What sixty million views bought

In August 2025, 62 people played at least one pickup match. In May 2026, 172 did. For the twenty months before that August, the monthly figure never left the 54 to 76 band.

Over the same stretch the community's YouTube channels published 290 Shorts, holding 60.7 million views between them, and 39 long-form videos holding 2.0 million. Both can be dated against the day each new player joined and against the day each new person joined the Discord.

The population tracked the 2.0 million. Most of this article is the work of showing that with something better than dates lining up. The rest follows the arrivals themselves: who they turned out to be, which mode absorbed them, and how many are still playing.

Two sources here are private analytics shared for this article: channel exports from Kronkleberry and justkoala, and the Discord server's join log.

Everything on the pickup side comes from the match record TronStats keeps: 1,877 fortress matches from 2022-09-30 onward and 6,682 TST matches from 2023-07-26 onward, through 2026-07-27. That record covers pickup servers only. Someone who installs the game, plays a public server and never joins a pickup does not appear.

Two flat years, then a wave

From January 2024 to August 2025, twenty consecutive months, an average of 6.9 players joined pickup per month. The busiest of those months produced 14. Then: 22 in September 2025, 41 in October, 51 in January 2026, 70 in April, 63 in May.

Those arrivals came in bursts. Every seven-day window holding at least six debuts and at least two and a half times the rate of the preceding eight weeks. Grouped into periods they make four waves, which the rest of this article refers to by name.

For scale: 262 people had ever played a pickup before September 2025, and 62 of them played one in August 2025. That is the scene these waves landed on.

Who the arrivals were

Growth in an old game can come from two places: people who have never played, and people coming back. Separating them is harder than it looks, and the answer is different for recent absences than for old ones.

Start with what the record can see. Counting every case where a player went 90 days or more without a match and then played again, it holds 198 such returns by 103 distinct players. Forty nine of them have done it more than once. The median dormancy is 168 days and the longest is 966. Of the 180 returns old enough to observe a full 90 days afterwards, 44 were a single day and nothing more, while 88 produced five or more active days.

What matters for the waves is the rate. Monthly returns have stayed between 0 and 12 for the entire history, including every month of the growth period. September 2025 had 11, April 2026 had 6, June 2026 had 7.

198returns after 90+ days dormant 0-12returns per month, every month on record

That count only sees a return if the player has a recorded match since 2022. Somebody who played this game in 2015, stopped, and came back in 2026 has no earlier row. They enter the data as a brand-new player, and nothing in a match log distinguishes them from one.

They are in there, and not in small numbers. Eleven of the 370 arrivals since September 2025 are known to be players who were around years ago and came back: DaWg, viper, Prestige, gyal, ilona, Solut, larry, may, naiss, frank and ?????. Six of them are in the fifteen highest-volume arrivals. Between them the eleven account for 29.8% of every match the arrivals have played, from 3% of their headcount.

Sixty million against two million

Nine community channels published 290 Shorts and 39 long-form videos over the same period.

Every view counted here is a YouTube view. At least one of these creators posts the same clips to TikTok, sometimes there first, and nothing in this corpus can see that. Where it happens a clip's real reach is larger than the number printed beside it, so the per-view figures below understate how far the Shorts travelled rather than overstating it.

The Shorts hold about thirty times the total views of the long-form videos. That gap is a tail rather than a rule: the median Short took 6,537 views against the median video's 2,224, and one clip, the 23.7 million view Short, is 39% of the Shorts column on its own. If reach were the thing that mattered, the debut curve would be a copy of the Shorts curve.

ShortLong-form video

The two long-form rules land on the two largest weeks in the record. The week of the 2026-04-14 video produced 37 debuts, the highest week there has ever been. The week after the 2025-12-30 video produced 22, joint second with the week that followed the April video. The 23.7 million view Short of 2026-03-11 sits on a week of 16, which then falls back to 5. The 6.7 million and 6.3 million Shorts of December sit on weeks of 3 and 4.

The monthly totals say the same thing. March 2026 published 25.2 million Shorts views and produced 34 debuts. April 2026 published 0.16 million Shorts views and one long-form video, and produced 70. July 2026 published 5.8 million and produced 12, the quietest month since August 2025.

Between watching a video and playing a pickup there is a step: joining the Discord where games are organised. That log exists, and it moves with the videos rather than with the views.

