PLAY'S · Meta · 27 Sep 2026 · Updated 29 Sep

Six days, four leaders, and a tier list that leaves one out

This site wrote on 25 September that the first reading of the set 37 format did not exist yet, anywhere. It exists now and it is six days long. The most-played deck's share has fallen on every one of those days, from 26.8% to 13.3%, and four different decks have held first place. That is what an unsettled format looks like when you count it. Here are the numbers, with every date attached.

The dates on this page

24 Sep 2026: Set 37 released during a maintenance window of 09:00–17:00 JST. DMPP-32 left New Division at the same maintenance. The format starts here.
24 Sep: First numbers — but only the hours after 17:00. 82 games.
25 Sep: First full day. 308 games.
26 Sep: Second full day. 333 games.
27 Sep: Third full day. 386 games.
28 Sep: Fourth full day. 216 games — the smallest full day so far.
27 Sep: This page first went up, covering three days.
28 Sep: Rewritten around four days.
29 Sep: Fifth full day. 294 games.
29 Sep: Rewritten once the 28 September report appeared.
30 Sep: The strategy site publishes its first tier ranking for this format.
1 Oct: This rewrite. The format is six days old.

What the numbers are

A player posting as @sumamame logs the decks they face in ranked play, and the tally is published through the strategy blog BEANS, one report per day, split by division. It is the same survey, by the same method, that this site used for the old format, which is what makes the two comparable. This is an encounter rate: how often you sat across from a deck. It is not a win rate, and nothing here says any of these decks is good.

Six days, side by side

New Division	24 Sep	25 Sep	26 Sep	27 Sep	28 Sep	29 Sep
Jokers	4.9%	4.2%	5.4%	7.5%	11.1%	13.3%
Cotton Zenith	4.9%	5.5%	7.2%	9.3%	13.9%	9.9%
Fire/Light Raid	24.4%	16.6%	16.5%	15.8%	11.6%	12.6%
Water Magic Tools	26.8%	20.5%	19.5%	13.7%	13.4%	9.2%
Baraghiara Loop	3.7%	9.7%	15.0%	10.4%	8.3%	9.5%
mono-Fire Balga	8.5%	11.7%	10.5%	7.0%	11.6%	6.5%
Water/Darkness Magic Tools	—	—	3.0%	—	4.6%	—
unlabelled "other"	8.5%	12.7%	15.1%	16.4%	15.8%	13.2%
games in sample	82	308	333	386	216	294

Six of the decks the source breaks out, chosen because they carry the argument below, plus its unlabelled remainder. Bold marks the most-played deck on each day. A dash means the source did not break the deck out that day. Figures from the BEANS New Division reports for 24 to 29 September 2026. The Jokers row changed label on 27 September, from a named build to the bare tribe, so its first three figures and its last three may not be counting quite the same thing.

The number that measures "unsettled"

Read only the bold figures, left to right: 26.8, 20.5, 19.5, 15.8, 13.9, 13.3. The most common deck in the format has been less common every day than the most common deck the day before. Over six days that share has halved exactly.

And the name on it has changed three times. Water Magic Tools led for three days, then Fire/Light Raid, then Cotton Zenith, then Jokers — one day each. The first two changes did not come from the new leader surging past: Raid took first place on 27 September while falling, because Water Magic Tools fell further. Jokers is the exception, and the only deck in the table that has risen on five of the six days: 4.9, 4.2, 5.4, 7.5, 11.1, 13.3.

First place on 28 September was one game The 28 September report counts 216 games. At that size, Cotton Zenith's 13.9% and Water Magic Tools' 13.4% work out to 30 games and 29. Raid and Balga, both at 11.6%, are 25 each.

So "Cotton Zenith is the most-played deck" is true of that report and should not be read as more than that. What the fifth day does show is that no deck is clear of the rest: the top four are within five games of each other.

The deck that climbed for five days and then did not

Cotton Zenith went 4.9, 5.5, 7.2, 9.3, 13.9 — and then 9.9. For five days it was the one deck in the table that was more common on each day than the day before, and it nearly tripled. On the day set 37 landed it was 4.9%, tied for fifth in that day's report. On 28 September it was the most-played deck in the format. On the 29th it gave back most of the previous day's gain.

The version of this page written on 29 September said that line "looks exactly like Baraghiara Loop's did then", meaning a climb that might be about to bend. It bent the next morning. That is the second time in this format that a deck has been the fastest riser on one report and a faller on the next.

This site has not yet written about what the deck does or which set 37 cards, if any, it uses, and an encounter tally cannot say. Its line was a climb, not a verdict — which is the whole argument of the section below it.

The loop deck peaked on the third day

Baraghiara Loop went 3.7, 9.7, 15.0, then 10.4, 8.3, 9.5. It quadrupled in three days, gave back almost half of that peak in the two after, and has since steadied.

