The Worlds Report - Issue #8
- Roblox's -4.18% week is a measurement event, not a demand event. Murder Mystery 2 alone accounts for -146.9M of the -164.2M decline, or 89.5% of it. Strip that one title out and the like-for-like comparison is -0.51%. Demand did not contract this week; our measurement of it did.
- The proof is in the shape of the day, not the size of the drop. Murder Mystery 2's overnight floor fell from 1,013,583 to 244,048 in a single four-hour interval and stayed down, while its afternoon peak only drifted from 1,062,508 to 814,048 across six days. Bots are the population that never sleeps. Roblox's own August 19 enforcement post warns that visit and player data in Creator Hub Analytics may shift for affected periods.
- Simulator's +23.4% is the same defect running the other way. Excluding Steal An Egg, created July 25 and the largest single-title move on any platform this week at +243.5M, the Simulator line is -0.1%. One month-old title is the entire move.
- Fortnite's GTA-6 theme halved before its own reveal, from 14.3% to 7.5% of CCU and from four islands in the top-15 down to two, with the August 27 showcase still three days ahead. The season took its place: v42.00 shipped August 20 and PvP went from 45.8% to 51.4%.
- Scorecard: 4 hit, 1 partial, 1 miss, 1 carried. Running 43 hit, 11 partial, 6 miss. Both breaks trace to the same August 19 enforcement action, and neither was a bad read of demand.
Running record: 43 hits · 11 partials · 6 misses
Last week's six calls. One grades later against the event it names and is carried, not graded. Of the rest, four hit, one partial, one miss. Both breaks landed on metrics that turned out to be measuring platform enforcement rather than player behaviour, which is a failure mode we have not had before and which we are naming rather than absorbing.
- ◐ PARTIAL
Do not price the Roblox discovery change into a build decision until it shows up in two consecutive weeks. Like-for-like does not fall below -2% and Simulator/Tycoon holds at least 24%. MED-HIGH
The demand half failed: like-for-like came in at -4.18%, below the -2% floor. The share half held comfortably at 34.0%. Condition not met. The driver is the August 19 enforcement action: Murder Mystery 2 is 89.5% of the entire decline, and excluding it the comparison is -0.51%. We grade the threshold as written and it failed. The honest gloss is that the metric measured our own pipeline this week rather than Roblox's demand.
- ✗ MISS
Build toward a dated seasonal event rather than a republish. Murder Mystery 2 remains a net accelerator in the window ending August 24. MED-HIGH
Murder Mystery 2 came in at -146.9M, from 523.4M to 376.5M. A decisive net decelerator. Two things keep this a clean miss rather than an excuse: the event ran to August 23 as we said, and the content-update field still reads August 14, so the game shipped nothing new and the call's premise was intact. It broke on a metric the enforcement action re-levelled. A visit-count call is only as good as the visit count, and we had no reason to think that was at risk.
- ✓ HIT
Treat the collect-sim cohort as update-responsive and volatile, and stop calling its direction. At least two of Grow a Garden 2, Fish It! and Fisch carrying an in-window update date are net accelerators. MED
All three carry an in-window content-update date, and two are net accelerators: Fisch +5.4M and Fish It! +2.3M, against Grow a Garden 2 at -8.2M. Threshold met. No dated public driver; three ordinary content updates, two of which worked. This is the first week in three we described this cohort correctly, and we did it by refusing to call its direction.
- ✓ HIT
Survival-to-Fortnite is still the one real opening, but build it as a genre bet rather than the boss-fight shape. Still zero Survival/Co-op islands in the Fortnite top-15. MED
Zero, for a ninth straight week. Fortnite Survival/Co-op is one island with any CCU at all, at 0.1%. The shape evidence gained its second observation, as the call asked: that island, ROSE Party Royale 2.0, runs 31.0 minutes per player against a 13.9-minute median. Two independent weeks now point the same way.
- ✓ HIT
On VRChat, build a room or a tool, and measure it against the platform rather than against zero. MED
Both halves. A private-rooms chill hangout is still carrying its crowd, and FISH ǃ [PYRAMID] kept handing back the crowd its update brought in. A ninth consecutive week for the destination-over-event read, now through four straight weeks of platform-wide softening.
