ContentsWhat this ranks, and what it does not

What this ranks, and what it does not How we counted Where the ranking comes from Tier list The three that climbed, and the one that didn't The Digtoise problem Purple, and last month's question answered The corpus lost 33 decks, and it moved two rows Red's workshop is a deck now The batch, one last time What to watch in week 38

The Palworld TCG board for week 37: three cards climbed, and the one to watch didn't move

The Palworld TCG board for week 37, built from 1,537 complete public decks created between 8 August and 6 September. Three cards climbed, Reindrix – Icy Gaze into S and Jormuntide Ignis and Pyrin Noct into A, and Aurora Guide fell off the bottom. The real story is Digtoise – Seismic Drillback, third

Last month's board had three cards on it that moved, and every one of them moved down. This one has four changes and three of them are promotions. Reindrix – Icy Gaze (ETD01-016) is in the top band for the first time. Jormuntide Ignis (EBP01-001) is back in A after a month away. Pyrin Noct – Steed of Azure Flames (EBP01-077) joins it there. Aurora Guide (EBP01-048) slipped off the bottom edge by a whisker.

None of that is the interesting part. The interesting part is a card that didn't move at all, sitting third in A, which the most recent week of building suggests is on its way out of the format entirely. We have spent most of this run trying to talk ourselves out of that finding and have not managed it.

What this ranks, and what it does not

There is still no competitive record for this game. The publisher's Deck Recipe page has a Featured Decks section and a Tournament Decks section, and on 7 September both say the same three words they have said every time we have looked: No entries found. The events page still puts the first Challengers Cup season in October and the road to Masters League 26/27 in November. Shop tournaments have been running since August. Nobody publishes what happens at them.

So this is not a results ranking. It is a census of what people are building, and the difference matters. A card everybody plays might be a card everybody is wrong about, and this measure cannot tell you which. It can tell you what the room believes, precisely, with a sample size attached.

Two lists turned up on 6 September with placements typed into their titles, one claiming first place and one third at the same store event. They rank nothing here. There is no field size, no report, no way to check any of it, and a text field is not a results feed. We mention them only because the one claiming to have won is a Red/Purple deck running four Mounted Machine Gun and four Primitive Furnace, which is the engine the numbers below have been watching get built for a month. Nice when the anecdote and the data point the same way. Still an anecdote.

A cozy game room features a wooden table set with board games, dice, and snacks, surrounded by chairs and bookshelves. A lamp illuminates the scene.

How we counted

The sample is 1,537 complete public decks created between 8 August and 6 September 2026 (UTC), meaning exactly 50 main-deck cards plus 10 souls. They come from the palworldtcg.gg public API, the upstream of the community lists we import and serve on our own deck pages. Of 2,118 public decks on that site when we measured on 7 September, 1,560 were created inside the window and 23 of those were unfinished.

That window is the same rule as last month's board, rolled forward: the thirty whole days ending the day before the period opens. It also means this board and the last one share twenty-three days of decks, so treat any small movement here as a nudge rather than a swing. Where a change is big enough to be worth a paragraph, we have gone and looked at the weeks underneath.

One genuinely new thing this time. ISO week 36 is closed, and it is the first week in this series we have been able to census properly. It holds 219 complete decks, spread evenly at 37, 39, 37, 27, 34, 19 and 26 across its seven days, with 211 distinct decklists among them and no minute containing more than two creations. No bulk import, no burst, no batch. Every board we have run since 24 August has been built on a window with a bulk import sitting inside it, and now there is a clean week to check the board against. That check is most of the rest of this article.

The legal pool is 148 distinct main-deck cards plus the Soul, taken from the publisher's card list. That list runs to 256 rows once every printing is counted and folds to 148 once alternate art and the nine non-Soul PR2026 promos are matched to the cards they reprint. Unchanged from last month and the month before: no new card has entered the format. The publisher's products page puts the next booster, Legends Awaken, on 30 October, with two Eternal Ascent trial decks on 18 December. Every one of the 148 appears somewhere in this census. The loneliest is Direhowl – Proud Fang (EBP01-099) on seven lists.

