What is the sample size?

267 ranked League of Legends games across 5 female players over 5 months.

What winrate delta was observed?

A mean delta of 5.9 winrate points in favour of off-period phases, consistent across 100% of the players studied.

Is the result statistically significant?

The paired t-test at the player level (n=3) gives p ≈ 0.048, just under the 0.05 threshold. The two-proportion z-test on games (n=267) gives p ≈ 0.22, not yet statistically conclusive.

Is this medical advice?

No. This is an independent, non-peer-reviewed pilot study. Correlations are suggestive, not proven science.

The pilot study

In early 2026, LoL coach Tony "Saskio" Chau went viral for tracking his duo partners' menstrual cycles over 5 months and correlating cycle phase with LoL performance data (winrate (WR), KDA, damage, deaths) across 345 ranked games and 5 female players. Using a custom Riot API pipeline, he found measurable correlations between cycle phase and in-game performance. He publicly shared detailed data for 3 of the 5 players; the table below covers that subset only.

PLAYER	ROLE / RANK	GAMES	OFF-PERIOD WR	ON-PERIOD WR	DELTA
Yuulu	Support · Master	147	57.5%	52%	−5.5pp
Sweatylilkoi	APC Supp · Master	68	61.2%	58.8%	−2.4pp
Asakifox	Mid (Ahri) · D4	52	59.6%	48.1%	−11.5pp

Combined (3 players, 267 games): weighted off-period WR 58.9% vs on-period 53.0%, a −5.9% WR delta. Across the full 5-player roster, the mean per-player WR delta is −9.58 points (standard deviation 2.41).

Overall duo WR went from ~52% pre-tracking to 55.1% after adapting champion picks to phase, up +3.1% WR over 4 months and 345 games.

Aggression signal (Yuulu): on-period damage/death score averages 8.6/10 vs 4.85/10 off-period. Saskio switches Vayne → Tristana to match the aggression level instead of fighting it.

Mean cycle length across roster: 28.04 days, standard deviation 0.31. Tight enough that next-cycle start is predictable to within hours.

Academic follow-up: QUT researcher Riley Dunn is running a formal study on cycle performance in female esports players (MOBA & FPS). Saskio is in discussionsto contribute his dataset. We're building the tool to make that kind of data collection possible at scale.

Statistical analysis

Paired t-test

Unit: players (n=3)

t = 2.42, df = 2

p ≈ 0.048

4.8% chance the effect is random. Crosses the p<0.05 threshold, but with only 3 players, one outlier could drive the result.

Two-proportion z

Unit: games (n=267)

z = 0.76

p ≈ 0.22

22% chance the effect is random. Not enough data to affirm a correlation statistically.

Effect direction is consistent across 100% of players, but detecting a 6% WR shift reliably requires ~800+ on-period games. p=0.22 isn't "no effect", it's "not enough data yet." Sample is small, self-selected, not peer-reviewed. Not medical advice. That's exactly what we're building this for.

← Back home