- ✓ HIT
Forward call registered August 10 for this issue: a first-party social island absorbs hangout demand rather than seeding a third-party lane. Fortnite Roleplay/Life/Social stays below 1.5% of CCU and zero hangout or social islands enter the top-15. MED
Both legs. The genre came in at 1.10%, and the top-15 is nine combat and parkour islands, two GTA driving islands, two tycoons, a hide-and-seek and a horror. Epic published its own social island, Sprite Garden, into the lobby Discovery menu on August 20, and the hangout demand went there. Epic's island never entered our own figures at all, so this is not that island's traffic showing up in our number. The lane simply stayed empty for everyone else.
Read every figure in this chart against the August 19 enforcement action, which is why we are showing the whole set rather than a selected three. Two of these lines are really title lines: Simulator's rise is -0.1% excluding Steal An Egg, and the platform total is -0.51% excluding Murder Mystery 2, which sits in Horror/Escape. The genres with the mildest moves, Roleplay at -7% and RPG/Adventure at -2%, are the closest thing this week has to an undistorted read, and they describe a roughly flat week. Roblox Obby is omitted entirely: a single large course carrying roughly 80% of the genre left our figures, so the remaining number measures our own coverage rather than the platform.
Survival / Co-op PvE
of new visits
Roblox
of CCU
Fortnite Creative
Survival-to-Fortnite is a ninth straight week of vacancy, and it is now the only gap on this board we are willing to call an opening. Roblox routes 9.7% of new visits into Survival and Co-op PvE. Fortnite routes 0.1%, which is one island carrying any CCU at all. Last week the prescription rested on a single data point and we said so. This week it has a second: that island, ROSE Party Royale 2.0, runs 31.0 minutes per player against a 13.9-minute median across Fortnite islands. Two independent weeks, both showing the same thing, that when a co-op island does appear in this vacuum it holds people roughly twice as long as the median island in a genre nobody is serving. What has changed is our confidence in the shape rather than in the gap. The gap has been stable for nine weeks. The shape claim, that the winning form is a longer bounded co-op session rather than a short boss fight, now has two observations instead of one, which is why this week's call grades the genre leg and the shape leg separately rather than as a single conjunction.
A concentrated curve concentrates its errors too
The two Roblox stories this week are the same defect pointed in opposite directions, and both are genre lines that are really title lines. Murder Mystery 2 dragged a genre and a platform total down; Steal An Egg dragged one up. On a winner-take-all curve a genre average is a claim about the largest title in it and very little else, which is why we lead with the single-title figure whenever one is available. It is also why a platform-level measurement change reaches a small creator through whatever incumbent sits above them in the same bucket. Fortnite's curve is the reason the season beat the IP this week. Its long tail means a creator does not need to win, only to exist while attention is routed past them, and a dated first-party season is the one event that reliably reroutes it. On this curve you do not lose to a competitor, you lose to a rotation, which is why the correct Fortnite play is cheap, immediate and disposable.
Roblox
top-10 share 57.2% top-1 9.8%
Winner-take-all. The top 10 hold 57.2% of new visits. When an enforcement action lands on a giant, it lands on the genre line too.
Fortnite Creative
top-10 share 38.3% top-1 5.6%
Long tail. The median island peaked at 6 CCU on 13.9-minute sessions. A fresh niche island is viable here, and a rotation will take it away.
VRChat, in brief
No VRChat platform event landed in this window, so everything here is organic behaviour rather than a response to something the platform shipped. The platform has now softened for four straight weeks, and that changes what holding still means. The worlds that gained through it were a bar and a shared-screen room, both places where the point is doing something alongside other people. The sleep-and-cuddle cluster gave back more than the platform around it, across three separate rooms. Comfort looked like the durable format through the spring; through late summer it is the activity rooms that are holding. The destination-over-event read held for a ninth consecutive week. A private-rooms chill hangout has now carried the same crowd for six weeks running without shipping anything at all, while the fishing world that grows by dropping content kept handing back the crowd its update brought in. That is what it looks like when a place becomes somewhere people intend to return to rather than somewhere they currently are. When the platform drifts down, level is a win, and this week the worlds that stayed level were the ones with something to do in them.
