PLAY'S · META · 15 OCT 2026

Week three: four decks led, and the one that led most ended lowest

In the format's first week four different decks took the daily lead. In the second, one deck took it six times out of seven. The third week went back to four. Raid took the most days of anyone — three — and on the last day of the week recorded the lowest reading it has had since the format opened. Dragerion, which was not a named slice on any chart before 1 October, set its highest.

What this page covers, and when Seven days: 8 to 14 October 2026, the third week of the set 37 New Division format, 1,575 games. The figures are one person's daily ranked survey, published the following morning; 8 to 10 October were read from the source charts on 11 October, 11 and 12 October on the 13th, 13 October on the 14th and 14 October on the 15th.

This week does not overlap the week two page, which ended on 7 October. Deck columns are family totals where the chart splits a deck across more than one entry, which is the same convention week two used.

Corrected 16 October Four things on this page were wrong when it went up on 15 October, and are fixed above. This note has itself been wrong twice. As first written on 16 October it said the table showed “eleven to thirteen” chart entries, when the range is ten to sixteen; and it said “all twenty-one days of the format now sum to 100.0”, when only the fourteen days read in full do. Both are corrected above. They are recorded here for the same reason the rest of the note exists.

It said “every day in this week’s table sums to 100.0”. The check is run on the whole chart, not on this table; the table shows seven columns out of the ten to sixteen entries a chart carries, so its rows sum to between 78.6 and 88.1.

Running that check properly then found a real gap: 9 October had been read as fifteen slices summing to 97.9, not 100.0. Going back to the source chart showed a sixteenth slice at 2.1 that had been missed. Its label is four characters ending の紋章 and the 604-pixel source image does not resolve the first one, so the figure is recorded and the name is not guessed. It falls outside all seven columns of the table, so no cell on this page changed; the 9 October row was and is correct.

Re-reading the rest of the series on 16 October found the same deck missed on 1 October, at 4.3, and on 2 October, at 3.2. Those two days feed the nine-day page and the week two page. It is not a column on either, and every published cell on both was re-checked against the corrected data: 54 cells and 49 cells, no differences. All fourteen days that were read in full — 1 to 14 October — now sum to 100.0. The seven days before that, 24 to 30 September, were deliberately partial reads of six columns each and have never summed to 100; they are marked tracked in the data file rather than full, and the check does not apply to them.

It said 8 to 12 October were read “between 10 and 13 October”. The earliest of those readings was taken on 11 October, not the 10th.

It said two smaller entries swung “two points rather than eleven”. One of the two swung 4.3 points, not two.

It said the ranked season “ends tonight” in one place while saying it “ended at 23:59 on 15 October” in another. The page went up at about 23:20 that night, before the season closed, so the two sentences contradicted each other. Both now read as past.

No figure in either table changed. The tables were generated from the data file rather than typed, and were re-checked against it cell by cell both before and after publishing.

The week, day by day

Date	Games	Leading named deck	Its share
8 Oct	227	Raid	21.6
9 Oct	238	Raid	14.7
10 Oct	222	HADES Dragerion	18.0
11 Oct	262	Dragerion and Baraghiara, tied	12.6
12 Oct	169	Cotton Zenith	15.4
13 Oct	114	Raid	19.3
14 Oct	343	HADES Dragerion	16.3

Leading named deck is the largest single named entry on that day's chart. The survey's own unnamed bucket is excluded from this column; on 11 October it was larger than either tied deck, and that is covered below.

Four names, and none of them held it

Raid led on the 8th, the 9th and the 13th. Dragerion led on the 10th and the 14th, and shared the 11th with Baraghiara at 12.6 each. Cotton Zenith led on the 12th. That is four names across seven days. The only back-to-back lead was Raid on the 8th and 9th; from the 9th onward the lead changed hands every single day.

Week two was the opposite shape: Raid took six of seven, five of them running. So across three weeks the format has gone four leaders, then one, then four again. Whatever settled in week two did not stay settled.

Raid led the most and finished at its floor

Raid opened the week at 21.6 and closed it at 10.2. That 10.2 is its lowest figure in all twenty-one days of this format. The fall is recent rather than long-running. Raid was at 22.0 on 1 October and 23.2 on the 5th, and the 21.6 it opened this week on is an ordinary figure for it — the fourth highest of its twenty-one readings. It is the 10.2 at the other end of the week that has no precedent.

Across the week it went 21.6, 14.7, 14.4, 10.7, 13.0, 19.3, 10.2 — down three days, up two, then down to the floor. It still led three of those days, which says as much about where the rest of the field was sitting as it does about Raid.

