The previous champion-map article used an X axis called "snowball" — a weighted average of largest killing spree and multi-kill. We buried a caveat back then: "snowball measures kill bursts, it does not directly mean ending games early." This time we put that sentence on trial, and the caveat turned out to be the whole story.
First, define win tempo:
Win tempo = median duration of that champion's WINS − global median win (17.6 min)
Negative = wins faster than average. If sprees really rolled into early wins, the spree axis and win tempo should correlate strongly. Measured: Spearman rho = −0.03 (n=173) — no relationship at all. The spree axis measures personal kill highlights; closing a game is a team property. The data says these are two independent skills.
Who actually closes games fast
−52s
−40s
−38s
−34s
−33sThe fastest closer is Yorick — spree rank 147. He pushes towers, not kill feeds, and his wins land almost a full minute under the global median. In ARAM, snowballing waves and towers closes games faster than snowballing kills. Jinx is the one big spree champ on the list (rank 20): she is the exception that holds both ends, not the rule.
Spree kings who win slow
+33s
+23s
+23s
+20s
+16sAssassins and cleanup carries farm huge sprees, but the sprees do not convert into tempo — their wins actually run longer than average. The extreme case is Shaco: top-quarter sprees, yet the 2nd-slowest closer of all champions (+54s) — and he is about to show up again on the late-game board below.
Zero sprees, fastest closers: the supports
−30s
−28s
−24s
−15s
−12sThis row is the key evidence: Taric, Sona and Ivern all sit in the bottom 12 of the spree board, yet their wins close faster than average — because a team fed by its enchanter closes fast while the kills land on someone else's scoreboard. Snowballing is a team property; the spree board only credits individuals.
So who are the real early and late champions?
The cleaner measure pulls the two ends apart directly: the left number is the win rate in games that end by 16 minutes, the right is in games past 22 minutes (the middle is dropped on purpose).
+16.1pp
+13.0pp
+12.8pp
−13.3pp
−12.4pp
−11.7ppOrnn is the biggest climber in the game: his Masterwork upgrades hand teammates a thousand gold of free item stats, so he loses short games hard (41%) and wins long ones (57.1%) — and he is also the 4th-slowest closer (+48s). A classic "wins by dragging it out" profile. The big Summoner's Rift dataset (8M games, Diamond+) also flags Ornn as one of the strongest even-state climbers — two completely different datasets landing on the same champion. Seraphine is the biggest early-game monster: she wins nearly two thirds (63.4%) of games that end by 16 minutes, then falls back to a coin flip past 22. And Shaco and Twitch reappear here — big sprees, slow wins, stronger the later it gets. Three signals, one conclusion: their sprees are farmed in long games, not rolled into fast ones.
One honest caveat
The win-tempo effect is small: every champion's median win sits within about ±1 minute of the global median — game length is decided by ten players, so any one champion's influence is diluted by construction. It is also descriptive: a long game is not necessarily one he dragged out; even matchups simply run longer. And the early/late tilt shifts with the patch window: Kassadin read as nearly flat in the previous window and −3.3pp mildly early in this one — the axis is very stable across patches overall (r≈0.86–0.93), but read individual mid-game-spike champions with care.
Data: matches collected by arammeta · queue 2400 (Mayhem) · patches 16.11–16.13 · 951,931 games (collector writes while we scan; the two passes differ by 38 games) · ≥500 wins and ≥500 losses per champion · win tempo = median win duration − global median win 17.6 min · spree axis vs win tempo Spearman rho = −0.03 · early/late = WR(≥22 min) − WR(≤16 min).