How it works
diffed.lol answers one question: did you get diffed? It backs that with counted statistics, not a black-box score. No machine learning, no opaque "OP rating". Everything below is arithmetic you could verify yourself.
1. The win-probability curve
We continuously crawl finished ranked games across every rank and count outcomes. For each game state (minute · team gold lead · objectives), we tally how often teams in that exact spot went on to win. Your match's curve is a lookup against that table. "Teams in your spot at 20:00 won 22% (n=3,400)." A frequency, not a prediction.
2. Tier-relative comparison
Your stats are compared to your own rank band (Iron-Bronze, Silver-Gold, Plat-Emerald, Diamond, Master+), your role, and your champion where the sample allows. A Gold player is measured against Gold players, never against challenger. Built from 470,073 counted games, refreshed continuously.
3. The metrics
Per game we measure CS/min, CS@10, gold/min, vision/min, kill participation, damage share and early deaths, each as a percentile and a gap vs your band's typical value. "The one thing to fix" is your highest-impact gap, with the cost quantified in gold.
4. Honesty
- Every card shows the sample size behind its benchmarks. Thin sample means a lower-confidence badge, or "too close to call".
- One game is noisy. A single match is directional, not your average.
- Blame attaches only to the person who looked themselves up. Teammate facts (KDA) are neutral, never accusations of intent.
- This is descriptive benchmarking, not an alternative ranking or MMR/ELO system.