The W23 Meta, In Layers
A multi-signal read of the W23 tournament week — eight resolutions side by side, ending in this week’s Strength Index.
An Awakenings feature — a multi-signal read of the W23 tournament week, ending in this week’s Strength Index.
This is an editorial: it makes an argument and lands on a position. The argument is that the W23 meta has structure at several resolutions at once, and that reading any one resolution by itself — even the most-cited one, win rate — misses things visible in the others. The position is that the Archive’s job is to publish those several resolutions side by side, week after week, and to mark which of them are doing real work and which of them are noise.
Where the games happened
ISO 2026-W23 saw the Archive log 27 events of five-plus rounds across 925 players and 2,561 games. Stratified by event size:
The 10–25-player bracket (13 events, 576 games) is the most numerous slice of the field but the smallest by volume. The 26–50 bracket (11 events, 1,056 games) sits in the middle. The 51–100 bracket (3 events, 929 games) is the smallest by event count but produces almost as many games as the middle bracket because the events are larger and longer. The 100+ bracket is empty this week — the two largest events on the schedule clocked in at 99 and 98 players, just below the line.
This stratification is worth doing because the question how does the data behave at different event sizes? is one the rest of the analysis will keep brushing against. The encouraging answer in W23 is that the texture of games is remarkably stable across the buckets. Average per-game VP margin is 27.6 in the 10–25 bracket, 27.9 in 26–50, and 28.0 in 51–100 — within a half-point of each other. Draw rates are 2.78%, 2.94%, and 3.01% — within a quarter-point. Whatever else is true about a five-round event at the local FLGS versus an eight-round GT with ninety-nine players, the games themselves are not, on these summary statistics, visibly different. That is a small result, but it is the kind of small result the article will lean on later: the same per-faction read holds against the same per-game texture whether the bracket is small or large.
What is different across the brackets is everything that isn’t a per-game statistic — who shows up, what they bring, how seriously they are trying to win the event. The win-rate stratification at that level is a question the Archive doesn’t yet have enough events at the largest sizes to answer cleanly — three events in 51–100 is too few to publish a per-bracket faction read against, and zero events at 100+ is, by definition, nothing. Worth flagging for W24 and onward.
Win rates and confidence intervals
The W23 Infinite Archive pull on 2 June carries 5,122 games across 28 factions; the field’s weighted-average win rate is 48.53%, drawn down slightly from neutrality by the 3% draw rate sitting in every game’s denominator.
The top is dense. Chaos Space Marines at 56.57% on 350 games anchors it; behind them, Drukhari (54.29% / 105), Emperor’s Children (54.02% / 261), Leagues of Votann (53.57% / 112), Deathwatch (52.94% / 68), Orks (52.70% / 222), and T’au Empire (51.49% / 268) all sit above the field by enough to notice. The bottom, by symmetry: Grey Knights 33.67% on 98 games, Imperial Agents 37.5% on a vanishing 8 games, Space Wolves 39.84% on 123, Dark Angels 42.07% on 145, World Eaters 42.69% on 171.
Two things to mark before reading any of these too hard.
First, every Wilson 95% confidence band overlaps the next two or three factions’ bands above and below. The same one-week-of-noise caveat that Six Bins and Swiss Isn’t Random keep flagging applies here unchanged. The ordering is what it is; the spacing between any two adjacent factions is not yet large enough to publish a between-factions difference at 95% confidence on one week of event-pull data. The bands narrow as games accumulate; for several weeks to come, the disciplined read is the band, not the centre point.
Second, the smallest samples publish the loudest swings. Imperial Agents at 37.5% on 8 games is not a number, it is a placeholder. Grey Knights at 33.67% on 98 is closer to a number but still has a ±10-point half-width. The two are visually striking and informatively thin. The competitive 40k scene reads weekly stats from many sources, and almost all of them present these factions’ numbers with the same fonts the high-N factions’ numbers get. The Archive’s ambition is to present them with the same caveats they need — desaturated bars on Figure 2 mark them visually; in prose, they are flagged as small-N each time they are cited.
Scoring and the texture of games
Win rate counts wins. It does not count how much. The W23 matchups data — every game’s VP a, VP b, faction a, faction b — lets us look at how much, and the answer is worth a section.
Average VP margin per game (their own VP minus the opponent’s, across every game played, restricted to factions with at least 100 W23 games):
- Highest: Orks +6.64, Drukhari +6.13, Chaos Space Marines +5.51, T’au Empire +5.27, Emperor’s Children +4.10.
- Lowest: Space Wolves −6.35, Adeptus Custodes −5.45, Adepta Sororitas −5.43, Space Marines −4.33, Dark Angels −4.16.
Two of those tops (Orks and Drukhari) are factions whose W23 win-rate read sits a few points above field but doesn’t tell you what kind of games they are winning. The VP margin says: when they win, they are winning by more than the field average; when they lose, they are losing by less. That texture is different from a faction that wins half its games narrowly and loses half its games narrowly, even if the win rate looks the same.
