Trending Roblox Games

The scoring formula, in full

Nothing here is secret, and publishing it is partly a way of inviting the argument. A game qualifies if it currently holds at least 500 concurrent players and has been tracked for at least seven days. For every qualifying game we compute four figures:

1. The 24-hour change, as a percentage of the game’s own player count, against the newest reading at or before the 24-hour mark. Weight: 0.45. This is the main term, because a day is the shortest period over which the daily cycle roughly cancels itself out.

2. The 7-day change, on the same basis. Weight: 0.30. This is what separates a game that is genuinely climbing from one having a good Tuesday.

3. Acceleration — the 24-hour rate minus one seventh of the 7-day rate. Weight: 0.15. Positive means the game is gaining faster now than it has been on average across the week, which is the difference between a rise that is continuing and one that is running out. This term is the reason a decaying spike drops off the chart quickly rather than coasting on last week’s numbers.

4. Proximity to its highest tracked count — current players as a share of the largest figure we have ever recorded for that game. Weight: 0.10. A game breaking into new territory is a different event from one recovering a third of the way back to where it was. This is deliberately the smallest weight, because it is the term most distorted by a short record.

All four are then converted to z-scores across the qualifying set for that day, and combined at the weights above into a velocity figure. The final score is that velocity multiplied by the base-ten logarithm of the current player count.

Two things we changed about this formula, and why

The shape above came from an earlier design, and two parts of it did not survive contact with the arithmetic.

Every term is standardised, not just the first two. The original converted the two percentage changes to z-scores but left acceleration and peak proximity as raw values. Those live on completely different scales: a z-score is unbounded with a standard deviation of one and routinely reaches plus or minus two, while peak proximity is a ratio between zero and one. Weighted at 0.10, a term that can never exceed one contributes at most 0.10 to a total whose first term swings across roughly 1.8. The weights were therefore describing an influence the terms did not have, and peak proximity was in effect switched off. Standardising all four makes the published weights true.

Only positive momentum is ranked. Multiplying velocity by the log of the player count behaves correctly while velocity is positive — a 20% rise on a million players outranks the same rise on six hundred, which is what anyone would want. Below zero it inverts: a velocity of −2 multiplied by a larger logarithm produces a more negative score, so the largest collapse sorts as the least notable event on the list. Rather than patch the sign, games at or below zero velocity are dropped. A game losing players is not trending, and it has a chart of its own.

There is one further guard. Because z-scores are relative, they will always produce a “top of the list” even on a day when nothing whatsoever happened. So a game must also be up in absolute terms and be beating the median qualifying game before it can appear. On a genuinely flat day, this chart is allowed to be short.

What the score cannot do

It cannot tell a durable rise from an event spike on the day it happens. Both look identical over 24 hours; only the acceleration term, reading across a week, begins to separate them, and it does so after the fact rather than in advance.

It cannot see a game we do not track. Our catalogue covers games holding a meaningful live player base, so a genuine breakout starts registering once it crosses into that set — which means the very earliest phase of a new hit is invisible here, and any site claiming otherwise is guessing.

It cannot distinguish players from attention. Advertised traffic, a viral video and organic word of mouth all deposit real players in a game and all score the same. Over a longer record the decay rates differ sharply, which is the useful part — but that requires the history this dataset is still accumulating.

And it cannot survive a broken pipeline quietly. A missed collection run makes a gradual move look like a step change, and a Roblox-side outage makes the recovery afterwards read as a surge. Failed requests are stored separately rather than as zeroes, which prevents the worst of it, but a suppressed count that Roblox reported in earnest is one we have no way to challenge.

More Roblox charts

All Roblox charts → · Most played right now → · Biggest player gains → · Biggest player drops → · Roblox player count board → · Most stable games → · Most volatile games →

Best Roblox games by genre → · Live Roblox player stats → · The full game database → · Working game codes →