When this page first went up on 27 September it had three of those points and described a climb. The next report turned the climb into a spike, and the one after that confirmed it. Cotton Zenith's line today looks exactly like Baraghiara Loop's did then.

The source calls it a loop deck The label in the tally is バラギアラループ — Baraghiara Loop. Not the creature, not a colour combination. The deck is being recorded under what it does.

Which settles something this site got wrong

The day-one page carried three player posts saying the Baraghiara deck ran itself out of cards and that they could not work out what it was for. It also carried the other half of the conversation, where players called it the strongest thing in the set. Those posts were from 25 September, when the deck was at 9.7%.

On the day players were saying it does not work, it was already the fourth most common deck in the format, and it was half again as common the next day. Day one is not a measurement

The complaint and the climb are both real. People were describing their own games accurately and the format was moving underneath them. It has since moved back. That is what a young format looks like, and it is the argument for not publishing a verdict during one.

The tier list arrived on day six

On 30 September the Japanese strategy site published its first tier ranking for this format, six days after release. It had ranked All Division two days earlier. Its top tier holds two decks: Nature Jokers and Fire/Light Raid.

Those are the two most-encountered decks in the 29 September report — 13.3% and 12.6%, first and second. On the top of the format, the ranking and the encounter data agree.

It does not rank Cotton Zenith at all Cotton Zenith was the most-played deck in the format on 28 September and 9.9% on the 29th. It is not in the first tier of that ranking, or the second, or the third.

That is not a contradiction — a tier list is an opinion about how strong a deck is, and an encounter tally is a count of what people queued into. A deck can be widely played and thought unremarkable, and the two sources are measuring different things. But it is the sharpest example on this page of why one number should not be read as the other.

The two sources also name decks differently — the ranking uses civilisations where the tally uses colours, so 火光レイド and 赤白レイド are the same deck written two ways. Matching them up is a judgement, not a lookup, and this page has only done it where the match is unambiguous.

A sixth of the format has no name

The survey's unlabelled "other" share went 8.5, 12.7, 15.1, 16.4, 15.8. It stopped growing on the fifth day, but it has now been above 15% for three days running — larger, on 28 September, than any deck the survey names.

A format settling down would push that number the other way, as players converge on fewer decks and the survey names them. Five days in, it has not started to.

The Grand Prix deck, and the one that replaced it

Water/Darkness Magic Tools won the Grand Prix on 13 September, eleven days before the set landed. It was never a common deck: the same survey had it at 6.2% and then 6.9% in the old format, sixth on the ladder on the day it won. In this sample it is broken out on only two of the five days: 3.0% on the 26th and 4.6% on the 28th. On the other three it is either absent or sitting inside the unlabelled "other" share; this data cannot tell the two apart, and it would be wrong to say the deck vanished.

The mono-Water build is the one near the top — first on each of the first three days, and within one game of first on the fifth. This is the same split the strategy site files under two different formats, with the Darkness version attached to the format that ended on 24 September. On 25 September this site could only say the two were probably different decks. Five days of encounter data say the newer one replaced the older one immediately.

And the Fire line was never missing

The day-one page said the set's new Fire cards went unmentioned in all twelve posts it read, and that this was a third of the set's coverage going undiscussed. Fire decks are 32.9%, 28.3% and 27.0% of the first three days — between a quarter and a third of the format.

The later days cannot be added to that line On 27 September the survey stops counting Fire in two buckets and starts counting it in five, splitting out variants it had not named before. Those five add to 32.4%, which looks like a jump — but the categories changed underneath the number, so it is not the same measurement. On 28 September it counts three, adding to 26.9%.

This is worth stating plainly because it is the kind of thing that produces a fake trend: a survey that reclassifies mid-week will show movement that no player caused. The Water/Darkness line above, broken out on some days and not others, is the same problem.

These are two different claims, and only one of them is settled Fire is plainly being played. None of the twelve posts named Raid, and Raid was in the top two on each of the first four days.

But an encounter tally records which deck you faced, not which cards were in it. Whether those Fire decks are running set 37's new Fire cards is not something this data can show, and both of these archetypes existed before the set. The day-one page's actual claim — that the new Fire cards were not being talked about — is still open.

Raid held still for three full days, then dropped The last reading of the old format tracked Raid at 19.8%, then 17.8%, then 17.9%. In the new format it read 16.6%, 16.5% and 15.8% on the first three full days — barely moved by a set rotating out and more than ninety cards coming in. On 28 September it is 11.6%, on the smallest full-day sample so far.

An earlier version of this page said the players who called Raid unchanged were right about it. They were right for three full days. Whether the fifth is a turn or a small sample is exactly what one more day cannot answer.

This page is part of the evidence

This article has now been rewritten twice in three days. It went up on 27 September with three days and a headline about a deck that had quadrupled. The next morning's report forced a rewrite around four days and a peak. The morning after that brought a new leader, decided by one game, and forced this one.