Dragerion set its high, and its second high

Dragerion reached 18.0 on the 10th, its highest reading of the format, and 16.3 on the 14th, its second highest. Both fall inside this week.

It is worth being exact about where this deck came from, because it is easy to overstate. Dragerion does not appear as a named slice on any chart before 1 October, when it read 4.9. That is not the same as the deck not existing: anything below the survey's cutoff sits inside the unnamed bucket rather than being absent from the format. What can be said is that it was not large enough to be named, and that thirteen days later it set the high.

The piece on Dragerion taking the lead was written on the 11th, from the 10th's reading. That lead lasted one day outright, before the tie on the 11th and Cotton Zenith on the 12th. The page has not been rewritten; this is the correction to it.

Cotton Zenith put its high and its low in the same week

Cotton Zenith read 4.8 on the 8th and 15.4 on the 12th. Those are its lowest and its highest readings of the entire format, four days apart. It led the 12th on that 15.4 and was back to 11.1 by the 14th.

Of the seven columns in the table below, Cotton Zenith is the only one whose format high and format low both fall inside this week. Two smaller entries outside the table do the same thing, across 4.3 and 2.1 points against Cotton Zenith's 10.6.

It is not, however, the widest-swinging column of the week. Raid moved 11.4 points between its own high and low, and the unnamed bucket moved 11.4 as well, against Cotton Zenith's 10.6. Three columns moved by more than ten points in seven days, which is the thing to take from this week rather than any one of them.

Every tracked deck, every day

Date	Raid	Jokers	Baraghiara	Cotton Zenith	HADES Dragerion	Magic Tools	Other
8 Oct	21.6	13.7	14.1	4.8	9.3	7.5	12.7
9 Oct	14.7	10.1	12.6	8.8	11.3	13.9	7.2
10 Oct	14.4	11.3	11.8	5.9	18.0	6.8	14.1
11 Oct	10.7	10.3	14.9	11.8	12.6	9.2	18.6
12 Oct	13.0	9.5	9.5	15.4	13.6	9.5	14.7
13 Oct	19.3	13.2	8.8	10.5	11.4	10.5	12.3
14 Oct	10.2	10.2	11.1	11.1	16.3	9.0	14.7

Baraghiara and Magic Tools are family totals, since the chart splits each across more than one entry on some days. Baraghiara is split on the 9th, 10th and 11th; Magic Tools on the 9th, 12th, 13th and 14th. Other is the survey's own unnamed bucket, printed as it appears.

On the 11th the largest slice had no name The unnamed bucket read 18.6 on 11 October, more than either of the two decks tied at 12.6. It was the largest slice on the chart that day. The leader column above reports the largest named deck, which is why the 11th shows a tie at 12.6 rather than Other at 18.6.

The same day shows why families and entries are not interchangeable. Baraghiara's family total was 14.9, above both tied entries — but that total is one 12.6 entry plus a separate 2.3 entry, and the tie is between single chart entries. Read as families the day looks like a Baraghiara lead; read as the chart labels it, it is a tie. Neither reading is wrong, and they are not the same statement.

The sample size moved more than the decks did

The survey covered 114 games on the 13th and 343 on the 14th. That is three times as many games in one day. The 114 is the smallest full day in the twenty-one; the 343 is the second largest, behind 386 on 27 September.

The ranked season HADES・ドラゲリオンカップ ended at 23:59 on 15 October, so the 14th was the second-to-last day of it. Players pushing for a rank before a season closes is the obvious reading of a jump like that, and it may well be the right one — but the survey publishes game counts, not reasons, and nobody has said so. It is recorded here as what it is: the two days sit next to each other.

It matters for reading the rest of the page. A 2.3 on the 14th is eight games; a 2.6 on the 13th is three. The small days are where single-digit entries appear and vanish, and the 13th is the smallest of them all.

What the fourth week changes

That season has now ended, which resets the ladder everyone on these charts is climbing. The next set is due in November, and the anniversary set after it changes what is legal in New Division at all. Until then the thing to watch is whether Dragerion holds the high it set twice this week, or whether week four goes back to four names again.

Where these numbers come from A single player's daily ranked survey, published as pie-chart images the following morning. Each figure here was read off the chart by hand, and the check is that the whole chart for a day sums to 100.0. All seven days in this week now pass it. The table above is a subset of those charts — seven columns out of the ten to sixteen entries a chart carries — so its rows run between 78.6 and 88.1, not 100. That check catches a misread digit. It does not catch two labels being swapped, so the deck names carry more uncertainty than the percentages do.

It is one survey of a few hundred games a day, not the publisher's own data, and it is not a tournament field. How the data works sets out what it can and cannot support.