7 min read
This article is currently shown in English. A translation is in progress.
Jungle is the role where it's hardest to know, mid-game, whether your last decision was correct. The feedback comes too late and too quietly. The fix is to build a habit of reviewing the match data afterward and asking whether the choices were actually rational. Here's the framework I use to evaluate jungle games from the JSON LoL2LLM exports.
Full clear into level-3 gank: stable. Maximizes XP and gold; weak early pressure. Default for scaling junglers (Karthus, Graves, Nidalee).
Three-camp into early gank: aggressive. Generates lane pressure but vulnerable to a counter-clearing enemy. Lee Sin, Elise, Zac territory.
Counter-jungle first: high risk, high reward. Steal camps to open an XP lead. Shaco, Kindred, top-tier Ekko games.
Which is correct depends on the matchup of your champion vs theirs. A useful prompt: "given my champion and the enemy jungler, which opening path was the correct one?" — checks whether your draft-time plan even made sense.
Gold-range median for jungle CS at 10 minutes is 40–50. Below 35 means you've burned too much time on ganks or got counter-cleared. Above 55 with zero kill participation means you over-farmed — you weren't affecting any lane.
Share of your team's kills + assists you were involved in. Jungle target is 60–75%. Below 50% you basically weren't on the map.
Caveat: KP scales with total team kills. 5/8 in a low-kill game (62%) and 12/20 in a bloodbath (60%) read the same percentage but mean different things. Always check raw kill count alongside.
The most concrete numeric on jungle "macro work." Targets through minute 15:
Voidgrubs: 3–4 as a team (out of 6)
Drakes: 1–2
Zero grubs and zero drakes means you lost early map control entirely. "But my KP was great" doesn't convert to wins under those conditions. Objectives are to junglers what CS is to laners.
The JSON doesn't carry coordinates, but each death time is recorded. Jumping to those timestamps in the in-client replay is one of the highest-leverage habits in solo queue.
The recurring patterns:
Invading the enemy jungle without knowing where their jungler is
Third gank into a losing lane — same trap, three times
Engaging at drake with a bad teamfight comp because the timer felt urgent
Asking the AI "at each of my death timestamps, what objective or fight was happening?" surfaces situational mismatches.
KDA lies more on jungle than any other role — great KDA losses and bad KDA wins are both common. The check I run on every jungle game:
jungle CS at 10 ≥ 40 AND kill participation ≥ 60% AND drake participation ≥ 50%
Pass all three: your jungle game succeeded, win or lose — the loss was downstream of you. Fail any one: there's room even if KDA looks fine. Hand the AI this rule of thumb explicitly and the analysis becomes a verdict, not a suggestion list.