ShortLong-form video

In the 21 days after each upload, the 0.39M video was followed by 372 Discord joins and the 1.3M video by 843. The 23.7M Short, with nearly fourteen times the reach of both videos put together, was followed by 236. Scaled to a million those become 957 and 626 against 10 to 58 for the six large Shorts, with the same caveat as before: the videos never had a million views to scale from.

The single clearest day in the whole record is 13 March 2026. The Short was watched 2,232,196 times, its best day. Thirteen people joined the Discord and four played their first pickup match.

What one upload does

Take the plainest version first, with nothing scaled. In the 21 days after it published, the 23,718,175-view Short was followed by 30 first-time players. The 388,753-view video, with a sixty-first of that reach, was followed by 40.

30debuts in 21 days, the 23.7M Short 40debuts in 21 days, the 0.39M video

Putting every upload on the same footing means dividing by reach, which is the chart below. One caution about reading it, because the unit is larger than some of the uploads: these are rates, not counts. The video's 102.89 is what 40 debuts becomes when scaled up to a million views. It did not bring 103 people; it brought 40, from a fraction of the audience the Short had.

Six Shorts holding 54.2 million views between them come out at 1.26 to 8.53 debuts per million. The two long-form videos, holding 1.7 million between them, come out at 51.27 and 102.89. The 23.7 million view Short of 2026-03-11, the largest thing anyone in this community has ever published, sits at the bottom of the chart.

One window was genuinely ambiguous and it is the one that settles the question. olive published a 6,285,525-view Short on 2025-12-27. Kronkleberry published a 388,753-view video three days later, on 2025-12-30. Both sit inside the January wave, so window totals cannot tell them apart. Daily onset can, because a Shorts feed delivers most of its impressions inside 48 hours.

ShortLong-form video

The Short delivered no players in 72 hours. The rise began four days after the video and ran for three weeks.

The April window has no ambiguity to resolve. Within ten days either side of it the largest thing anyone published was a 50,032-view Short four days later, under 4% of the video's reach. The long-form video went up on 2026-04-14 and produced seven debuts the same day, against a baseline under one a day. The two largest debut days in the whole record are the 17th and 18th.

September 2025 cannot be read this way at all, and it is worth saying why. That month was a near-daily run of about 25 uploads in 24 days, so no single one of them has a window to itself. Nothing in it can be attributed, in either direction.

Where in the funnel the loss happens

Putting views, Discord joins and pickup debuts on one axis, normalised to a million views of each format, gives the whole path from a video to a game.

The two lines are close to parallel from the second stage onward, and that is the finding. Almost the entire difference is spent in one step.

A long-form viewer is 36 times more likely to join the Discord than a Shorts viewer. After that the two groups behave almost identically: they convert into players at the same rate, and they pick up TST at the same rate. The fortress gap, 67.9% against 57.5%, rests on 109 and 87 players and is small enough to be noise.

So Shorts are not producing worse players. They are producing far fewer entrants, and the ones who arrive are much like anyone else. Whatever a twelve-minute video does that a twenty-five-second clip does not, it does before anybody reaches the Discord.

One caution about that middle step. Discord joins and pickup debuts are two counts over the same window, not one group of people tracked through it. Nobody is followed from a join to a first match, so "Discord to a pickup" compares two populations that overlap by an unknown amount rather than measuring a rate at which individuals progress.

Six signals, one axis

Each of these is measured in something different: views, concurrent players, joins, people, matches. Scaling every one to a percentage of its own peak puts them on a single axis. That trade is worth naming: normalising compares shape and throws magnitude away entirely, so a series that ran from 4 to 46 and one that ran from 52,538 to 12,524,685 look the same size here.

The peaks do not line up, and where they fall is the argument of this article compressed into one row each.

March is the high-water mark for attention and for nothing else. Every human measure peaks a month later, alongside a video with 5% of the Short's reach. Then the ordering changes character: fortress peaked back in January, when that cohort of newcomers arrived, and TST peaks last, in a month with no upload at all, because by then the growth had stopped being an event and become a population.

Read the tail with care. TST's rise through May and June is the same people playing more, not new people arriving; new players were already falling by then. And a monthly bucket is coarse enough that a video published on the 30th belongs almost entirely to the following month, which is why January's numbers move on December's upload.

