid you, and the gap between the two numbers is the interesting part.
Riki is not their carry. It is the enemy OpenDota put in the same position as you (position 3), because the honest comparison is against whoever contests the same ground.
| The fight | Window scored | You, Razor | Riki | |
|---|---|---|---|---|
| 9:27 – 10:04 | 10:04 – 13:04 | 19 lh 596 gpm | 9 lh 293 gpm | held |
| 10:18 – 11:08 | 11:08 – 14:08 | 20 lh 577 gpm | 10 lh 247 gpm | held |
| 28:28 – 29:15 | 29:15 – 32:15 | 17 lh 415 gpm | 19 lh 644 gpm | held |
| 40:06 – 41:21 | 41:21 – 44:21 | 16 lh 948 gpm | 6 lh 212 gpm | held |
| 42:46 – 43:26 | 43:26 – 46:26 | 19 lh 1,084 gpm | 4 lh 168 gpm | held |
| 47:35 – 48:33 | 48:33 – 51:33 | 17 lh 732 gpm | 0 lh 35 gpm | held |
3RikiOne trigger you can perceive, one action nobody else can move, one rate a detector can check. Run one at a time.
win
against Abaddon at 247 gpm
assumed role: OfflaneWave access is the moment your hero can kill a whole creep wave from outside its reach, without trading health for it. Before that moment every wave costs you time and position; after it, waves become gold at a rate that compounds. The clock below is a proxy for that moment: the first of six items to finish — Manta Style, Aghanim’s Scepter, Mjollnir, Radiance, Battle Fury, Maelstrom. Those six are the ones the corpus raises most often in wave-clearing context, which makes them a measure of what gets said rather than of what clears a wave. A hero whose spells already clear one crosses the threshold without buying anything on the list, and the proxy will miss it.
no item none of the six landed. Read it as a question rather than a verdict: either the waves were walked into all game, or this hero already cleared them with a spell
6:50Arcane Boots
12:10Force Staff
16:23Oblivion Staff
18:58Witch Blade
24:04Dragon Lance
24:04Hurricane Pike
25:20Aghanim's ShardEvery item over 1400 gold, stamped with the clock time it finished. The ones ringed in teal are the six above.
The comparison against public games comes from OpenDota’s scenarios endpoint. It carries no rank filter, so it is every bracket rolled together rather than the players above you, and it counts games in seven coarse buckets that stop at minute 30. Read it as roughly where the timing sits, not as a rank.
Each bar is where this game sits among games on this hero in your own bracket, not across every bracket, so it reads straighter than the all-bracket number on the card. The notch is the middle of that bracket. Nothing here is coloured for good or bad; a row marked ↓ is one where the small number is the good end. Lane efficiency 54%, position 3, teamfight participation 40%, 0 buybacks.
Between 5:00 and 10:00, Abaddon earned -238 more gold than you — 217 a minute against your 264. This is the window your own earnings fell furthest behind the enemy in your position, which is the only gold comparison in this report that is about you rather than about all ten players.
| Minute | You earned | Abaddon earned | Difference |
|---|---|---|---|
| 5:00 – 6:00 | 255 | 90 | +165 |
| 6:00 – 7:00 | 302 | 344 | -42 |
| 7:00 – 8:00 | 384 | 251 | +133 |
| 8:00 – 9:00 | 124 | 215 | -91 |
| 9:00 – 10:00 | 258 | 185 | +73 |
Nothing explains it. You were alive, out of every fight, and your buildings were standing. Nobody stopped you, and the map still paid them more than it paid you. This is the only branch that is entirely yours.
Where the five minutes went, from the replay position track: 23s dead; 219s in top lane, your half; 27s in top lane, their half; 12s in their jungle, standing still for 68s of it and walking for 208s.
Teal is your team, ember is theirs, and your own name is marked. Places are approximate: the map is cut into bases, lanes, river and the two jungles, and the diagonal decides whose half you were on.
Last hits taken by you on Silencer and by Abaddon in the 3 minutes after each fight ends. This is loop R1 measured at the moment it usually leaks.
