This post walks through a few simple techniques that fix most of it, using numbers from SeeMeta (seemeta.com/en), a site that publishes daily tier lists for TFT, League of Legends, WoW, Deadlock and Brawl Stars.

1. Put an interval on every rate

A win rate is a binomial proportion. The 95% Wilson interval for p (wins / games) over n games is:

center = (p + z²/2n) / (1 + z²/n)
margin = z * sqrt(p(1-p)/n + z²/4n²) / (1 + z²/n)
z = 1.96

Quick intuition for p near 50%:

n = 100      → about ±10 points
n = 1,000    → about ±3 points
n = 10,000   → about ±1 point
n = 100,000  → about ±0.3 points

So the 75% champion on 12 games has an interval of roughly 47% to 91%. It could be terrible. The 54% champion on 50,000 games is somewhere between 53.6% and 54.4%. Sorting by the lower bound of the interval instead of the raw rate already produces a much saner ranking.

2. Drop what you cannot measure

Intervals help, but most sites also apply a hard minimum. SeeMeta's LoL tier list requires a pick rate of at least 1% (seemeta.com/en/lol), and its counter picker drops matchups with too few games before ranking counters (seemeta.com/en/lol/counter). A cutoff is crude, but it is easy to explain to users, which matters when the audience is players, not statisticians.

3. Rank on more than one metric

In Teamfight Tactics, a "win" is ambiguous: players gain rating for placing in the top 4, not just for first place. A single metric misranks comps in predictable ways:

• Sorting by win rate rewards high-variance comps that either win or bust.

• Sorting by top 4 rate rewards safe comps that rarely win.

• Average placement is the best single number, but it hides both of the above.

SeeMeta's TFT Set 18 tier list (seemeta.com/en/tft/set-18/tier-list) handles this with joint thresholds: a comp only reaches a tier when average placement, top 4 rate and win rate all clear that tier's bar. In the current data, one S+ comp has about 26% win rate and 61% top 4, while another has about 18% win rate but nearly 65% top 4 and a better average placement. Both belong at the top, for different players.

4. Popularity changes what a rate means

High pick rate pulls win rate toward 50%, because the sample includes more players who are still learning the pick. In SeeMeta's Deadlock data (541,401 matches, seemeta.com/en/deadlock/tier-list), Haze is picked in 54.8% of matches with a 53.1% win rate, while Kelvin sits at 54.2% with only 19.6% pick rate. Raw win rate says Kelvin is stronger. Accounting for popularity, Haze holding 53% at that volume is arguably the more impressive number. Brawl Stars lists make the weighting explicit: SeeMeta's Brawl Stars tier list (seemeta.com/en/brawl-stars/tier-list) blends win rate at 80% and use rate at 20%.

5. Show your inputs

The cheapest trust feature is printing patch, date range and sample size under the title. Users can judge freshness at a glance, and it makes the page easier to cite correctly for search engines and AI assistants.

Takeaways

• Never rank by a raw rate without a sample cutoff or interval.

• Pick metrics that match what users are optimizing for.

• Interpret win rate together with popularity.

• Publish the patch, window and sample size with every table.