The death review

The four rules

The review is rules, not a model, and the rules are short enough to check against any game.

Traded

A death is traded when the dying player's team gains a kill, an epic monster, a tower, an inhibitor or a plate within 60 seconds. The rest are untraded: the team got nothing for them. A death that bought a Baron is a trade; a death in the river to nobody's benefit is not.

Converted

A takedown is converted when the player's team takes an epic monster or a building within 60 seconds of it. A kill that turned into a tower is worth more than a kill that turned into a recall, and the rule says which was which.

Cost and gain

Every death carries the win chance it took from the team and every takedown the win chance it gave, read from the win-chance model at that moment with the kill's bounty included. The victim's loss and the takers' gain are the same number, seen from each side.

Contested

An epic monster is contested when players from both teams are credited on it. A profile shows how many of a player's contested objectives their team took.

Why these rules

The trade rule follows the idea of a “worthless death” used in professional analysis: a 2025 study of 37,388 professional games found it among the measures that best separated players, ahead of raw deaths. The cost rule follows Maymin's 2020 finding that kills and deaths weighed by the win probability they moved track team results far more closely than a plain kill to death ratio. Both say the same thing from different directions: what a death cost depends on when and why it happened, and a count throws that away.

The minute is a judgement. Shorter, and a death that set up a Baron 70 seconds later reads as wasted; longer, and the next unrelated fight is credited to it. Sixty seconds is where the examples read right most often on our games, and it is stated so that a reader who disagrees can see exactly what they disagree with.

What our games show

Across 2.9k ranked games and 162k deaths, 74.8% of deaths were traded and 44.0% of 391k takedowns were converted. Of 13k contested objectives, 7.5k went to the player's team, which is close to half because every contest has a team on each side. Those rates are the baseline a profile is read against: a player whose deaths are traded 80% of the time is doing better than most.

How a profile is placed

A profile shows four rates: untraded deaths as a share of deaths, win chance lost per 30 minutes, converted takedowns as a share of takedowns, and win chance gained per 30 minutes. Each of a player's games is placed against every game Riftline holds in the same role, as a percentile, and the percentiles are averaged, the same way the Riftline score is built. “Better than 64% of mid laners” means the same thing on every profile. For the two rates where lower is better, the percentile is turned around, so higher is always better on the bar.

The rates are withheld until 10 reviewed games in a role stand behind them. Three games place nobody: one death in the river would move the untraded rate by a third.

Where it is wrong

The trade rule is at team level. If a teammate on the other side of the map takes a tower thirty seconds after a pointless death, that death reads as traded, because the rule cannot tell a trade from a coincidence. The cost of a death comes from a model that cannot see items, positions or respawn timers, so a death at 35 minutes with Baron up costs the same as one with Baron down. Only ranked solo and flex games with a stored timeline are reviewed, and a profile's rates describe those games, not a whole season. When the model is refitted with a change in what it reads, every review is recomputed against the new version, so figures can move a little from one night to the next without any game changing.

Everything above is a rule that can be checked on a game page. Open a game's story, pick a player, and each death is listed with what happened in the minute after it and what the model says it cost.