The verdict column reads last hits, not gold, on purpose. Gold in this window includes kills, assists and buildings, so a team that won the fight and kept fighting can show a fat gold rate having touched no wave at all — which is the exact leak this table exists to catch. The gold rate sits beside it because it says whether the map paid you, and the gap between the two numbers is the interesting part.
Abaddon is not their carry. It is the enemy OpenDota put in the same position as you (position 3), because the honest comparison is against whoever contests the same ground.
| The fight | Window scored | You, Silencer | Abaddon | |
|---|---|---|---|---|
| 6:46 – 7:32 | 7:32 – 10:32 | 5 lh 255 gpm | 5 lh 217 gpm | held |
| 10:44 – 11:33 | 11:33 – 14:33 | 8 lh 449 gpm | 12 lh 216 gpm | lost |
| 16:44 – 17:29 | 17:29 – 20:29 | 14 lh 318 gpm | 12 lh 205 gpm | held |
| 20:03 – 20:49 | 20:49 – 23:49 | 18 lh 757 gpm | 1 lh 285 gpm | held |
| 21:47 – 22:26 | 22:26 – 25:26 | 6 lh 726 gpm | 4 lh 257 gpm | held |
| 23:43 – 24:37 | 24:37 – 27:37 | 23 lh 587 gpm | 0 lh 159 gpm | held |
4Monkey KingOne trigger you can perceive, one action nobody else can move, one rate a detector can check. Run one at a time.
win
against Dragon Knight at 344 gpm
assumed role: OfflaneWave access is the moment your hero can kill a whole creep wave from outside its reach, without trading health for it. Before that moment every wave costs you time and position; after it, waves become gold at a rate that compounds. The clock below is a proxy for that moment: the first of six items to finish — Manta Style, Aghanim’s Scepter, Mjollnir, Radiance, Battle Fury, Maelstrom. Those six are the ones the corpus raises most often in wave-clearing context, which makes them a measure of what gets said rather than of what clears a wave. A hero whose spells already clear one crosses the threshold without buying anything on the list, and the proxy will miss it.
Manta Style at 21:54 the first of the six to finish. Of 334 public games that bought it on this hero, 54% bought it earlier than you
8:33Power Treads
15:08Yasha
21:54Manta Style
24:18Mithril Hammer
28:39Black King Bar
35:08Aghanim's ScepterEvery item over 1400 gold, stamped with the clock time it finished. The ones ringed in teal are the six above.
The comparison against public games comes from OpenDota’s scenarios endpoint. It carries no rank filter, so it is every bracket rolled together rather than the players above you, and it counts games in seven coarse buckets that stop at minute 30. Read it as roughly where the timing sits, not as a rank.
Each bar is where this game sits among games on this hero in your own bracket, not across every bracket, so it reads straighter than the all-bracket number on the card. The notch is the middle of that bracket. Nothing here is coloured for good or bad; a row marked ↓ is one where the small number is the good end. Lane efficiency 55%, position 3, teamfight participation 80%, 0 buybacks.
Between 1:00 and 6:00, Dragon Knight earned 418 more gold than you — 320 a minute against your 237. This is the window your own earnings fell furthest behind the enemy in your position, which is the only gold comparison in this report that is about you rather than about all ten players.
| Minute | You earned | Dragon Knight earned | Difference |
|---|---|---|---|
| 1:00 – 2:00 | 144 | 248 | -104 |
| 2:00 – 3:00 | 292 | 360 | -68 |
| 3:00 – 4:00 | 196 | 417 | -221 |
| 4:00 – 5:00 | 380 | 410 | -30 |
| 5:00 – 6:00 | 173 | 168 | +5 |
Nothing explains it. You were alive, out of every fight, and your buildings were standing. Nobody stopped you, and the map still paid them more than it paid you. This is the only branch that is entirely yours.
Where the five minutes went, from the replay position track: 267s in bottom lane, your half; 13s in bottom lane, their half; 11s in your jungle, standing still for 66s of it and walking for 233s.
Teal is your team, ember is theirs, and your own name is marked. Places are approximate: the map is cut into bases, lanes, river and the two jungles, and the diagonal decides whose half you were on.
