The biggest mover in our entire six-week limiteds census lost 29 favorites. It started with 83,795.
That's a decline of 0.035%, and it was the most dramatic thing that happened to any of the 475 items we watched between June 28 and August 12, 2026. We banked snapshots on 32 distinct days across that window - not all 46; the collector missed a stretch in early July and a few days elsewhere, and we'd rather tell you that than smooth over it - and for every item with at least 20 snapshot days we compared the first reading against the last. Favorites, the one public number in our census that still updated at all: 404 items flat, 69 down, 2 up.
The two gainers were Pinstripe Fedora and Banana Cat. Each gained exactly one favorite. In six weeks.
The shape of the stillness
The median absolute change in favorites across all 475 items is zero. Not close to zero - zero. Of the 71 items that moved at all, 61 moved by one or two favorites. Eight moved by three to five. Two broke double digits: the Nova Grunge Gemstone Sword's −29 mentioned above, and Frozen Horns at −17 - on a base of 413,866, a decline of 0.004%. The largest move in percentage terms belongs to Black Valkyrie, which shed 5 of its 426 favorites, a heart-stopping −1.2%. All told, 30 items out of 475 moved by more than a tenth of a percent of their base. Nothing moved by more than two percent.
Movement, such as it is, runs almost entirely one direction. Sixty-nine down, two up: people occasionally un-favorite an item they favorited years ago, and essentially nobody new arrives to favorite a classic limited. That asymmetry is probably the census's one genuine finding about the market rather than about the instruments - the audience for these items isn't churning, it's very slowly leaving.
The columns that didn't even pretend
Favorites was the lively column. The others didn't twitch once.
RAP, for the 167 items in the deep set that carry it, showed exactly one value per item across all 32 days. Not one item showed a second reading. We thought we knew why - Roblox's legacy resale feed stopped recording in early 2025 - but that explained our collector, not the market: the feed we read had stopped, while Roblox's current resale endpoint kept recording sales. Same story for the sales counts that rode the same feed: 167 items, zero changes, six weeks, all of it an artifact of the endpoint.
One column did appear to move, and it's worth the disclosure because it's the kind of thing that fools people, including us for about an hour. The remaining-quantity field changed for 308 items - all on June 29, the second day of the census, all landing on exactly the item's total quantity, and never again. That's not the market doing anything. That's our own collector changing how it read the field, one day into the window. A less suspicious query would have reported "65% of limiteds saw supply movement" and been completely wrong. Longitudinal data punishes you for trusting your own pipeline slightly less than it punishes you for trusting anyone else's.
Stopped clocks, and the one that ticks
Put the columns together and the picture looked stark. RAP: frozen since early 2025, in the feed we read. Sales: frozen with it. Favorites: technically live, but moving at homeopathic doses - a median change of zero over six weeks. A six-week census says the favorites needle is painted on. Judging a limited by its favorites count in 2026 is reading a stopped clock and crediting it for consistency; judging it by the legacy feed's RAP was the same mistake, and it was ours.
The contrast is what makes this interesting rather than merely bleak, because Roblox does have public numbers that are alive. Live player counts move constantly - when we measured the weekend cycle, the swing was about 8% inside a single week, and update spikes move some games 50% in a day. One week of player-count data contains more movement than six weeks of the limiteds data we were collecting by a factor we didn't bother computing because one of the factors is approximately zero.
This is also why BloxToolbox has never shipped a limiteds-history sparkline, and now you've seen the query that justifies the refusal. A sparkline over the data we had would be 475 flat lines with axis labels, sold as insight. We won't chart it. We did, however, just ship a history tool for the signal that earns one: the new game player-count history tool charts the numbers that actually move, from the same first-party snapshot bank this census came out of.
So what still carries information about a limited, if its favorites count doesn't? The things you can watch happen: live resale listings and how fast they clear, the trade offers an item actually attracts, and player counts for the games that drive its demand. All harder to read than a favorites counter. That's roughly the point - the favorites counter went dead, the RAP we were reading had stopped, and the market kept moving anyway.
Six weeks is one census, and a longer window may catch drift this one couldn't. The snapshots keep banking either way. But if you've been refreshing a limited's page waiting for its numbers to tell you something, we can now put a measurement on what you're waiting for: about one favorite, possibly two, and it'll probably be leaving.
Corrected September 11, 2026: what changed.