Celeste, eight days on: the nerf did not cost her wins, it cost her players
We promised a two-week review; here is the day-eight checkpoint. Win rate unchanged at every rank, ranked and unranked alike. But Celeste went from 45% to 40% of matches overnight — one player in six dropped her.
On August 25 we wrote that Celeste's nerf had a tipping point at 167 spirit power, and we promised a two-week review. That review is still coming; in the meantime, eight days of matches already say something, and it is not what we expected.
Wins did not move
Take the broadest measure. In ranked, Celeste won 52.5% of her games between patch 44 and patch 45; she has won 52.7% since, over 29,249 games. In unranked, 52.1% before, 51.5% after, over 97,432 games.
Neither gap means anything. Over the same period, heroes the patch did not touch saw their win rate drift by up to 1.2 points in ranked and 1.0 in unranked — that is the background noise, the normal breathing of the numbers from one week to the next. Celeste sits inside it, on both sides.
We looked tier by tier, since that is where the nerf was meant to bite. Among Initiates, 50.0 → 50.2%. Among Phantoms, 55.0 → 54.9%. Among Ascendants, 59.0 → 59.0% on 500 games. By whole band — bottom, middle, top of the ladder: +0.2, +0.2, +0.3 points. Nothing. The nerf did not cost Celeste a single win rate anywhere.
A word on our own automated measure, since we put it live this week on every patch page: at day 2, it showed −0.80 points across all queues, just beyond the noise. By day 8, the gap has closed. Two days of data were too few — which is exactly why the page keeps measuring at day 7 and day 14.
What moved is the number of players
The figure that matters is elsewhere. Before the patch, Celeste appeared in 32.9% of ranked matches; since then, in 28.0%. In unranked, 50.1% before — one game in two — and 45.3% after. In both queues, five points fewer: one Celeste player in six stopped picking her.
What makes that figure solid is not its size: other heroes moved as much over the same period with no patch involved — Warden, Mina and Paradox each gained five to six points of presence. It is the timing. Day by day, Celeste had been climbing without a break since August 8: 37% presence, then 40, 44, 47, up to 48.2% on August 20. On the 22nd, patch day, she stands at 45.3%. On the 23rd, at 40.4%. She has not crossed 40% since.
A curve that rises for twelve days and snaps on the night of the patch is not noise.
Why wins hold when players leave
The two results read together. A nerf that mainly weakens the early game — which is what we computed on August 25 — first discourages those who played Celeste opportunistically, because she was climbing. Those who stay are the ones who played her before the wave, with tier 2 on the ultimate and the right reflexes. The win rate of a hero whose casual players walk away does not fall: it concentrates.
The top of the ladder points the same way. Among Phantoms, Celeste appeared in 36% of matches; she is at 23%. Among Ascendants, from 49% to 30%. Those populations are small and their numbers swing a lot from week to week — we would not publish them on their own. But they fall in the same direction as everything else, and harder.
As for the 167-spirit threshold, the share of Celeste games ending above it had not changed as of August 25: 32.3% before, 32.8% after. Players did not shorten or lengthen their games to reach it. They left, or they stayed.
What we cannot claim
We do not know where those players went. Warden, Mina and Paradox rose at the same time, but nothing says they are the same people, and none of the three was touched by the patch. Nor do we know whether Celeste's presence will recover: eight days is enough to see a drop, not to know whether it holds. And the Ascendant win rate rests on 500 games — it is stable, but it could be so by chance.
The two-week review will tell whether the stability of wins holds, and whether presence settles at 40% or keeps sliding. It will be on the patch page from September 6, measured automatically, and we will write it up right after.
Methodology. Readings taken on August 30, 2026 around 11:00 UTC from the `match_player` table of deadlock-api.com's public SQL endpoint. "Before" window: from patch 44 (August 12, 22:57 UTC) to patch 45 (August 22, 21:40); "after" window: from patch 45 to the reading. Background noise: largest absolute win-rate change among heroes not modified by the patch with at least 2,000 games in each window (37 control heroes). Presence: share of matches Celeste is in, computed by dividing her appearances by total appearances over twelve players; daily curve from August 6 to 29, the 30th being incomplete. Rank tiers from the match's average badge. Share of games above 167 spirit carried over from our August 25 reading.