Rarely in the first weeks after launch. Only 2 of 14 Steam games with 200+ reviews that opened September 2026 with a Mixed label on their English reviews showed Mostly Positive by 1 November. Both had started within 0.1 points of the 70% line. For those 14, reviews written after day 14 ran a median 1.0 point more positive than launch reviews, against 0.4 points less for all 95.
What happened to the 14 launches that opened Mixed?
We took the 95 September 2026 launches with at least 200 reviews from the October edition of the Steam launches in numbers index, which lists every launch with its first-14-day review rates and is free to download as CSV. For each game we asked Steam’s public review endpoint for two windows of English reviews from Steam purchasers. Window one is days 0 to 14. Window two runs from day 15 to day 60, or to 1 November for games released later in the month.
14 of the 95 had the Mixed label after window one, and none started lower than Mixed. The full list, game by game, is further down.
Nine of the 14 had a better second window than first. The median gain was 1.0 point, with the middle half of games between -4.0 and +3.4. Only 2 rose by more than chance would explain, and 2 fell by more. The store label moved for the two games that were already sitting on the line. The all-language label that store pages show by default gives the same count with different games: 14 were Mixed at day 14, and 2 (Moonlighter 2: The Endless Vault and Train Sim World 7) had moved up to Mostly Positive by 1 November. One game went the other way on that label: WARDOGS showed Mostly Positive on day 14 and Mixed by 1 November, after its later English reviews fell from 79.8% to 63.0% positive.
Why does a better second month hardly move the label?
Mostly because of volume. Across the 95 games, days 0 to 14 produced 65.8% of all English reviews they had collected by 1 November (184,386 of 280,190). For the median Mixed starter, the later window added 1 review for every 4 written at launch.
Anymaker shows what that means. Its later reviews were 69.8% positive, 21.5 points above launch, the largest gain in the sample. Its cumulative English score on 1 November was still 52.7%, because 53 new reviews cannot outweigh 211 old ones. Steam does show a separate score for the past 30 days at the top of the store page (Steam review score), and that is where a turnaround like this becomes visible first.
The practical reading for a studio is that the lifetime label is mostly fixed by the build that players meet in the first two weeks. That is the build our Launch Health Check is meant for: a release candidate checked at a fixed $699 by up to two testers over about 30 hours, with a Team Lead and a report within 3 business days. If the game is already out on Mixed, every fix lands in front of the players writing the 30-day score. A patch that breaks saves made on the live build costs more reviews than it wins back, which is why the patch day QA checklist checks old saves in the new build before anything goes to the default branch.
How many new reviews does it take to climb out of Mixed?
Valve does not publish the label bands. In every pull for this article, though, the switch between Mixed and Mostly Positive fell at 70%: the highest share labelled Mixed was 69.997%, the lowest labelled Mostly Positive was 70.3%. With that line, the climb is plain arithmetic. New reviews needed = launch reviews × (0.70 - launch share) ÷ (new share - 0.70).
This table assumes a game launched at 60% positive. It is arithmetic, not a forecast.
| Reviews at launch (60% positive) | New reviews needed at 75% positive | At 80% positive | At 90% positive |
|---|---|---|---|
| 300 | 600 | 300 | 150 |
| 1,000 | 2,000 | 1,000 | 500 |
| 3,000 | 6,000 | 3,000 | 1,500 |
Put next to the measured pace, the gap is wide. A 300-review launch that keeps the median Mixed starter’s rate would add about 80 reviews in the following weeks. Even at 90% positive, that is half of what the climb needs. The comebacks that make the news run on a different clock. No Man’s Sky took two years to get from Overwhelmingly Negative to Mixed and three more to reach Mostly Positive, as PCGamesN summarised Sean Murray’s account in September 2021.
Do games that start well keep their score?
Mostly yes, and they drift down slightly more often than up. Across all 95 games, the later window was a median 0.4 points less positive than launch, with the middle half between -3.8 and +2.0. 19 games fell by more than chance would explain and 9 rose. For the 62 games that opened with Very Positive or Overwhelmingly Positive, the median change was -0.6 points, with 12 clear drops and 4 clear rises.
The English labels barely moved. Of the 81 games that started at Mostly Positive or better, 2 slipped from Very Positive to Mostly Positive, 1 rose from Mostly Positive to Very Positive and 4 rose from Very Positive to Overwhelmingly Positive. The rest kept the label they had on day 14. A likely reason for the small downward drift is who reviews first: players who bought on day one are often the ones already waiting for the game. That is a hypothesis; this data cannot test it.
So the 1.0-point median gain among Mixed starters should be read against a baseline that runs slightly negative. Part of it may be a pull toward the middle, since a game measured at a low point has more room to look better next time by chance alone.
Did Early Access or patching change the later reviews?