Steam is the most detached measure of the set, since third-party concurrency trackers can see nothing of the Discord, the videos or the pickup record. Its three largest month-over-month rises are the three long-form months: +206% in October 2025, +113% in January 2026, +231% in April 2026. The month of the 23.7 million view Short rose 76%. September 2025, which published 11.9 million views of Shorts, is the lowest month in the series at 3.82 average concurrent. The game's all-time concurrency peak falls in April 2026; the two trackers I checked disagree on its value, 146 against 121, but agree it is the record.

Fortress and TST are not the same problem

The two modes moved in opposite directions for three years. Fortress went from 556 matches in 2023 to 459 in 2024 to 326 in 2025. TST went from 464 to 1,826 to 1,885 across the same years.

The structural difference is headcount. Fortress is 12 players, six a side, and 1,509 of the 1,877 recorded matches have exactly 12 distinct players in them. TST is 8, and 5,764 of 6,682 have exactly 8. Averages come out slightly above both, 12.2 and 8.1, because substitutes rotate through a match. Fortress needs half as many people again as TST, every time, before a game can start.

The difference shows up in how often each mode happens rather than how large it is. In 2026, TST ran on 205 of the 208 elapsed days. Fortress ran on 143. And when fortress does run, it runs in a short block: matches per active day has stayed between 2.3 and 3.1 every year since 2022, while TST's climbed from 3.5 to 12.2. Fortress has to be assembled. TST is simply running.

One clock or two

The same difference shows up on the clock. Hours below are given as GMT and EST together, in that order, so 4a/11p means four in the morning GMT and eleven the previous evening EST. Two notes before reading them: the underlying timestamps are UTC, which GMT matches exactly, and EST here means a flat five hours behind, so between March and November, when US Eastern runs on daylight time, the local clock reads an hour later than the label.

In plain terms: fortress is a night that starts after dinner on the US east coast and finishes in the small hours. It lives between 11pm and 6am GMT, which is 6pm to 1am EST, and peaks at 2am to 5am GMT, 9pm to midnight EST. Between 9am and 4pm GMT, the small hours through late morning EST, it has run exactly one match in four years.

TST covers that same block and adds a second peak at 8pm to 9pm GMT, 3pm to 4pm EST. That is mid-afternoon on the east coast and early evening in western Europe, which is the more likely reading of who is filling it.

The split has barely moved in three years. Below, the afternoon block is 5 to 10pm GMT, which is 12 to 5pm EST, and the night block 11pm to 6am GMT, which is 6pm to 1am EST.

TST's two blocks have held at roughly 42% and 47% every year since 2023, and its single busiest hour moved out of the late block into the earlier one in 2024, where it has stayed. Fortress has never had a second window.

One row is worth pausing on. 2025 was fortress's worst year, and it is also the year fortress was most concentrated: 85.9% of its matches in the late block and only 12.0% in the earlier one. When the mode is short of people it fires only in its single best hour. In 2026, with the waves arriving, the earlier window came back to 19.3%. A format that needs 12 people gets one shot a day at assembling them, and loses even that when numbers thin.

This also constrains how the next section can be read. If fortress only ever happens in one narrow window, then any two fortress regulars are being sorted by the clock before anything else sorts them.

Who keeps each mode alive

Both modes rest on a small group. Through 2025, between 12 and 16 players accounted for half of all play in each mode each year. The waves widened that to 21 in fortress and 25 in TST during 2026, the healthiest that number has ever been.

Attendance rates make the fortress dependency concrete. In 2025, Sanity appeared in 84.0% of every fortress match played, 274 of 326. In 2023 it was Ampz at 81.3%, 452 of 556. TST has never had a figure like that; its highest is Sanity at 56.8% in 2024, and in 2026 the top attendee is Izzimahizzi at 40.5%.

84.0%of 2025 fortress matches included Sanity 40.5%of 2026 TST matches included the top attendee

The decline is not the fortress core leaving. Take the fifteen players with the most fortress matches ever recorded. All fifteen were active in each of the five half-years from January 2024 to June 2026, and fourteen of fifteen in the second half of 2023. Nobody left. What changed is where their time went: fortress accounted for 60.1% of their play in the second half of 2023 and 27.6% by the first half of 2025.

So the question is whether fortress needs these particular people, or whether it simply happens on weeks when the scene is busy anyway.

