Behind in top lane:
stop the bleed, let the map carry.
Choose how far behind you are and when. The model forecasts your chance of winning the game, then shows what comparable states look like if the deficit stays flat, keeps growing, or starts closing.
Compare every role’s macro job →Which developments actually changed the path to victory?
Descriptive lift shows the full observed pathway. Adjusted ELR holds concurrent teammate, fight, structure, and objective pathways fixed to isolate the model’s expected-loss reduction.
| What happened? | Descriptive lift | Adjusted ELR | 95% model interval | Overlap states |
|---|---|---|---|---|
| Swing the fightskill differential improves by at least two | +17.2 pp69.9% forecast | +18.4 pp | 17.7–19.0 pp | 1,22964% of cohort |
| Team gold comebackteam gold differential improves by more than 500g | +15.1 pp64.5% forecast | +12.0 pp | 11.4–12.6 pp | 81442% of cohort |
| Teammates create goldnon-top teammate gold differential improves by more than 500g | +13.2 pp63.2% forecast | +6.3 pp | 5.9–6.8 pp | 59731% of cohort |
| Mid creates a leadmid gold differential improves by more than 300g | +8.9 pp55.0% forecast | +2.1 pp | 2.0–2.2 pp | 1,88298% of cohort |
| Bot creates a leadbot-duo gold differential improves by more than 500g | +7.6 pp56.0% forecast | +1.0 pp | 0.9–1.0 pp | 1,75391% of cohort |
| Take turret groundrelative structure count improves by at least one | +7.5 pp72.4% forecast | +10.9 pp | 10.3–11.5 pp | 93048% of cohort |
| Take objective groundrelative neutral-objective count improves by at least one | +6.1 pp55.8% forecast | +7.9 pp | 7.4–8.4 pp | 1,23664% of cohort |
| Beat the lane deficittop gold differential improves by more than 250g | +5.4 pp42.5% forecast | +0.3 pp | 0.3–0.4 pp | 72338% of cohort |
| Cross-map dividendtop does not materially recover while teammates, objectives, structures, or fights improve | +4.4 pp53.4% forecast | +4.5 pp | 3.4–5.6 pp | 603% of cohort |
Adjusted ELR balances the 10:00 state, excludes the studied role from teammate gold, holds other measured pathways fixed, and only publishes well-supported comparisons. Unobserved intent and opportunity can still move the number.
at 3:00
110 supported game states.
from here
95% interval: 30.8%–36.8%
at the last supported minute
This is a descriptive path through comparable observed states.
How eventual-win forecasts change from this state
No absolute catch-up: remain in the selected 250g band.
Move one bracket farther behind every two minutes.
Move one bracket toward even every two minutes.
Every percentage is a forecast of winning the entire match.
What being behind means at each minute
Forecasted probability of eventually winning, evaluated from the complete observed game state inside each top gold bracket.
| Minute | -1,500g | -1,250g | -1,000g | -750g | -500g | -250g |
|---|---|---|---|---|---|---|
| 3:00 | — | — | — | — | 33.7% | 48.0% |
| 4:00 | — | — | — | 37.6% | 41.5% | 47.5% |
| 6:00 | — | — | 41.3% | 37.0% | 43.5% | 50.6% |
| 8:00 | 31.5% | 34.7% | 37.2% | 41.2% | 43.7% | 46.0% |
| 10:00 | 34.1% | 36.4% | 40.0% | 42.4% | 45.5% | 47.8% |
| 12:00 | 30.4% | 38.1% | 40.9% | 40.7% | 47.0% | 50.3% |
| 14:00 | 38.5% | 40.6% | 40.3% | 45.5% | 44.4% | 47.3% |
| 16:00 | 44.8% | 39.2% | 42.1% | 45.5% | 48.0% | 50.9% |
| 18:00 | 33.9% | 44.4% | 45.6% | 42.6% | 50.2% | 47.5% |
| 20:00 | 40.2% | 41.2% | 43.9% | 49.1% | 49.9% | 48.8% |
You do not need to beat your lane opponent to recover the game.
Holding a −500g deficit is not gold recovery. The path to victory can still run through the other four players: teammate gold, a bot- or mid-lane lead, neutral objectives, turrets, or a favorable fight while top stops bleeding.
The pathway table shows the full descriptive route and an adjusted expected-loss reduction. The adjusted number holds teammate gold, the 10:00 state, and overlapping pathways in check so one teleport or dragon is not treated as the whole story.
minute state map
Eventual win probability across supported 250g role-opponent bands.
without catching up
Follow the same absolute deficit band as the game advances.
states required
Cells below the publication threshold are omitted rather than guessed.
A forecast of game state, not a death tax.
At each minute the model predicts the eventual winner from the full observed state: role gold, XP, level, CS, team economy, objectives, structures, side, patch, and matchup. The table averages those forecasts inside 250g bands.
Later “held-line” states only include games that stayed in that band, so the path is descriptive. Use it to see viable trajectories, not as a promise that avoiding one death creates the full forecast change.
3,430 top minute-3 states · top-lane-win-path-v4