Awakenings The Second Week
DISPATCH #003 · PUBLISHED 2026-05-26

The Second Week

Nearly double W21’s field — but every faction’s W22 Wilson 95% interval overlaps its W21 interval. The disciplined read is that not one mover separated cleanly, and the Defiler-five anecdote stops being a blanket.

Issue window: ISO 2026-W22 · pull dated 25 May 2026 · 9,454 games across 28 factions

Last week the Archive published its first proper data dispatch — The Baseline Week — and made one thing very clear: with only a few thousand games on the board, almost any read of the meta was going to come with a caveat the size of the read itself. This week the field nearly doubled. We are now looking at 9,454 games across 28 factions, up from 5,712 in W21. So the natural question for a second dispatch is the disciplined one: which of last week’s stories survived contact with twice the data, and which were just the first thousand pairings making faces in the dark?

The short answer is that the meta is starting to look like a real distribution rather than a set of provisional bullet points — but the bands are still wide enough that no single mover separated cleanly from its baseline week, even the ones that look like they should have.

What W22 looks like

Horizontal bar chart of Wilson 95% confidence intervals for all 28 factions in the Archive's Week 22 performance pull. Bars are sorted by win rate from Chaos Space Marines at the top down to Imperial Agents at the bottom; the field average is drawn as a dashed gold line near 48%. The top three factions are picked out in bright phosphor green, the bottom three in danger red, and several thin-sample bars with half-widths over 10 points are desaturated to flag uninformative N.
The Archive’s Week 22 pull — every faction’s win rate with its Wilson 95% confidence interval. The wide bars are the uninformative ones.

The top of the table held. Chaos Space Marines finished W22 at 58.59% on 611 games, almost exactly the 59.05% they posted last week on 442. T’au Empire sit at 54.17% on 504 (W21: 55.03% / 358). Deathwatch — last week’s 56.67% read on a thin 90 games — landed at 54.50% on 200 this week, which is the kind of mean-revert one expects when a small-N high number gets a second weekend of pairings. Emperor’s Children firmed to 52.64% on 454. The shape at the top is the same shape as W21.

The bottom told a slightly different story. Space Wolves, who finished W21 at 38.89% (N=144) and looked unambiguously bad, are at 45.13% on 277 this week — a six-point move that nudges them roughly into Aeldari / Black Templars territory. Aeldari, not Wolves, own the bottom of the read-the-band group at 41.67% on 264. Imperial Agents are nominally below them at 34.55%, but on N=55 — we return to that below. World Eaters, who sat near the field average last week, dropped to 43.33%. The cellar’s address changed, not its existence.

The field’s weighted average was 48.41% — down a touch from 49.16% in W21. With draws in the denominator, both numbers are consistent with the population property the data has to satisfy. The drift between them is nothing.

The disciplined read: nothing separated

This is the part that is easy to skip and shouldn’t be.

If you take W21 and W22 side by side and ask “for which factions are the two weeks not consistent with the same underlying win rate?”, the answer is none of them. Every single faction’s Wilson 95% confidence interval in W22 overlaps its 95% interval in W21. The biggest movers in absolute terms — World Eaters down 6.67 points, Space Wolves up 6.24, Tyranids down 4.60, Space Marines up 4.57, Thousand Sons up 4.37 — all of them sit inside the band the prior week could have generated by sampling alone.

Eight biggest week-over-week movers, excluding Imperial Agents. For each faction a W21 confidence band and a W22 confidence band are stacked with their sample sizes labelled. Every pair of bands overlaps, demonstrating that even the largest absolute swings sit inside the prior week's sampling envelope.
The eight biggest W21→W22 movers (Imperial Agents excluded as uninformative). Every faction’s W22 band overlaps its W21 band.

That does not mean nothing happened. It means that doubling the games once does not yet give the data the resolving power to call a six-point swing on 250 games the work of the faction rather than the work of the calendar. Six Bins and The Big Soup Problem both say the same thing in slightly different words: at the volumes a single week of competitive 40k produces, the noise floor is structurally tall. The bands narrow as the games accumulate; they do not narrow on the schedule a weekly post would like them to.