Analyst notes
Two of last week’s calls broke, and both broke on the same August 19 enforcement action rather than on a bad read of what players want.
Call 2 said like-for-like Roblox demand would not fall below -2%. It came in at -4.18%. Call 3 said Murder Mystery 2 would remain a net accelerator. It came in at -146.9M. Both are graded as written, and both failed.
But neither metric was measuring demand that week. A pre-registered metric can be invalidated by a change in how a platform counts, with no change at all in what players did. We have written eleven weeks of calls on the assumption that a visit count measures behaviour. For one week in August it measured enforcement.
The way to tell them apart is the daily shape. Real players leave in the afternoon too, so an audience that is genuinely shrinking pulls the whole curve down. Murder Mystery 2’s afternoon peak fell about 23% while its overnight floor fell about 76%. That is a population being removed once, not an audience leaving continuously.
Our two structural calls are now nine weeks old. Calls 5 and 6 have held every week since late June, through a Fortnite season launch and four weeks of VRChat softening. They are about platform shape, which changes slowly, and that is why they keep holding while the week-to-week calls break.
What this costs going forward is a baseline, not a thesis. Every Roblox week-over-week figure that straddles August 19 compares an inflated week to a cleaned one.
Each call below gets graded in next week's issue against the metric named on it.
-
Re-baseline every Roblox number you own to August 18, and do not read the next two weeks as demand. The August 19 action is a level shift, not a trend. Murder Mystery 2's afternoon peak fell about 23% while its overnight floor fell about 76%, which is the signature of a population being removed once rather than an audience leaving continuously. HIGH
Graded next week on: Grade 2026-08-31: the like-for-like comparison returns to within 2.0% week over week, and Murder Mystery 2's 16:00 ET peak stays within 15% of its August 23 level of 814,048.
-
On Fortnite, the tradeable event is the season launch, not the IP. The GTA theme halved with its own showcase still ahead, while an ordinary parkour island went 80 to 536 and set its all-time peak on launch night. We are naming the confound in advance, because the last time we wrote a Fortnite call without pricing in a scheduled event, it broke: the GTA VI showcase airs August 27, inside the window this grades, so a fall in PvP share may be the showcase rather than season decay, and we will separate the two before grading. MED-HIGH
Graded next week on: Grade in the issue covering August 27: Fortnite PvP/Fighting/Shooter holds at least 48% of CCU.
-
Do not build a Roblox merge or collect sim off this week's +23% Simulator line. The genre excluding Steal An Egg is -0.1%. One month-old title is the entire move, and it has already taken most of its lifetime visits in seven days. MED-HIGH
Graded next week on: Grade 2026-08-31: Roblox Simulator/Tycoon like-for-like new visits excluding Steal An Egg stay within 5% week over week.
-
On VRChat, build the reason to be there, not the comfort. Through four weeks of platform-wide softening, the rooms that gained were a bar and a shared screen, while the sleep-and-cuddle cluster gave back more than the platform around it. Comfort is not a retention mechanic in late summer. An activity with other people in the room is. MED
Graded next week on: Grade 2026-08-31: at least two of YTS 2.1, The Black Cat and Popcorn Palace are still carrying their crowds, and at least two of everyone is asleep except you, place to eat cereal and Just A Cabin have given back more.
-
Take Survival-to-Fortnite as a long-session co-op island. The shape claim now has its second observation: Roblox Survival at 9.7% against Fortnite at 0.1% is a ninth straight week of vacancy, and the one island in it runs 31.0 minutes per player against a 13.9-minute median. MED
Graded next week on: Grade in the issue covering August 27, as two separate legs: first, still zero Survival/Co-op islands in the Fortnite top-15 by peak CCU; second, any Fortnite Survival/Co-op island appearing anywhere in our figures, not only in the top-15, reports at least 20 minutes per player. If no such island appears at all, the second leg is reported as untested rather than passed.