Colour identity comes from the cards rather than the label the builder typed; 22 decks disagree with their own tag. Blue is in 958 of the 1,537 decks (62.3%), Purple 751 (48.9%), Red 699 (45.5%), Green 520 (33.8%). One deck in ten is mono-colour, and half of those are Purple.

One limitation to keep in your pocket. The 1,537 decks contain 1,418 distinct decklists, so roughly one in thirteen is an exact card-for-card copy of another list in the same window. This is a census of lists, not of builders, and a popular deck that gets cloned counts every time.

Where the ranking comes from

Same measure as the last four boards, deliberately. For each card we take its in-colour inclusion rate: of the decks that could legally play it, what share did. Colourless cards are eligible everywhere. Bands sit on plain fractions, S at two decks in three, A at one in two, B at two in five, C at one in three, and inside a band the order is the order of raw share, ties broken by card number. Freezing the measure is the only thing that makes "moved a band" mean anything at all. Change the maths and every movement becomes something you invented.

Thirty-three of the 148 cards clear one deck in three. Red puts nine cards on the board and the other three colours put eight each, which is very nearly last month's too-neat symmetry making a comeback. The seven Colourless cards again contribute nothing. The best of them, Elphidran – Gentle Radiance (EBP01-098), is in 18.9% of all 1,537 decks despite being legal in every one. Cards anyone may play remain cards nobody does.

Tier list

Eight in S, seven in A, thirteen in B, five in C. No band was padded and none came out empty. Foxparks – A Toasty Hug (ETD01-004) is in 591 of 699 Red decks and remains the most agreed-upon card in the game, now by a wider margin than last month: 3.8 points clear of second place where it was 1.7 clear before. The trial decks keep punching above their weight. TD01 and TD02 supply 48 of the format's 148 cards and hold ten of the thirty-three board seats, five of them in S.

The three that climbed, and the one that didn't

Reindrix – Icy Gaze (ETD01-016) is the headline promotion, into S at 68.5% of 958 Blue decks, and it is worth being precise about how it got there, because the answer is mostly the calendar. Reindrix was in 51.5% of Blue decks during 1 to 7 August, the week this window drops. Losing that week lifts its thirty-day average more than anything that happened in the last seven days did: in week 36 itself it reads 67.6%, down from 75.9% the week before. So it is a real S-band card by our measure and a card that had a slightly quiet week, at the same time. Both things are true and the board only shows you one of them.

Jormuntide Ignis (EBP01-001) returns to A at 50.07% of 699 Red decks. That is 350 lists where 350 is the minimum, which is to say it cleared the line by a single deck. We demoted it last month by a margin nearly as thin. A card oscillating across a band boundary by one or two lists is not telling you anything about the format, and we would rather say so than dress a wobble up as a trend. Read it as a card that lives on exactly one Red deck in two.

Pyrin Noct – Steed of Azure Flames (EBP01-077) is the promotion with an actual story. Across weeks 32 to 36 it reads 43.1%, 46.2%, 52.9%, 52.1% and 57.8% of Purple decks, up in three of those four steps and up nearly fifteen points end to end, and it is now Purple's second most-played Pal behind Blazehowl Noct – Darkflame Defender (ETD02-017). In a colour that has spent a month shedding cards, one seven-cost Pal has been walking the other way the entire time.

Aurora Guide (EBP01-048) drops off the board at 33.2%, two decks short of the 320 it needed. It is the second month running that a card has left by a margin you could fit in a lunch break, after Maraith went by four decks in August. The bottom of C is a turnstile and we would not read much into either exit.

A person sorts trading cards on a table, with a steaming mug, a croissant, and stacks of books nearby.

The Digtoise problem

Digtoise – Seismic Drillback (EBP01-050) is third in A on this board, in 281 of 520 Green decks, 54.0%. On the 219 decks of week 36 alone it is in 24 of 78 Green decks, 30.8%, which does not clear C. Two other cards also fall off a week-36-only board, Victor's Strategy (EBP01-046) and Blazehowl – Hellflame Defender (EBP01-014), but both of those were already in the bottom half. Digtoise is the only card that goes from A to nowhere.

The week-by-week line inside Green decks is 50.2%, 54.0%, 62.0%, 59.3%, then 30.8%. A drop of that size in a single week, on a sample of 78, is worth several attempts to explain away, so here are the three we made and what happened to each.

