Among eligible teams after their first Baron capture. 95% interval 79.0%–80.6%.
Objective atlas RANKED TIMELINE STUDY · baron-conversion-v1.2
What happens after a team takes Baron?
Follow the capturing team from its pre-Baron position through the first fight, three-minute structure conversion, five-minute economy, and final result.
Out-of-sample state probability before the capture.
Predictive state change across 8,881 model-eligible captures.
Out-of-sample state probability after the transition.
THE CONVERSION GAP
A clean Baron is a different game from a leaking one.
“Conversion” here means finishing the match with a win, not merely taking a tower.
41.4% of eligible captures are clean conversions and finish with a win 87.7% of the time. When doing Baron, focus first on a deathless exit, then a structure within three minutes while keeping gold movement positive. The remaining 58.6% win 70.8%; their excess win chance above 50% is 44.8% lower. Therefore, when contesting a secured Baron, focus on forcing buff-holder deaths, denying the first structure, and stopping the capturer’s gold state from improving.
of eligible captures · 3,531 cases
87.7% final winof eligible captures · 5,007 cases
70.8% final winReduction in the win rate above 50%, clean group versus remainder.
+16.9 PP absolute gapProtect the exit first: avoid a capturing-team death in the first minute, then turn the buff into at least one structure by three minutes without letting the gold state stall.
The observed pressure points are the inverse: force buff-holder deaths, deny the first three-minute structure, and flatten or reverse the capturing team’s gold movement. These are associations, not proof that any one intervention causes the result.
STARTING POSITION
Baron matters differently depending on the state you bring into the pit.
Starting stateNPre-statePost-stateMovementFinal win
Out-of-sample, match-grouped model with isotonic calibration. These are predictive state movements, not randomized treatment effects.
CONVERSION CLOCK
Value either becomes permanent or leaks away.
Economy uses complete minute snapshots. Kills, structures, inhibitors, and subsequent objectives use exact event timestamps.
mean gold-difference change
- Structure
- 32.2%
- Inhibitor
- 5.4%
- Next objective
- 7.8%
- Game ended
- 4.4%
- Capturer deaths
- 0.9
- Opponent deaths
- 1.4
mean gold-difference change
- Structure
- 82.7%
- Inhibitor
- 43.8%
- Next objective
- 28.5%
- Game ended
- 34.5%
- Capturer deaths
- 2.66
- Opponent deaths
- 4.32
mean gold-difference change
- Structure
- 88.6%
- Inhibitor
- 57.4%
- Next objective
- 41.3%
- Game ended
- 53.5%
- Capturer deaths
- 3.87
- Opponent deaths
- 6.1
median · 8,854 observed
IQR 0.8–1.8mmedian · 7,225 observed
IQR 1.5–4.6mmedian · 4,948 observed
IQR 1.4–4.3mmedian · 9,628 observed
IQR 2.4–8.3m CONVERSION FAILURES
How a secured Baron falls apart, and what it usually costs.
These are overlapping observed patterns, not mutually exclusive causes. “Clean conversion” is the comparison case; every other row describes a specific way value leaked after capture.
Observed patternFrequencyFinal winvs clean3m gold Δ3m inhibitorEnded by 3m
Clean Conversion
No capturing-team death within 60 seconds, at least one structure by three minutes, and either a positive gold-difference change or a game end.
Frequency41.4%3,531 cases
Final win87.7%
Versus cleanReference
3m gold change+3,974g
Inhibitor by 3m59.7%
Game ended by 3m39.5%
Lost Exit Fight
The capturing team suffers more deaths than its opponent within 60 seconds.
Frequency21.8%1,862 cases
Final win57.8%
Versus clean-29.9 PP
3m gold change+424g
Inhibitor by 3m19.4%
Game ended by 3m14.4%
No Structure By Three Minutes
No tower, inhibitor, or Nexus structure is taken within three minutes.
Frequency17.5%1,490 cases
Final win55.8%
Versus clean-32 PP
3m gold change+46g
Inhibitor by 3m0.0%
Game ended by 3m25.8%
Economy Flat Or Worse
The capturing team's gold difference is unchanged or worse at the three-minute snapshot.
Frequency14.6%1,247 cases
Final win51.6%
Versus clean-36.1 PP
3m gold change-1,610g
Inhibitor by 3m13.2%
Game ended by 3m0.0%
Compound Collapse
The team loses the 60-second exit fight, takes no structure, and has an unchanged or worse gold difference after three minutes.
Frequency4.3%366 cases
Final win42.9%
Versus clean-44.8 PP
3m gold change-2,128g
Inhibitor by 3m0.0%
Game ended by 3m0.0%
Frequency uses captures with complete 60-second and three-minute event windows. Gold consequences use the first complete snapshot after three minutes and report coverage in the publication.
THE EXIT FIGHT
The first deaths after capture predict whether Baron converts or reverses.