Early Access games moved down more. The 17 Early Access games had a median change of -4.9 points from launch to the later window, against -0.4 for the 78 full releases. A permutation test of the two medians gives p = 0.003, so the gap is unlikely to be chance in this month. It is still 17 games from one month. Among the Mixed starters only 3 were in Early Access, too few to say anything about them as a group.
Patching gives no signal here. Using the patch-type announcements counted in the index for each game’s first 30 days, games with 3 or more had a median change of -0.4 points (72 games) and games with fewer had -0.6 (23 games), p = 0.89. Among the 14 Mixed starters, 13 had 3 or more announcements, so the one game with fewer leaves nothing to compare. Patch counts and early reviews are covered separately in Steam game patches vs reviews, which looks at the first 14 days rather than at what comes after.
Which of the 14 moved, game by game?
All 14 Mixed starters, ordered by their launch share, with the later window next to it.
| Game (release) | Days 0-14: reviews, positive | Later window: days covered, reviews, positive | Change | Outside noise? | English label on 1 Nov |
|---|---|---|---|---|---|
| Anymaker (18 Sep, EA) | 211, 48.3% | 29.5 days, 53, 69.8% | +21.5 | yes, up | Mixed |
| EA SPORTS FC 27 (24 Sep) | 2,020, 48.4% | 23.5 days, 1,624, 45.8% | -2.6 | no | Mixed |
| Backrooms: Code Yellow (25 Sep, EA) | 391, 51.4% | 22.5 days, 48, 54.2% | +2.8 | no | Mixed |
| SpeedRunners 2: King of Speed (3 Sep) | 311, 54.3% | 44.5 days, 39, 56.4% | +2.1 | no | Mixed |
| NBA 2K27 (3 Sep) | 650, 55.5% | 44.5 days, 795, 50.4% | -5.1 | borderline | Mixed |
| BLACKWOOD (16 Sep) | 942, 55.6% | 31.5 days, 119, 59.7% | +4.1 | no | Mixed |
| Crimson Moon (1 Sep) | 2,125, 60.2% | 46 days, 130, 60.8% | +0.6 | no | Mixed |
| Halloween: The Game (8 Sep) | 5,951, 65.2% | 39.5 days, 2,008, 61.4% | -3.8 | yes, down | Mixed |
| Train Sim World 7 (15 Sep) | 247, 65.6% | 32.5 days, 141, 68.8% | +3.2 | no | Mixed |
| Minecraft Dungeons II (29 Sep) | 5,964, 67.7% | 18.5 days, 534, 63.1% | -4.6 | yes, down | Mixed |
| The Guild 1 Remake: Europa 1410 (17 Sep, EA) | 322, 68.3% | 30.5 days, 45, 62.2% | -6.1 | no | Mixed |
| The Crust (10 Sep) | 258, 69.4% | 37.5 days, 155, 69.7% | +0.3 | no | Mixed |
| Police Chief Simulator (17 Sep) | 326, 69.9% | 30.5 days, 115, 71.3% | +1.4 | no | Mostly Positive |
| SILENT HILL: Townfall (23 Sep) | 2,913, 70.0% | 24.5 days, 848, 76.1% | +6.1 | yes, up | Mostly Positive |
“Change” is the later window’s positive share minus the launch share, in percentage points. “Outside noise” means a two-proportion z-test gave |z| of 1.96 or more; NBA 2K27 sat at z = -1.93. SILENT HILL: Townfall shows 70.0% after rounding, but its 2,039 positive reviews out of 2,913 are 69.997%, and Steam labelled that Mixed.
What can this data not tell you?
Five choices in the method shape the result, and each one points to a way the numbers could change.
| Choice in the method | What it means for the result |
|---|---|
| Later window of 17.5 to 46 days (median 32.5) | Only games released on 1 to 3 September got close to day 60. A rerun of the same script on 1 December 2026 gives every game the full 60 days and may move a few labels. |
| English reviews only | The index is built on English reviews. On the all-language label that store pages show by default, 2 of 14 also moved up, but they were different games. |
| Steam purchasers only | Reviews from key activations do not count toward the store score, so they are not counted here either. |
| 14 Mixed starters, 17 Early Access games | Small groups. The z-test per game and the permutation test per group sort signal from noise, yet several tests on one month will throw up a false positive now and then. |
| Patch count from announcements | It counts Steam announcements with words like patch, fix or update near the top. Silent builds are invisible to it, and it says nothing about how big a fix was. |
The 95 games and their launch-window rates are in the CSV on Steam launches in numbers, and the later windows can be pulled again from the same public endpoint with the dates given above. For the October picture of launch bugs that feeds these numbers, see the October 2026 Steam Launch Bug Index, and for the first patch timing, the day-one patch data.