The mix changed. Green's partner colours shuffled between weeks 35 and 36: Green/Purple fell from 46 decks to 23 while Green/Red rose from 15 to 21, and if Digtoise lived mainly in the pairing that shrank, the fall would be an accounting artefact rather than a decision. This is exactly the trap that ate last month's Purple story, so we tested it first. Standardising week 36 to week 35's pairing mix gives 27.5%, which is lower than the raw number, not higher. And it falls in every pairing: Blue/Green 61% to 28%, Green/Purple 57% to 17%, Green/Red 67% to 48%. The mix is not doing this.

Green is being splashed rather than played. If builders were reducing Green to a few cheap cards alongside a real Blue or Purple deck, a six-cost Green Pal would be the first thing out. Mean Green cards inside a Green deck across weeks 32 to 36: 26.9, 26.5, 27.3, 25.7, 25.2. Essentially flat, and the mean cost of those Green cards did not move either, 4.16 to 4.15. Worse for the theory, when we split Green decks into heavy (20 or more Green cards) and light, the fall is concentrated in the heavy ones, 71.3% to 42.6%, a drop with a z of −3.38. The decks most committed to Green are the decks cutting it.

The Collecting package was cut. Digtoise spends two Ingredient to buff itself and pick up Breakthrough, so a retreat from that whole cluster of cards would drag it out with them. Copies of Collecting-aptitude Green cards per heavy-Green deck: 14.74 in week 35, 14.00 in week 36. Flat. Not that either.

What is actually left is small, specific and slightly odd. Green's heaviest decks tripled their eight-cost count, from 0.54 copies per deck to 1.59, and put their structures back up from 4.83 to 6.07 after three weeks of decline. Mammorest – Roar of the Wilds (ETD02-007), Warsect – Iron Fortress (EBP01-054), Vaelet – Bewitching Blossom (EBP01-055) and Broncherry – Brimming Adoration (ETD02-006) all gained ten points or more. A list like Consistify shows the shape: 28 Green cards, no Digtoise, four Ranch (EBP01-066) and four Breeding Farm (EBP01-063) holding the ground while eight-cost bodies arrive on top.

Digtoise is not alone in this, either. Lily's Strategy (EBP01-070) sits fourth in A at 53.5% and reads 39.7% on week 36, a two-band gap of its own, and its decline started earlier: 58.3% in week 34, 41.6% in week 35. Green's board position is the most out-of-date thing on this page.

So the honest summary is that heavy-Green decks appear to have swapped a midrange card for a top end, that we can see the swap and not the reason, and that 54 heavy-Green decks in one week is not enough to name a cause. Digtoise stays in A because the measure is frozen and the window is thirty days, and that is the correct thing for the board to do. If week 37 reads anything like week 36, it will not be in A next month, and you will have known about it here first.

Purple, and last month's question answered

Last month we found that Purple's apparent collapse was mostly a denominator trick. Mono-Purple decks, which run about fifty Purple cards and therefore touch nearly every Purple card in the game, had stopped being built, and their absence deflated every Purple inclusion rate at once. We flagged that the whole reading rested on a single thin week and asked for a second one before anyone built on it.

Here is the second one. Mono-Purple decks as a share of all complete decks, weeks 32 through 36: 4.7%, 7.8%, 6.1%, 2.6%, 2.7%. The population halved and has now stayed halved. Purple's inclusion rates have reset downward and should be read at their new level rather than treated as a slide in progress.

The four cards we said survive the control mostly kept going. In two-colour Purple decks, Dark Cannon (EBP01-096) went 20.2% to 12.6% and Maraith – Twilight Messenger (EBP01-085) 17.4% to 12.6%, both continuing to fall with the mono-Purple effect held constant. Katress – Abyssal Sorcerer (EBP01-075) drifted from 11.8% to 10.7%. Tombat – Out of Nowhere!? (EBP01-076) bounced, 10.1% to 14.6%, so call that one unresolved. Three of four holding up is a decent return on a control.

One correction, because it is ours to make. Last month's article reported four mono-Purple decks in week 35. Running the same query on today's corpus returns ten. We cannot reconcile those from here, and the reason is the next section.