Neither team records a death during the selected window.
80.6%final win2,040 observed cases
- 3m gold change
- +2,815g
- Structure by 3m
- 84.4%
- Inhibitor by 3m
- 45.0%
- Game ended by 3m
- 33.0%
The capturing team kills at least one opponent without losing anyone.
88.6%final win2,437 observed cases
- 3m gold change
- +3,730g
- Structure by 3m
- 90.7%
- Inhibitor by 3m
- 56.7%
- Game ended by 3m
- 43.2%
Both teams lose players, but the capturing team kills more opponents than it loses.
86.6%final win1,648 observed cases
- 3m gold change
- +3,152g
- Structure by 3m
- 89.8%
- Inhibitor by 3m
- 49.7%
- Game ended by 3m
- 31.8%
Both teams record the same non-zero number of deaths.
76.3%final win960 observed cases
- 3m gold change
- +2,026g
- Structure by 3m
- 81.4%
- Inhibitor by 3m
- 35.4%
- Game ended by 3m
- 18.9%
The capturing team suffers one more death than the opponent.
68.0%final win1,106 observed cases
- 3m gold change
- +1,304g
- Structure by 3m
- 73.5%
- Inhibitor by 3m
- 25.7%
- Game ended by 3m
- 15.4%
The capturing team suffers at least two more deaths than the opponent.
46.2%final win842 observed cases
- 3m gold change
- -676g
- Structure by 3m
- 54.3%
- Inhibitor by 3m
- 11.3%
- Game ended by 3m
- 13.2%
STEAL, CONTEST, OR CONCEDE?
The denominator matters more than the highlight.
A successful steal is observable in a completed capture. A steal success rate requires every failed attempt too. This corpus does not yet identify those opportunities reliably, so no strategy receives a made-up percentage.
All qualified five-player or partial-team contests, whether they secure Baron or fail.
Best evaluated against denial, fight, and eventual game-state outcomes—not Baron ownership alone.Every attempt where exactly one champion enters contest range while teammates remain disengaged, including misses and deaths.
Must count the solo entries that miss, die, or force a disengage, not just the steals seen in the capture log.Every one-person attempt plus a predeclared post-attempt death or escape window, not successful steals alone.
A subset of one-person attempts; it also needs an explicit survival window and the value of the life traded.Every start where the first team disengages and the opposing team either captures or fails to finish.
Requires health-path and disengagement detection to distinguish a takeover from a fresh start.All Baron opportunities where the defending team concedes, including the cross-map value actually gained.
Must value towers, inhibitors, dragons, waves, and gold gained while Baron is conceded.What can be compared now
The post-capture branches above show what happens after ownership is known. They cannot answer whether sending one player was better than contesting as a team, because failed attempts and the alternative actions are missing. That comparison will publish only after the AI’s attempt detection passes the page’s stated precision, recall, labeling, and opportunity gates.
STATE DEPENDENCE
The same capture lands in different games.
StateNFinal win3m gold Δ3m structure
COMPOSITION
Which roster profiles convert pressure?
Official champion tags are patch-versioned. Adjusted lift compares each profile with the expected result for its capture minute, pre-Baron gold band, dragon score, and patch.
Official tag profileNFinal winAdjusted lift3m gold Δ3m structure
Data Dragon content hash · 66db52223a150af1. Strategic archetypes are withheld until their champion assignments have a reviewed, versioned rulebook.
NOTES
How the numbers are built
Definitions
Capture. A team's first observed BARON_NASHOR kill in the official Riot timeline.
Baseline. Last complete team snapshot at or before capture, no more than 60 seconds old.
Horizon. First complete snapshot at or after the target, no more than 60 seconds late.
Events. Kills, structures, and objectives use their exact observed timestamps.
Interpretation. Observed and predictive association; not an isolated causal effect.
CompositionAdjustment. Direct standardization by capture-minute band, pre-capture gold band, dragon score, and patch; multi-label official tags.
Conversion-gap calculation
Clean conversion requires no capturing-team death within 60 seconds, at least one structure by three minutes, and either positive three-minute gold movement or a game end. The remainder is every other eligible capture. “Excess win chance” is final win rate minus 50%; the displayed reduction is (clean excess − remainder excess) ÷ clean excess.
Population
Queue 420; patches 15.11, 15.14, 15.15, 15.16, 15.17, 15.18, 15.20, 15.23, 16.1, 16.10, 16.11, 16.12, 16.13, 16.14, 16.15, 16.16, 16.17, 16.2, 16.5, 16.6, 16.7, 16.8, 16.9; regions EUW1, KR, NA1.
Coverage
Subgroups below 100 observations are hidden. Exact death-event coverage is 100.0%.
What is not estimated
A numerical “start Baron versus do not start,” organized-contest, one-person-steal, or suicide-steal effect is not published until the AI’s replay-level detection is solid. Completed captures alone cannot supply the failed-attempt denominator.