The blowout panels say the same thing more sharply. Define a blowout as a 40+ VP margin (the field’s median per-game margin is 23; the 75th percentile is 42, so 40+ is meaningfully past the typical game’s ending) and ask which factions win or lose those games most often relative to their total games played:
- Most likely to blow out an opponent (% of all games won by 40+ VP): T’au Empire 23.1%, Orks 22.1%, Emperor’s Children 19.5%, Chaos Space Marines 17.1%, World Eaters 17.0%.
- Most likely to be blown out (% of all games lost by 40+ VP): Space Wolves 18.7%, Adepta Sororitas 18.7%, Adeptus Custodes 18.1%, World Eaters 18.1%, Aeldari 17.3%.
The single most informative finding here is the double-presence: World Eaters appear on both lists. Roughly one in six of their games is a 40+ VP blowout they delivered; roughly one in five is a 40+ VP blowout they took. That is what a high-variance, narrow archetype looks like in the data. They either crush their opponent or get crushed. The W23 win-rate read (42.69%) compresses that into a single number, but the games it describes are visibly different from, say, Space Wolves’ W23 games (39.84% WR) — both lose more than they win, but only one of them is winning the games it does win by substantial margins.
The bottom of the VP-margin list is the more troubling read. Space Wolves, Custodes, Sororitas, Space Marines, and Dark Angels are not just losing more games than they win; they are losing them by larger average margins than the field. That isn’t yet a separation result — the bands overlap — but it is a consistent enough finding across two metrics (WR and VP margin) to flag.
Who moved between weeks
The W22 read is now an event week ago. W22-to-W23 movers, restricted to factions with at least 50 games in both weeks:
- Top three up: Leagues of Votann +6.63 points (46.94% → 53.57%), Adeptus Mechanicus +5.98 (44.70% → 50.68%), Orks +5.54 (47.16% → 52.70%).
- Top three down: Grey Knights −11.85 (45.52% → 33.67%), Genestealer Cults −7.62 (52.21% → 44.59%), Space Wolves −5.29 (45.13% → 39.84%).
As before, every one of these movers — up or down — has a W23 Wilson band that overlaps its W22 band. The shifts are large enough to be eyebrow-raising; they are not yet large enough to be separation results. The same disciplined read from W22 still applies in W23: the bands are narrowing, the ranking is shuffling, no single faction has cleanly separated from its prior week’s read. The Grey Knights −11.85 swing is the loudest in the field this week and still consists of two reads (W22 N=134, W23 N=98) whose CIs touch.
Who’s actually being played
Field share — each faction’s portion of total tournament games this week — is its own signal. It tells you which armies the competitive scene is choosing to bring before any game is played.
| Faction | Field share | vs 1/28 baseline |
|---|---|---|
| Space Marines | 10.39% | 2.91× |
| Chaos Space Marines | 6.83% | 1.91× |
| Astra Militarum | 6.68% | 1.87× |
| Tyranids | 5.74% | 1.61× |
| Necrons | 5.25% | 1.47× |
| T’au Empire | 5.23% | 1.47× |
| Emperor’s Children | 5.10% | 1.43× |
| Death Guard | 4.78% | 1.34× |
| Adeptus Custodes | 4.63% | 1.30× |
| Orks | 4.33% | 1.21× |
| Imperial Knights | 4.02% | 1.13× |
| Thousand Sons | 3.63% | 1.02× |
| World Eaters | 3.34% | 0.93× |
| (and 15 more, descending to Imperial Agents 0.16%) | ||
Space Marines dominate, run at 2.91× the equal-share baseline, and they’re closer to a sixth of the field than to a typical faction’s share. CSM, Astra Militarum, Tyranids, Necrons, T’au, EC all sit above 1.4×. The bottom of the field — Imperial Agents, Deathwatch, Adeptus Mechanicus, Genestealer Cults — sits at 0.4× or lower. These are not “underplayed” in any deficit sense; the scene chooses what to bring, and the popularity profile is one of the things competitive 40k as a hobby produces.
Worth flagging: representation is not win rate. Space Marines, the most-played faction by a substantial margin, finished W23 at 47.74% (below field). Astra Militarum, also heavily over-represented, sat at 46.20%. Sample size is large enough that those reads are real reads. The two most-played factions in W23 are both under field by a couple of points. The conclusion is not Space Marines are weak — they are not — but rather that representation and performance can comfortably move in opposite directions, and have done so this week.
Non-performance signals
The Archive carries two interest signals — Google Trends search index and YouTube view counts — and one Practice signal (community tier lists). The W22 Attention capture is the most recent of each; W23 Attention will publish later this week.
