Method / Lane labels
Lane labels
Every game with a timeline says whether a player won, drew or lost their lane by 14 minutes, in five words: won big, won, even, lost, lost big. The label is a measurement against the same role, not an opinion, and this page says exactly where the lines are drawn.
What is measured
At 14 minutes, the game's timeline records every player's gold, experience and minions killed. For each player, Riftline finds their opposite number, the enemy in the same role, and computes the player's share of the pair's total on those three: gold, experience and CS, equally weighted. A share of 0.50 is dead even; 0.54 is the lane shown as “54 : 46” on a match row. Fourteen minutes because that is when plates fall and the first tower usually follows, so it is the last moment at which the lane is still mostly a lane.
Junglers are paired with the enemy jungler and supports with the enemy support, on the same three numbers. A support's share moves less than a top laner's, because supports share gold and experience with their carry, and that difference is exactly why the label is placed against the same role rather than against a fixed number.
Where the lines are
The question the label answers is not “how big was the lead” but “how unusual was it for this role”. So the distance from even is placed against every measured lane Riftline holds in the same queue and role, as a percentile. The closest 30% of lanes are even. The widest 10% are won big or lost big. Everything between is won or lost. That split, a third even and a tenth big, is the one STRATZ uses for lanes in Dota, where the same question has been asked for years.
On Riftline's games those lines land about 0.02 and 0.09 from even for most roles: a lane at 51 : 49 is even, a lane at 56 : 44 is won, and a lane at 60 : 40 is won big. The exact values are re-measured with the score every night, so they drift as the corpus grows, and a label given last month may sit on the other side of a line today.
When there is no label
A role with fewer than 200 measured lanes in the corpus gets no labels, because a percentile against a handful of games describes the handful. A game without a stored timeline has no 14-minute figures and no label; profile games get their timeline at the nightly run, or when someone opens the game's story. A game that ended before 14 minutes, or where a player had no opposite number in their role, gets none.
On a profile
A profile adds a player's labels up per role: so many won, so many even, so many lost, with the big ones counted separately, over the newest games with a timeline. The bar beside a role is those counts in proportion. It answers the question “does this player usually win lane” with a count rather than a feeling, and with the number of games it was counted over.
Where it is wrong
A lane is not always a lane. Roaming mid laners, lane swaps and a jungler camping one side all show up as a lane won or lost by a player who was not the reason. A support who roamed to win two fights elsewhere reads as behind in their own lane. An early tower dive that cost a death but took a plate reads as even. A player who was counter-picked and survived to 49 : 51 did well, and the label cannot know it. The label is a fact about the numbers at minute 14, and the death review and the win-chance curve are there for the rest of the game.
None of this is hidden in a score. The 14-minute share is printed on every match row beside the label, so a reader who disagrees with a label has the number it came from.