6 min read
In Iron through Gold, KDA and win rate look obvious but correlate with rank surprisingly weakly. The signals that actually separate stagnating players from improving ones are five quieter numbers. All five are present in the JSON LoL2LLM exports, so an AI can compute them against role expectations in one prompt and tell you which one to fix first.
Total game CS is noisy because game length varies; CS at 10 minutes predicts skill far better. The Gold-range median for laners (top, mid, ADC) is 70–80 by minute 10. Below 50 indicates a persistent issue with either wave management or matchup awareness.
"I'll catch up later" doesn't work. Twenty CS down at 10 minutes is roughly a Long Sword's worth of stats at first item, and that gap widens to a full item by mid-game team fights. Treat early CS as a compounding asset, not a checklist item you tick off when convenient.
For non-support roles, hitting at least 2 control wards a game is the line that splits ranks. Most Silver players buy 0–1. Gold should be 2+; supports should be 5–8.
One control ward eliminates almost all ambush surface area before a drake. 75 gold to delete the "invisible enemies" problem is the highest gold-efficiency item in the game — but only if you actually place it. Carrying one in your inventory is not the same as benefiting from it.
Most low-rank players die first within five minutes. The cause is almost always one of three: pushing without enemy jungle vision, ignoring the opponent's summoner timers, or misjudging an early-game lane bully.
Gold-tier target: push your first death past 8 minutes. Setting only that as a goal across ten games shifts win rate by 5–10% for many players, because an early death is a triple loss — gold to enemy, forced recall, lost minion wave.
Total damage divided by deaths. KDA hinges on kill count; this captures "how much work did you do before dying." For ADC and AP carries:
Under 5,000: you're melting before reaching the fight
5,000–8,000: standard Gold-range
Over 8,000: Plat+ behavior
Only two ways to move this number: do more damage, or die less. When you prompt the AI, asking it to "compute damage-per-death and tell me which side has more room to improve" isolates your real bottleneck.
Not an individual stat, but the team that takes first tower wins about 70% of games on current patches. Support, jungle, and top especially — whether you can rotate to first-tower pressure after your lane resolves is rank-correlated.
First towers usually fall between 9 and 13 minutes. The habit of glancing at the map the moment your wave is shoved, and traveling to whichever turret is shaking, is the single cheapest macro upgrade in low ranks. You can ask the AI "how did I contribute to first tower?" and it'll read turret damage and rotation timing from the JSON.
Climbing by chasing KDA is slow. Climbing by tracking these five — all unflashy, all compounding — is fast. After every game, prompt the AI: "compare my numbers on these five metrics against role expectations and tell me which one has the most room to grow." That output is your practice target for the next session.