Roblox has a weekend. Measuring it honestly is the hard part

The most interesting thing in three weeks of Roblox player data is how little of it moved.

Since June 28 we've been banking daily snapshots of live player counts for the games in our directory - our own numbers, pulled from Roblox's public games API, the one data feed on this platform that definitely still works. Eleven snapshots later, across a constant panel of 109 games we caught at both ends of the window (June 28 and July 18, 2026), here's what the ledger says.

The top 50 is a fortress

Rank every game in the panel by live players on June 28. Do it again on July 18. Forty-six of the top fifty are the same games. Three weeks of updates, events, a World Cup happening in the background, whole genres allegedly dying on social media - and the leaderboard shrugged and kept its seats.

The churn happens at the edges, and it's violent when it happens:

Anime Vanguards is the sharpest lesson: down seventy percent from its late-June level. That's not a dying game - it's the standard update cycle we documented in the first ten days of this dataset. A big update spikes the count, the spike decays for weeks, another update spikes it again. If you catch a game mid-decay and call it dead, you'll be wrong by next patch.

The weekend bump: about 8%, measured carefully

Everyone assumes Roblox is busier on weekends. Our snapshots can almost prove it - and the "almost" is worth your attention, because the obvious way to check is wrong.

Our snapshots aren't taken at the same time every day; build times drift across the afternoon. Player counts swing hard by hour of day, so comparing a 9 p.m. Sunday snapshot against a noon Tuesday one measures the clock, not the calendar. So we kept only the six snapshots that landed in the same narrow window (roughly 14:00-16:10 UTC): four weekdays and two weekend days, across a 104-game panel present in all six.

Result: the four weekday snapshots average 3.76 million concurrent players across the panel; the two weekend snapshots average 4.06 million. Call it a +7.9% weekend bump.

Now the honesty. Two weekend data points is a small sample, and the spread overlaps - the busiest weekday snapshot (4.00M, a Monday in late June) actually beat the quieter of the two weekend ones (3.87M). And because our matched weekend snapshots happen to fall later in the window than most of the matched weekdays, a slow calendar drift - summer holidays ramping up, say - could be wearing a weekend costume. The right way to say it: the bump is around 8 percent, probably, with error bars wide enough to fit a small game inside. Ten more Saturdays in the dataset will tighten that number, and we'll be collecting them.

Which games actually have a weekend

Inside that hour-matched panel, the bump isn't evenly spread. Some games barely notice Saturday; some transform:

Fisch going from ~75,000 on a weekday afternoon to ~201,000 on a weekend one is the biggest single swing we measured anywhere in this dataset. Long-session games - fishing, roleplay, racing - seem to be what people save for days they actually have time, though with two weekends of data we'll hold that theory loosely.

And the other end of the list is the data trap this whole note exists to flag. All Star Tower Defense and Anime Vanguards show weekend ratios far below 1, which reads like "weekday games." They aren't. They're games that were declining across the whole window - and since our matched weekday snapshots came early in the window and the weekend ones late, any sinking game automatically looks weekday-loyal. Same numbers, two stories, and only one of them survives contact with the trend line. Any time someone shows you a day-of-week split, ask what direction the underlying series was moving.

What this is good for

If you run or follow a game, the practical readings are: judge a game against its own update cycle, not the calendar week; expect roughly a twelfth more players on weekend afternoons; and treat any single-snapshot comparison - including ours - as a photograph of a moving object. Our sister directory's trending page ranks games by live players every day if you want to watch the churn yourself.

The dataset keeps growing a snapshot at a time, which is the only way a dataset like this can grow. When we have ten weekends banked instead of two, the 8% gets error bars worth printing. Until then it's the best honest number we have - and better than the vibes it replaces.