Here is one way to ask. Take the eight players with the most fortress matches ever. For each of them, sort every week into two piles: weeks they turned up for anything at all, and weeks they appeared in neither mode. Then count how much fortress everybody played across the two piles. Weeks are only ever compared against other weeks in the same calendar quarter, so the three-year decline cannot pass itself off as an effect. TST is the control: if the answer were simply that busy weeks are busy, both modes would move together.

The first two columns are fortress matches per week played by the whole community, not by the player named in the row.

Read the top row as: across the weeks Sanity showed up, the community played 16.2 fortress matches a week; across the weeks Sanity did not, 7.3.

Every one of the eight is positive, by a median of 5.6 matches a week, which is about half the volume of a week when they are around. The control does not follow: TST runs from -8.3 to +11.5 with no consistent sign.

Two limits. The quarter comparison removes the era trend but not week-level common causes, so a tournament weekend that draws a regular in and fills the lobby at the same time would look like this. And Sanity's row rests on only 14 absent weeks. These are associations, not per-person coefficients.

What the data cannot settle is the direction. A regular is present because fortress is running, and fortress runs because regulars are present. Whether fortress thinned because people stopped organising it or because it stopped being available to join is not answerable from match rows, and I am leaving it open rather than picking the flattering version.

People arrive in pairs

Treating regulars as interchangeable units of attendance misses something. On any given fortress night, who turns up is not independent of who else turned up.

How much more likely one player is to turn up on a night a particular other player turned up, against a night that player skipped?

The window is 2026-02-01 to 2026-07-27, and the 42 regulars in it are the ones already playing when it opened and still playing in May, so nobody is judged on nights they were never around for. Ordered so that A following B and B following A count separately, they make 1,722 pairs. The median answer is 1.18 times, and thirty-eight pairs sit at three times or more.

The obvious objection is that busy nights pull everyone in at once, so this could be two people independently liking a full lobby. Splitting the nights into those that ran one or two matches and those that ran three or more, the pattern holds inside both. morbit played 84.8% of big nights when Sanity was there and 7.7% when Sanity was not; on small nights, 40.0% against 9.4%.

On nights their usual partner was absent and they skipped fortress, what these players did splits by person. morbit was playing TST on 66.7% of those evenings, Magi and stereo on roughly half. Shoebat was on TST for 0% of them, and orly for under 14%. Some people move to the other mode; others simply are not there.

Two limits on all of this. Nobody in the data is absolutely dependent on anyone: of 1,722 pairs, zero had a follower who never once played without their anchor. And a shared schedule looks identical to a shared preference from here. Two players in the same time zone are online in the same hours, and nothing in match rows can separate that from choosing to play together.

The newcomers went into fortress

Fortress recovered in 2026: 450 matches and 265 distinct players in seven months, against 326 matches and 118 players in all of 2025. Much of that is visible in a single week.

Four players who had never touched pickup before, arriving within eight days of each other starting eleven days after the 2025-12-30 video, now appear in between a quarter and nearly half of every fortress match played in 2026. viper is the third-highest fortress attendee of the year, behind Izzimahizzi at 56.2% and Sanity at 44.4%, and did not exist in the record in December.

Those four are the visible tip. Of the 370 people who first appear from September 2025 onward, 32 have passed 100 matches and 77 played within the last 30 days. That second number is the honest one: a fifth of the wave is still here.

What they add up to is larger than their headcount suggests. In 2026 the wave newcomers filled 12,508 of 32,714 player-slots, and 2,802 of the year's 2,957 matches included at least one of them. The scene did not gain a fringe. It gained about two fifths of everybody on the grid.

38%of 2026 player-slots filled by wave newcomers 32newcomers past 100 matches

The fifteen highest-volume arrivals, ordered by total matches and counted per distinct match. Six of them are the returning veterans named earlier, not beginners. Because they arrived as much as eight months apart, the last column is the share of days since each player's own first match on which they played, rather than a raw count.

That last column is what the raw counts were hiding. DaWg has been on the grid for 73% of the days since arriving ten months ago. But gyal and Toasty, the two most recent arrivals here and modest against anyone with a longer run, have played on 90% and 87% of the days available to them. The mode mix varies far more than the volume does: since January, Prestige has played three fortress matches and viper has played 191.

The slopes are the point. gyal played 179 matches in the first month after arriving in May and 405 by the end of June, a steeper start than anyone else on the chart. DaWg, ten months in, is still climbing at roughly the rate they set in the first month. These are not people trying the game; they are people who now play it.