-
The Roblox Kids and Select bar came down, and the gate is not where most write-ups put it. The Highly Engaged Player threshold halved from 500 to 250 on August 19, the second loosening since the definition was relaxed on July 6. But highly engaged is not engaged: Roblox's own documentation requires account tenure, playtime in your game, and platform spend, so the real bar is 250 unique plays by age-checked users who have made a purchase somewhere on Roblox in the last 60 days. And the engagement bar is the second gate, not the first: publishing to these audiences at all requires ID verification through a government ID or a linked parental account, two-factor authentication, and either Roblox Premium held for two consecutive months or a one-time refundable fee. There is also a paid expedited-review route that skips the engagement requirement entirely. This is a real opening and it is cheaper than it was, but it is not free, and anyone selling it as just 250 players has not read the requirements. One dated qualifier, published August 25 and after the window this issue measures: Roblox now bars experiences from Kids and Select that combine a media feed, autoplay or self-scrolling, and a reward for continued watching, and removes non-compliant experiences already there. If your retention design is a watch-for-rewards loop, that route is closed. MED
Graded next week on: Grade 2026-09-14, allowing the 60-day window to accumulate: at least two Roblox worlds first seen in our figures on or after August 19 reach the top-100 by new visits.
- WARN This week's Roblox headline is a measurement caveat, and we are leading with it rather than burying it. Roblox ran an enforcement action against bot accounts on August 19 and stated that visit and player data in Creator Hub Analytics may shift for affected periods. Every Roblox week-over-week figure in this issue straddles that date. We report the like-for-like number as written, at -4.18%, and immediately give the single-title breakdown that explains it.
- INFO The evidence that this was enforcement and not demand is the daily shape. Murder Mystery 2's overnight floor fell from 1,013,583 to 244,048 in one four-hour interval and stayed down, while its afternoon peak drifted from 1,062,508 to 814,048 across six days. Bots have no daily cycle, so removing them flattens the floor and leaves the ceiling. Public reporting independently pins the same collapse to bot removal.
- INFO The two Roblox columns sit on different bases, by design. Share covers this week's full set of games; week-over-week compares only games present in both weeks. Every Roblox direction claim in this issue rests on the like-for-like column.
- WARN Two genre lines this week are really title lines. Simulator's +23.4% is -0.1% excluding Steal An Egg. The platform's -4.18% is -0.51% excluding Murder Mystery 2. We give the single-title figure in both cases because a genre average in a winner-take-all curve is mostly a claim about its largest title.
- WARN Roblox Obby is omitted from the table this week. A single large course carrying roughly 80% of the genre dropped out of our figures, so the remaining number measures our own coverage rather than the platform. We would rather show a gap than a wrong number.
- INFO No per-genre week-over-week is shown for Fortnite. The two weeks share no island codes at all, so any cross-week genre change would compare two entirely different sets of islands. Fortnite claims here rest on within-week share, top-15 composition, and per-island movement inside the week.
- INFO The GTA cohort figure is a composition comparison, not a same-island delta. The move from 14.3% to 7.5% compares the cohort's share of CCU across two different sets of islands. Island first-seen dates carry no information on Fortnite, because effectively the whole set is new every week.
- INFO Fortnite genre placement follows an island's tags rather than its theme, which is why the GTA cohort sits in RPG/Adventure this week and sat in Simulator during the previous wave.
- INFO Spot-checks: three run, all clean, at 1.2%, 0.2% and 4.1% against the live figures.
- INFO VRChat is checked on favorites and not on presence. Live presence is not verifiable without a logged-in session, which is one reason the VRChat read in this issue is qualitative, and policy is the other.
- WARN Two coverage gaps we are flagging against ourselves. A large Roblox Obby title left our figures entirely, and a number of worlds were captured only once this week rather than across the window. Both narrow what the Roblox figures can support, and both are ours rather than the platform's.
- INFO Method caveats. Fortnite island codes rotate through discovery, so per-island week-over-week is not possible. VRChat reads stay qualitative and carry no per-world figures. Newly seen means a world first showed up in our figures, not that it launched. Figures refresh roughly every four hours; spot-checks are order-of-magnitude against readings that may be a few hours old.
Get the next issue by email
Every Tuesday, about 10:00 ET.
Free and weekly, with double opt-in. Unsubscribe any time.