Do Iron and Diamond players show the same mastery curves? We analyzed more than 7.8 million data points across 1.2 million matches, from Iron through Master+.
Dataset
Two data sources were combined: the new per-tier analysis (Iron through Diamond, from mastery-enriched match blobs) and the original Master+ dataset (530K matches from the first crawl era). This gives us complete coverage across all 8 rank brackets.
| Tier | Matches | Median Cap | Mean Cost |
|---|---|---|---|
| Iron | 66,893 | 6 | -10.5% |
| Bronze | 46,816 | 6 | -10.7% |
| Silver | 57,697 | 6 | -10.0% |
| Gold | 285,392 | 6 | -8.5% |
| Platinum | 141,506 | 6 | -8.7% |
| Emerald | 82,456 | 6 | -9.6% |
| Diamond | 14,032 | 6 | -16.6% |
| Master+ | 530,540 | 6 | -12.3% |
Methodology
Mastery curve: For each (champion, tier) pair, we compute win rate at each mastery level (1 through 40+), then aggregate into 2-level buckets (1-2, 3-4, 5-6, etc.) for statistical power. Monotonic smoothing ensures the curve never dips.
Mastery cap (Spearman): We use Spearman rank correlation on a sliding window to find the mastery level where win rate stops correlating with experience. For each candidate cutoff, we test whether the remaining data points above that cutoff still show a significant positive trend (p < 0.05, rho > 0.1). The cap is where the correlation disappears.
Cost of learning: Measures the win rate penalty for playing a champion below your mastery cap. Computed as (low_mastery_WR - plateau_WR) x 100. A cost of -8% means first-timers lose 8 percentage points of win rate compared to experienced players on the same champion.
Sample size thresholds: Based on binomial CI width analysis, we require 267+ games per mastery bucket for +/-3pp precision (95% CI), and 385+ games per group to detect a 5pp WR difference with 80% power. Champions with fewer than 200 total games in a tier are excluded.
Finding 1: Mastery Cap Compresses at High Elo
The median mastery cap is 6 across all tiers, but the variance decreases dramatically at higher elo. In Iron, some champions show improvement up to mastery 26. In Diamond and Master+, virtually every champion plateaus by mastery 6.
In this dataset, high-rank mastery curves reached their estimated plateau earlier. The difference was statistically significant: a Mann-Whitney U test comparing Low (Iron/Bronze/Silver) with High (Diamond/Master+) mastery caps gave p = 0.0005.
However, this is driven by ~35 complex champions. 80% of champions plateau at mastery 6 regardless of rank.
Finding 2: Cost of Learning Is Stable Across Ranks
The median cost of learning stayed near -8 percentage points across tiers. In this measure, first-timing carried a similar win-rate gap in Iron and Diamond.
Mann-Whitney U tests show no significant difference between tier groups (p = 0.06 to 0.18). The Spearman correlation between tier rank and cost is weak (rho = 0.11), though technically significant (p = 0.0001) due to the large sample size.
Champions With the Biggest Rank-Dependent Mastery Curves
Some champions showed very different estimated caps across ranks. Viego reached level 26 in Iron and level 6 in Diamond, a 20-level gap. General game knowledge may explain part of that compression, but this analysis did not isolate the cause.
Finding 3: Assassins and Fighters Have the Highest Skill Ceilings
Champion class is a statistically significant predictor of mastery cap (Kruskal-Wallis p = 0.01). Assassins (31% with cap above 8) and Fighters (26%) take the longest to master, while Tanks plateau almost immediately (4%).
The measured cost of learning was similar across classes, from -7.5 to -9.9 percentage points. The clearer difference was how long the mastery curve continued before reaching its estimated plateau.
Role (top, mid, jungle, etc.) is NOT a significant predictor (p = 0.39). Mid lane appears highest only because it has the most assassins and mages.
Mastery Cap by Champion Class (Gold)
| Class | Champions | Median Cap | % Cap > 8 | Median Cost |
|---|---|---|---|---|
| Assassin | 16 | 6 | 31% | -9.9% |
| Fighter | 50 | 6 | 26% | -8.7% |
| Marksman | 28 | 6 | 25% | -7.8% |
| Mage | 36 | 6 | 19% | -7.5% |
| Support | 18 | 6 | 11% | -7.6% |
| Tank | 24 | 6 | 4% | -8.7% |
One-Trick Potential: Champions That Keep Rewarding Mastery
For this analysis, an OTP signal required both a high mastery cap—improvement beyond level 10—and a win-rate gain above 10 percentage points across the curve. Highlighted cells show where both conditions were met in each tier.
Each cell shows mastery cap / WR gain from untrained to plateau. Highlighted cells indicate the OTP signal (cap > 10 and gain > 10pp).
Data Quality and Limitations
Bootstrap stability: The Spearman mastery-cap estimate is noisy. Even popular champions with more than 50,000 games in one tier produced wide 95% bootstrap intervals in some cases, such as [6, 20]. Treat individual cap values as estimates, not exact breakpoints.
Diamond data: With 140K total games and 50 champions above 1K games, Diamond data supports cost-of-learning analysis but has limited power for detecting non-trivial mastery caps.
Master+ data: Sourced from the original 530K-match analysis (predominantly Master/GM/Challenger), providing rich per-level data with up to 150 mastery levels per champion.
Win rate sanity: Overall win rates across all tiers are 49.6-49.9%, confirming internal data consistency.