Celeste, fourteen days on: the wave is gone, and the top of the ladder dropped her
The promised review. Her ranked win rate did not move by a hundredth. But her presence did not hold the 40% mark: it is back to its early-August level, and among Ascendants it has nearly halved.
On August 30, eight days into Celeste's nerf, we left three questions open: would her win rate hold steady, would her presence settle at 40% of matches, and what would the top of the ladder say with more than 500 games? Here are the answers, at fourteen days.
Wins did not move, and this time it is final
In ranked, Celeste won 52.5% of her games between patch 44 and patch 45. Since then, over 54,453 games, she wins 52.5%. The gap is three hundredths of a point. In unranked, 52.1% before, 51.6% after, over 184,779 games — half a point, when heroes the patch did not touch moved by up to 1.8 points over the same period.
By rank band, the same picture: +0.1 point at the bottom of the ladder, +0.2 in the middle, −0.5 at the top, against background noise of 1.3 to 3.2 points depending on the band. Our automated measure, on the patch page, says the same at day 7: −0.34 points across all queues, within the noise. The −0.80 signal at day 2 was indeed an artefact of two days of data, as we suspected.
A ten-line nerf, six outright nerfs, and a win rate that did not blink. To understand it, look elsewhere.
Presence did not settle: it gave everything back
At day 8, Celeste had gone from 45% to 40% of matches overnight. We wondered whether she would hold that level. She did not.
Since August 30, her presence has slipped by a third of a point a day: 39.4% on the 30th, 38.9 on the 31st, 38.5, 38.3, 37.9, and 36.9% on September 4. That figure has a precise meaning: it is where she stood on August 6–8 — 37 to 38% — before the wave that carried her to 48% on August 20. Two weeks after the patch, nothing of that wave remains. Neither the climb nor the 40% plateau.
In ranked alone, the drop is of the same order: from 32.9% to 28.2% of matches. In unranked, from 50.1% to 44.4%.
The top of the ladder dropped her hardest
That is the third answer, and the sharpest. On August 25 we wrote that Celeste was a top-of-the-ladder problem: 55% wins among Phantoms, 58% among Ascendants. The nerf was meant to bite there. It did not lower her wins — it drove her players away, and all the more the higher you go.
Among Initiates, her presence falls from 40.5% to 38.8% of matches. Among Oracles, from 29.6% to 21.0%. Among Phantoms, from 35.9% to 23.0%. Among Ascendants, from 48.5% to 26.9% — nearly halved, now on 955 games against 500 a week ago. The higher the rank, the cleaner the drop.
Ascendants are also the only tier where the win rate moves: 59.0 → 56.5%. On 955 games, in a band where untouched heroes swing by more than three points, that is not a result — but it is the one place where the direction is right, and the one place Valve was aiming at.
What it says about the nerf
Read the two results together. An early-game nerf like this one mostly bothers players who picked Celeste because she was climbing — and they left, bringing her presence back to its pre-fashion level. Those who stay are the ones who played her before, with tier 2 on the ultimate and the reflexes to match: their win rate does not fall. The 167-spirit threshold changed no habits either: 32.6% of ranked games ended above it before the patch, 31.9% after.
Valve got what a nerf rarely gets this fast: less Celeste, without making her a weak hero. The top of the ladder understood it before anyone else.
What we cannot claim
This reading was taken on the morning of September 5, a few hours short of fourteen full days — the "after" window covers 13.4 days, and nothing will move by tonight. We still do not know where the departed players went, nor whether the slide stops at 37% or continues below the early-August level. And the Ascendant −2.4, at 955 games, remains a direction, not a measurement.
The patch page will keep measuring without us; we will not write a day-21 piece. Celeste's next appointment is the late-September update — if Valve touches her again, we now know what to look at first: not her wins, her players.
Methodology. Reading taken on September 5, 2026 at 08:00 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, 13.4 days. Background noise: largest absolute win-rate change among unmodified heroes with at least 2,000 games in each window (37 controls per queue, 32 for the top band). Presence: Celeste's appearances over the number of matches (total appearances divided by twelve); daily curve from August 6 to September 4, the 5th being incomplete. Tiers from the match's average badge. Automated day-7 measure: patch 45 page, all queues, noise ±0.87.