The bands are narrower now than they were a week ago. Chaos Space Marines’ half-width is down to roughly four points; T’au’s, Necrons’, Space Marines’ — all four points or less. By the third or fourth weekend of W22 volumes, some of these will start telling a stable story. They are not telling it yet.

The Defiler-five, revisited

In The Baseline Week we noted that the MFM v4.1 patch that legalised the Defiler should, in theory, have lifted the five Chaos factions that can field one — Chaos Space Marines, Thousand Sons, World Eaters, Death Guard, and Emperor’s Children — and added that the lift would not be even because army composition, detachments, and what else got changed in the patch all matter. W22 shows just how uneven that is.

Faction W21 WR (N) W22 WR (N) Δ
Chaos Space Marines 59.05% (442) 58.59% (611) −0.46
Thousand Sons 47.83% (230) 52.20% (318) +4.37
Emperor’s Children 51.42% (282) 52.64% (454) +1.22
Death Guard 49.85% (339) 47.29% (442) −2.56
World Eaters 50.00% (180) 43.33% (330) −6.67

The blanket framing from W21 — “the Defiler-five benefited, not equally” — stops being a blanket in W22. Three of the five sit above the field average; two sit below; one of those two is approaching the bottom quartile. Whatever the Defiler changed for these armies, it changed for them differently, and “they all play the same unit” looks less and less like the variable that matters. The factions themselves, their detachments, and the rest of the patch are doing more of the work.

This is exactly the place the methodology editorials — Big Soup, Swiss Isn’t Random, What the Numbers Can Bear — warn about. A single-unit story is a clean narrative. The data underneath it is rarely clean enough to support the narrative as told. None of the W21→W22 shifts above clears the 95% bar.

A worked example of what not to do

Imperial Agents finished W21 at 54.84% on 31 games. They finished W22 at 34.55% on 55. That is a twenty-point drop and it is, charitably, not interesting. The Wilson half-width on the W22 read alone is roughly twelve points; the W21 half-width was wider still. A faction that goes from one wholly uninformative number to another wholly uninformative number twenty points away is the data telling you it does not have enough Imperial Agents games to say anything yet. The right entry in this column for that faction is N=55, no read. Anything else is storytelling on top of dice.

It is worth marking because the temptation in a weekly post is to write about the biggest swing. The biggest swing this week is the one the data is least able to support.

Coverage and context

Last week’s W21 Awakenings pull from Trends and YouTube is in for context as well. Search Trends came back clean for all 29 canonical factions; YouTube attention is dense on Space Marines, Astra Militarum, Aeldari, Imperial Knights, and Death Guard — roughly the same shape as W20 — and Veizla’s Shuffle Wars III team coverage continues to dominate the video recap feed. The Goonhammer Competitive Innovations column continues to publish weekly. Practice — Breaking Heads (S-tier: CSM, AM, Thousand Sons, Emperor’s Children, anchored 15 May at a German Super Major) remains the most recent tier list of record; the divergence between Practice and the W22 Performance read narrows a little if you weight Thousand Sons up, and widens a little if you weight Astra Militarum up.

Nothing in attention or practice is doing anything that asks the Performance read to revise its priors.

What W22 actually adds

The honest summary of this dispatch is short.

The top is stable: Chaos Space Marines hold, T’au and Deathwatch hold near the top with usable N. The cellar shuffled — Space Wolves out, Aeldari and Imperial Agents in (one of those reads is real, one is N=55). The Defiler-five anecdote does not generalise. And the single sharpest discipline a second week imposes is that everything that looked like a story in W21 still has the same caveats in W22; they have not been earned away by twice the games. They will be earned away by more games, more weeks, and the band-narrowing that goes with both. We will keep posting the bands and keep marking which bands have narrowed enough to say something. Most weeks, the answer will be: not many of them yet.

The Archive is in the business of saying what it can support. The Baseline Week was the first week we could support anything. The Second Week is the first week we have something to compare to. Both are useful. Neither is a verdict.

Until next week.

The Editor


Data sources for W22: Infinite Archive performance pull (BCP, 5+ round events, ISO 2026-W22). Coverage references: Goonhammer Competitive Innovations, Veizla, 40K Fireside, Art of War 40k. Methodology companions: The Big Soup Problem, Six Bins, Swiss Isn’t Random, What the Numbers Can Bear.