The loop closes on the creators. Kronkleberry, whose channel produced both long-form videos and the 23.7 million view Short, is Kronkleberry, who has played fortress since 2022-10-03 and attended 35.3% of 2026's fortress matches. Koala, SanityTron and ellis are koala, Sanity and ellis. gyaltron is gyal, who joined in May 2026 and is already the fourth highest-volume newcomer of the entire wave. Bozo, who arrived on 2026-01-10 through someone else's video, published a competitive Retrocycles guide of his own on 2026-04-29.

Retention is the part that has not been solved

How much of each arriving cohort is still playing three months later:

Survival drifted from about a quarter to about a seventh across two years, and the most recent cohort is up rather than down. A drift, not a collapse.

By wave, the differences are real but not in the direction of size:

The January wave is the smallest of the three large ones and by some way the best retained. The April wave brought twice as many people and kept a third as many of them at 90 days, though its figure covers only the 103 arrivals old enough to measure.

The recent cohorts are better than the 2025 trough on the measures visible this early. The first group below is 90 players, the second 131.

Part of that is supply rather than stickiness: arriving during a busy period means more games are available to join.

Which mode holds them

The obvious next question is whether one mode keeps newcomers better. The answer, on two different tests, is that it does not.

The first test is what was running the night they arrived, which is close to random from the newcomer's point of view: you play whatever the lobby is playing. Within each wave, the mode of a player's first match makes almost no difference to whether they are still there 30 days later.

The second test looks like a strong result and is mostly an illusion. Players who tried both modes retain far better than players who tried one: in the January wave, 65.6% of the both-modes group was still playing at 90 days against 0% of the TST-only group. But someone who played on one evening cannot have tried two modes. The group is partly defined by the thing being measured.

Splitting on how often a newcomer showed up in their first fortnight, and measuring mode breadth inside that same fortnight, separates the two.

Inside each band the three mode groups sit within a few points of each other. Between the bands there is a factor of four. What predicts whether a newcomer is still playing three months later is not which mode they found, it is whether they came back twice in the first fortnight. Mode breadth was carrying that signal, not producing it.

The two bands are coarse, and the underlying relationship is not. Counting the separate days a newcomer turned up in that first fortnight, against whether they were still playing at 90 days:

Every step up adds survival, from 8.4% for the people who played one day and did not come back inside the fortnight, to 60.0% for those who turned up on seven days or more. There is no threshold anywhere in it. It is a gradient rather than a switch that flips at some particular session, which means there is no single number of games to aim a newcomer at. Restricting the same measure to arrivals from September 2025 onward gives the same shape on smaller groups: 7.1, 12.2, 17.6, 26.1, 60.0.

What the curve cannot show is which way the arrow points. Somebody who turns up on seven separate days in a fortnight is already somebody the game took hold of, so this may be measuring enthusiasm the player arrived with rather than a habit the scene built. Nobody was assigned a number of sessions.

If the arrow does point the way it reads, the practical version is that getting a newcomer into fortress specifically matters less than getting them back at all.

The counterweight sits in the same data. 42.0% of the players who debuted in the second half of 2025 played exactly one match, ever. Most arrivals across the whole period are one evening long, and 16.7% of the 2026 first-half cohort reaching three months is an improvement on 2025 rather than a good number.

There is no era to measure that against, because the record does not reach back far enough to hold one. Whether roughly one in six is normal for a twenty-year-old game with a twelve-player lobby, or whether it is the thing to work on next, is not answerable from this data.

The 2026 cohorts cannot be read at twelve months until 2027. Two things can be said now, and this article has spent most of its length on the smaller one.

Most arrivals still leave, and the rate at which they leave has barely improved. But a modest rate applied to a much larger intake leaves more people behind than a good rate applied to a small one. The 2024 first half kept 15 players past three months. The 2026 first half kept 28.

Everything else follows from that. 2026 has run 2,957 matches in seven months, against 2,211 in the whole of 2025 and 2,285 in 2024. Fortress, after three straight years of decline, has already played more this year than in all of 2025. Two fifths of every player-slot filled in 2026 belongs to somebody who arrived within the last twelve months.

So the answer to the question in the title is that sixty million impressions bought a fraction of what two videos did. What those two videos coincided with is a scene larger than at any point this record covers.