Last hits taken by you on Razor and by Dragon Knight in the 3 minutes after each fight ends. This is loop R1 measured at the moment it usually leaks.
The verdict column reads last hits, not gold, on purpose. Gold in this window includes kills, assists and buildings, so a team that won the fight and kept fighting can show a fat gold rate having touched no wave at all — which is the exact leak this table exists to catch. The gold rate sits beside it because it says whether the map paid you, and the gap between the two numbers is the interesting part.
Dragon Knight is not their carry. It is the enemy OpenDota put in the same position as you (position 3), because the honest comparison is against whoever contests the same ground.
| The fight | Window scored | You, Razor | Dragon Knight | |
|---|---|---|---|---|
| 6:02 – 6:54 | 6:54 – 9:54 | 6 lh 349 gpm | 16 lh 338 gpm | lost |
| 7:46 – 8:20 | 8:20 – 11:20 | 6 lh 471 gpm | 9 lh 262 gpm | lost |
| 10:28 – 11:23 | 11:23 – 14:23 | 13 lh 468 gpm | 14 lh 352 gpm | held |
| 12:45 – 13:36 | 13:36 – 16:36 | 20 lh 632 gpm | 13 lh 299 gpm | held |
| 17:17 – 18:01 | 18:01 – 21:01 | 25 lh 659 gpm | 8 lh 348 gpm | held |
| 20:32 – 21:25 | 21:25 – 24:25 | 35 lh 784 gpm | 13 lh 321 gpm | held |
| 23:13 – 24:00 | 24:00 – 27:00 | 16 lh 492 gpm | 11 lh 438 gpm | held |
| 24:52 – 25:32 | 25:32 – 28:32 | 10 lh 479 gpm | 10 lh 485 gpm | held |
| 28:52 – 29:46 | 29:46 – 32:46 | 11 lh 790 gpm | 7 lh 285 gpm | held |
| 32:54 – 33:38 | 33:38 – 36:38 | 16 lh 824 gpm | 12 lh 469 gpm | held |
| 35:51 – 36:32 | 36:32 – 39:32 | 0 lh 0 gpm | 0 lh 0 gpm | held |
2Oracle
1Skywrath Mage
1SvenOne trigger you can perceive, one action nobody else can move, one rate a detector can check. Run one at a time.
win
against Centaur Warrunner at 476 gpm
assumed role: OfflaneWave access is the moment your hero can kill a whole creep wave from outside its reach, without trading health for it. Before that moment every wave costs you time and position; after it, waves become gold at a rate that compounds. The clock below is a proxy for that moment: the first of six items to finish — Manta Style, Aghanim’s Scepter, Mjollnir, Radiance, Battle Fury, Maelstrom. Those six are the ones the corpus raises most often in wave-clearing context, which makes them a measure of what gets said rather than of what clears a wave. A hero whose spells already clear one crosses the threshold without buying anything on the list, and the proxy will miss it.
Manta Style at 27:33 the first of the six to finish. Of 334 public games that bought it on this hero, 83% bought it earlier than you
11:51Power Treads
21:48Yasha
27:33Manta Style
31:45Mithril Hammer
34:08Black King Bar
37:51Sange
37:51Sange and Yasha
44:34Aghanim's ScepterEvery item over 1400 gold, stamped with the clock time it finished. The ones ringed in teal are the six above.
The comparison against public games comes from OpenDota’s scenarios endpoint. It carries no rank filter, so it is every bracket rolled together rather than the players above you, and it counts games in seven coarse buckets that stop at minute 30. Read it as roughly where the timing sits, not as a rank.
Each bar is where this game sits among games on this hero in your own bracket, not across every bracket, so it reads straighter than the all-bracket number on the card. The notch is the middle of that bracket. Nothing here is coloured for good or bad; a row marked ↓ is one where the small number is the good end. Lane efficiency 39%, position 3, teamfight participation 40%, 0 buybacks.