Each rewrite was caused by the next day's number, not by anything this site learned about the cards. The corrections below are dated so that anyone can see the sequence. A format that a single daily report can overturn is not one that has a shape yet.

What this does not tell you

These are encounter rates, not win rates. A deck can be everywhere and lose.
One collector, all rank bands mixed. Master and Legend games are pooled with everything else, so this does not describe the top of the ladder.
82 games on 24 September is a very small sample, and the hours after a maintenance are not a normal day. Treat that column as a direction, not a figure. 216 on 28 September is the smallest full day; at that size one game moves a deck by almost half a point.
A tier list exists as of 30 September, and this page is older than it in places. The strategy site ranked All Division on 28 September and New Division on the 30th. It is one site's opinion, published six days into a format whose most-played deck has changed four times, and it says so itself: its own note calls the rankings the editorial team's own assessment and asks readers to treat them as a reference only.
The source's own categories are not fixed. Fire was counted in two buckets on Saturday, five on Sunday and three on Monday; the Jokers entry changed its label on Sunday; and one deck is broken out on some days and not others. A figure is only comparable to the days that counted it the same way.

When the next reading is

This site will re-measure on 3 October 2026, at nine days old. By then there will be enough points to say whether Cotton Zenith's climb keeps going or turns the way Baraghiara Loop's did, and whether Raid's drop on the 28th was the start of something.

The day-one page asked whether "nothing has changed" was a mood or a fact. Five days in, the honest answer is still that the question is too early — and there is now a number for how early. The most-played deck is 13.9% of games, one game ahead of the next, and the one tier ranking that now exists for it is six days old and leaves out the deck that led on day five.

Corrected 27 September, shortly after publication Three things were tightened. This page said the Grand Prix deck "does not appear at all" on 24 and 25 September — the survey does not break it out on those days, which is not the same as it being absent, because it may sit inside the unlabelled "other" share.

It also treated "Fire decks are a quarter of the format" as settling whether set 37's new Fire cards were being ignored. Those are different claims: an encounter tally records which deck you faced, not which cards were in it.

And the collector's method is now described as what can be seen — logged ranked games — rather than "every ranked game".

Rewritten 28 September This page went up on 27 September with three days of data and a headline saying the loop deck had quadrupled. The 27 September report, published the following morning, put it at 10.4% — down from 15.0%. The page was rewritten around four days, and the third day became visibly a peak rather than a direction.

Rewritten 29 September The 28 September report added a fifth day and a third most-played deck. The headline, standfirst, table and closing section have been rewritten around five days, and the page now leads with the falling top share rather than with Baraghiara Loop.

Two claims from the 28 September version no longer hold and have been changed rather than left standing: that Raid "has not moved all week" — it fell to 11.6% on the 28th — and that the unlabelled share "has grown every single day", which it did for four days and not the fifth. Nothing else that was true on 28 September has been removed.

Corrected 29 September, evening This page said no tier list existed. One does. The Japanese strategy site published a set 37 tier ranking for All Division on 28 September. It is New Division — the format this page is about — that still has none, which is a sharper fact than the one that was here before: the same site has ranked the older, deeper format and not this one.

The heading "Raid held still for four days" was also wrong. Raid read 24.4% on 24 September and 16.6% the next day. It held still for the three full days after that, not four. The body of that section always said three; the heading and one sentence did not.

Rewritten 1 October The 29 September report added a sixth day and a fourth most-played deck. Jokers reached 13.3% and took first place; Cotton Zenith, which led on the 28th, fell from 13.9% to 9.9%. The previous version of this page said Cotton Zenith's line "looks exactly like Baraghiara Loop's did then" — it bent the next day.

And a tier ranking for this format now exists, published 30 September, which the caveat list previously said did not. That section has been rewritten rather than removed, because what the ranking leaves out is more interesting than its arrival.

The headline changed from five days and three leaders to six and four. That is the fourth rewrite in five days, and the reason is the same every time: the next report.

The next reading is published. Days seven, eight and nine carry the format to 2 October on a separate page; the six days on this page are left as they were.

Sources

Encounter rates and sample sizes for 24 to 29 September 2026: the BEANS New Division reports for those dates, compiled by @sumamame across all rank bands. Six of the listed decks and the unlabelled remainder are quoted here; the full tallies are the source's own work and are not reproduced. The game counts for 28 September are this site's arithmetic from the published percentages and sample size.
Release date and maintenance window, and DMPP-32 leaving New Division: the calendar, from the publisher's own announcements.
The tier ranking: the Japanese strategy site's All Division ranking page, whose update log records publication on 28 September, and its New Division ranking page, whose log records publication on 30 September; checked that evening. Its top tier and the absence of one deck are cited here; the ranking itself is that site's work and is not reproduced.
Old-format Raid figures: the last reading of the previous format.
What the Baraghiara loop actually does: the set coverage, checked against the publisher's rules FAQ.