Sample size: do not let a rare high win rate fool you
A high win rate over dozens of games is not the same as the same result over tens of thousands. Sample size, pick rate, and skill bracket help show whether a rare setup is repeatable or simply had a short winning streak.
Key takeaways
- With fewer games, a handful of wins or losses can move the displayed rate sharply.
- All-ranks data reflects everyday play; high-rank data is better for studying execution by experienced players.
- After changing the skill bracket, recheck sample size, pick rate, and builds instead of watching rank alone.
First check how many games produced the number
Sample size is the number of games behind a result. A 55% rate over many thousands is harder for one player or one streak to move than 55% over a few dozen. More games do not guarantee a correct answer, but fewer games make large swings much more likely.
Low sample does not mean fake data
Small samples can reveal new builds, rare champions, and unusual pairings. Treat them as ideas worth testing, not settled defaults. ARAMKit warns below 200 games to signal uncertainty, not because 199 games are invalid and 201 are automatically reliable.
All ranks for normal play, high ranks for practiced execution
All-ranks data covers more common players and team compositions and usually has a larger sample. High-rank data is useful for seeing how experienced players execute a setup. Neither is the “correct” dataset: choose the view that matches the question and your own familiarity.
When high-rank positions jump, check how much the sample shrank
High-rank games are only part of the full dataset, so rare champions, augments, and builds may drop to very few games. A sudden climb can reflect skill, but it can also be normal small-sample movement. Compare win rate, pick rate, warnings, and common builds together.
Winning games are more likely to finish expensive builds
Costly items and complete three-item cores often appear after a team has earned enough gold and survived long enough to buy them. Their samples already favor games that reached those conditions, so the final win rate cannot be credited entirely to the item or augment.
Keep comparison conditions aligned
Compare options within the same champion, skill bracket, augment stage, and purchase slot whenever possible. Mixing a rare high-rank third item with a common all-ranks first item makes it impossible to tell whether the gap comes from players, timing, or the item itself.
A rare setup needs more than one spike
A few new games can move a rare setup several ranking positions. Keep watching whether its win rate stays in a similar range as the sample grows and whether pick rate and common pairings support the same story. One brief jump is a lead, not proof.
Match your decision to the strength of the evidence
A large sample, meaningful pick rate, and metrics pointing in the same direction support a normal default. A plausible setup with a moderate sample is worth trying. A tiny sample with one spectacular column belongs on a test list until more games support it.
Common mistakes
- Treating every low-sample result as false, or treating a rare high win rate as a hidden best answer.
- Assuming high-rank data is always more useful regardless of your own execution.
- Changing brackets and comparing rank without rechecking sample size, pick rate, and builds.
- Combining different augment stages, purchase slots, or brackets just to make the sample look larger.