The corpus lost 33 decks, and it moved two rows

Last month we noted that this corpus is mutable, having found five decks arriving retroactively inside a closed window. This month it went the other way, and further.

Week 36's article reports 1,845 decks created between 1 and 30 August at the moment it measured. The same closed window on today's corpus holds 1,812. Thirty-three lists that existed on 31 August are not there now, deleted or made private, and that is 1.8% of the sample the last board was built on.

It is enough to move rows. Comparing this board to the published week-36 one gives six differences, but two of them appear as soon as you recompute the old window on today's data: Leezpunk – Treasure Bandit (ETD02-014) reads A on the published board and B when its own window is recounted, and Wumpo Botan – Tropical Sentinel (EBP01-062) leaves the board the same way. Those two are the deletions, not the week, and we have not counted them as movement.

The practical version: a figure quoted from a closed window is accurate as of when it was read, not true forever. We are going to start saying the measurement date next to every number that matters, which is why this piece does.

A cluttered desk with stacks of files and an open photo album, beside a cup and a globe, illuminated by window light.

Red's workshop is a deck now

The one prediction we made last month held. In two-colour Red decks, Mounted Machine Gun (EBP01-015) went from 32.9% in week 35 to 43.2% in week 36, and Primitive Furnace (EBP01-016) from 50.9% to 54.7%. Furnace makes the board at 47.6% across all 699 Red decks. Machine Gun still does not, at 30.3%, because weeks 32 to 34, when it sat in barely one two-colour Red deck in four, are still in its denominator.

What makes it an archetype rather than two cards trending is that something is paying for them. Pump-Action Shotgun (EBP01-020), which has been in S on every ranked board we have published, fell in those same two-colour Red decks from 82.7% to 72.6%, and Single-Shot Rifle (ETD01-010) fell from 72.4% to 56.2% across all Red decks. Both of the cards on the way out are Gear, at seven cost and four. Both of the cards coming in are Structures, at two and four. Red is swapping things it spends for things that sit on the board and pay, and that is a deck making a decision rather than a drift.

Blue MMG is the clean version of it, four Furnace and four Machine Gun with twenty Blue cards behind them. The Red/Purple machine gun list is the greedier one, cutting Pump-Action Shotgun to two copies to make room. If you want one thing to build this week, it is this, and it is also why Single-Shot Rifle sits sixth in S on a board whose most recent week puts it fifteen points lower.

The batch, one last time

The 24 August bulk import is still in the census: 119 of the 1,537 decks in this window, 7.7%, created inside a single minute and tagged for a tournament whose results nobody published. We have kept it in for three boards now, on the grounds that carving a group out of a census after the fact turns a measurement into an opinion, and we say so every time.

Its grip is loosening on its own. Removing it changes exactly one row, sending Jormuntide Ignis back to B, against three rows last month. The colour skew it introduces is now under one percentage point on every colour, against 4.4 points last month. As the window rolls past 24 August it will stop mattering entirely, which is the tidiest possible way for this problem to end.

The only new arrivals in that group since we last looked are copies of the list claiming to have won it. The description naming that player is now attached to five separate decks, the newest dated 31 August and 3 September, and the exact fifty cards appear six times across the whole corpus. It is not the most-copied list we hold, which is a Red/Blue Chillet deck sitting on nine identical entries, but it is the only one whose copies keep a placement claim stapled to them as they spread.

What to watch in week 38

Digtoise, obviously. One more week at week-36 levels turns the Green section above from a curiosity into a demotion, and if it recovers instead we will have learned that a 78-deck week can be wrong by twenty points, which is worth knowing about every other figure in this piece.

Then Single-Shot Rifle, for the same reason in a different colour. It is sixth in S on the thirty-day board and 56.2% on the clean week, and Red's engine build is the thing eating it.

Then the Deck Recipe page. The moment it stops saying No entries found, this series changes from a census of what people build into something closer to journalism, and every band on this board becomes a hypothesis to check rather than a finding to report.

Dawn of Palpagos landed on 30 July. Five and a half weeks in, the format has picked eight cards it will not argue about and is having a genuine fight over everything else. The fight is the good bit.