Between 22:00 and 27:00, Centaur Warrunner earned 1,358 more gold than you — 764 a minute against your 492. This is the window your own earnings fell furthest behind the enemy in your position, which is the only gold comparison in this report that is about you rather than about all ten players.
| Minute | You earned | Centaur Warrunner earned | Difference |
|---|---|---|---|
| 22:00 – 23:00 | 387 | 360 | +27 |
| 23:00 – 24:00 | 729 | 654 | +75 |
| 24:00 – 25:00 | 445 | 655 | -210 |
| 25:00 – 26:00 | 586 | 601 | -15 |
| 26:00 – 27:00 | 317 | 1,552 | -1,235 |
Explained: you were in a fight. 109 of the 300 seconds sat inside a recorded teamfight. Read this against the table above instead: the question is not what you earned during the fight, it is what you took in the minutes after it.
Where the five minutes went, from the replay position track: 46s dead; 101s in your jungle; 41s in top lane, your half; 23s in your base, standing still for 89s of it and walking for 165s.
Teal is your team, ember is theirs, and your own name is marked. Places are approximate: the map is cut into bases, lanes, river and the two jungles, and the diagonal decides whose half you were on.
Last hits taken by you on Razor and by Centaur Warrunner in the 3 minutes after each fight ends. This is loop R1 measured at the moment it usually leaks.
The verdict column reads last hits, not gold, on purpose. Gold in this window includes kills, assists and buildings, so a team that won the fight and kept fighting can show a fat gold rate having touched no wave at all — which is the exact leak this table exists to catch. The gold rate sits beside it because it says whether the map paid you, and the gap between the two numbers is the interesting part.
Centaur Warrunner is not their carry. It is the enemy OpenDota put in the same position as you (position 3), because the honest comparison is against whoever contests the same ground.
| The fight | Window scored | You, Razor | Centaur Warrunner | |
|---|---|---|---|---|
| 1:28 – 2:12 | 2:12 – 5:12 | 7 lh 194 gpm | 14 lh 302 gpm | lost |
| 7:26 – 8:15 | 8:15 – 11:15 | 7 lh 186 gpm | 14 lh 339 gpm | lost |
| 14:31 – 15:17 | 15:17 – 18:17 | 25 lh 416 gpm | 22 lh 389 gpm | held |
| 22:22 – 23:10 | 23:10 – 26:10 | 33 lh 587 gpm | 29 lh 637 gpm | held |
| 25:27 – 26:28 | 26:28 – 29:28 | 22 lh 405 gpm | 15 lh 667 gpm | held |
| 30:44 – 31:44 | 31:44 – 34:44 | 37 lh 969 gpm | 26 lh 643 gpm | held |
| 36:42 – 37:34 | 37:34 – 40:34 | 25 lh 1,240 gpm | 0 lh 279 gpm | held |
| 38:54 – 39:50 | 39:50 – 42:50 | 13 lh 896 gpm | 6 lh 483 gpm | held |
| 40:52 – 41:43 | 41:43 – 44:43 | 12 lh 834 gpm | 7 lh 631 gpm | held |
| 43:47 – 44:27 | 44:27 – 47:27 | 0 lh 0 gpm | 0 lh 0 gpm | held |
4Viper
1Witch Doctor
1Centaur WarrunnerOne trigger you can perceive, one action nobody else can move, one rate a detector can check. Run one at a time.
Read an arrow as "changes in the direction of" for + and "changes against" for −. Every finding above is tagged with the loop it belongs to, so a number always names a mechanism. The drawing holds the four that the detectors read; all six are named underneath.
Thresholds are first guesses. They were tuned to fire on the one known loss and report nothing on the wins, which is a sample of six. Expect to move them once there are twenty games.
Benchmark percentiles cover every bracket rather than yours, so they read harsh. Use them for ordering, not as a target.
Career averages come from games played before a six-year break, on an older map and an older patch. They tell you what the account did, not what the player can do now.
Post-fight conversion counts last hits in a fixed window after each fight. A fight that runs into a Roshan attempt scores as a farming failure when it was not one.
Nothing here reads position yet. Where you stood and how long you were visible are the measurements that would turn the information-leak findings from an inference into a number.