Across the W22 Attention window, the search-interest leaderboard had World Eaters (94 / rank 2 of 29), Death Guard (88 / rank 10), and Chaos Space Marines surprisingly low (74 / rank 22). YouTube views inverted the picture at the top: CSM at 1.88M views (rank 2) leading, Death Guard at 551k (rank 9), and World Eaters at the bottom of the ranked field (55k / rank 26). Search measures who is being looked at; YouTube view counts measure who is being watched; the two are different questions and they produce different orderings.
Practice — the Breaking Heads tier list of 15 May, anchored to a German Super Major — currently puts Chaos Space Marines, Astra Militarum, Thousand Sons, and Emperor’s Children at S; Death Guard at A; World Eaters at C. The Practice signal moved faster than the W22 Performance signal did (this is the subject of the Defiler-Five companion section below), and its ordering is the closest single-week predictor of the W23 Performance read among the signals the Archive publishes.
The Defiler-Five, briefly
The most-discussed Chaos balance change of the last six weeks — the MFM v4.1 Defiler legalisation — is worth a short standalone read, because it is the cleanest worked example in the current data of why multi-signal reads matter.
The five Defiler-eligible factions finished W23 at five visibly different places: Chaos Space Marines 56.57% (still top of the field), Emperor’s Children 54.02% (firmed up further from W22), Thousand Sons 50.54% (cooled from W22’s 52.20%), Death Guard 50.20% (rebounded from W22’s 47.29%), World Eaters 42.69% (continued to slide). On the Strength Index below, three of them are STRONGER, one is AVERAGE, one is WEAKER.
The cross-source check on the Defiler-five — Archive (5+ round events), Warpfriends (20+ player / 5+ round events), and 40kstats (open BCP pool, no event-quality floor) — sharpens the point. The Archive’s W23 split is mirrored almost exactly in Warpfriends. In 40kstats’s much larger and much less filtered pool, the five cluster tightly around 50% — CSM 51.06%, World Eaters 49.84%, and everything in between within four points. The split lives at the competitive tail. Open the event-quality filter and most of it disappears. The full version of this argument, with figures and the cross-source progression, lives in the companion piece The Defiler-Five, Read Twice.
The methodology point that travels: factions are not interchangeable recipients of unit changes. Whatever the Defiler did landed inside five different rosters, against five different detachment lineups, played by five different player populations, and in five different parts of the meta. The result has been five different arcs through the patch window. The unit-blame story worked for one weekend in the W21 dispatch; two weeks of cross-source data have refused to honour it.
This week’s Strength Index
The Archive’s Strength Index is its composite — a single per-faction z-score combining Theory (datasheet-roster-derived archetype score), Performance (the multi-source win-rate composite), Practice (tier-list placements), and Attention (search + YouTube). The site weights them 25 / 40 / 25 / 10. The computed result for W23, plotted with the heuristic ±0.5σ band cuts:
The W23 split is 6 STRONGER, 16 AVERAGE, 6 WEAKER.
- STRONGER (composite z ≥ +0.5): Chaos Space Marines (+1.159), Emperor’s Children (+0.946), Thousand Sons (+0.693), Deathwatch (+0.583), Necrons (+0.546), Drukhari (+0.506).
- WEAKER (composite z ≤ −0.5): World Eaters (−0.537), Adepta Sororitas (−0.544), Space Wolves (−0.577), Genestealer Cults (−0.719), Grey Knights (−1.264), Imperial Agents (−1.398).
- The middle 16 — from Death Guard (+0.489) at the top of AVERAGE down to Dark Angels (−0.496) at the bottom — are the bands that this week’s data cannot meaningfully separate.
Two things to mark about the SI as a read.
First, the bands are heuristic, not statistical separation. A composite z of +0.52 is “above the gold dashed line”; a composite z of +0.48 is “below” it. The line is drawn to give the eye a reference, not to claim a difference between those two factions. The data does not currently support that claim; the bar lengths inside each band can be read, but the band labels are a rough framing.
Second, the SI is most useful as a multi-signal agreement check — when a faction is in STRONGER on all four input z-scores, that is a much firmer reading than when its composite leans STRONGER because one component is strong and the other three are mid. CSM, for example, lands STRONGER on Performance, Practice, and Attention’s YouTube side; its Theory z-score is modest. Drukhari lands STRONGER on Performance with a Practice and Attention near zero — the bar is real, but it is one strong signal carrying a composite over the line. The site’s full Strength Index page surfaces the per-signal contributions for exactly this kind of audit; this week’s printout above mirrors what readers will find there.
Closing
The shape of W23, in one paragraph: 27 events, 925 players, 2,561 games, no events of 100+ players this week. The field’s weighted average is 48.53%. CSM continues to anchor the top of every event-quality-filtered signal. Three of the Defiler-Five sit in the STRONGER band, one in AVERAGE, one in WEAKER. The biggest movers are large but none has separated cleanly from W22 yet. The most informative non-WR finding of the week is the World Eaters double-presence on the blowout panels: they both deliver and absorb 40+ VP margins at top-five rates. The most informative representation finding is that the two most-played factions in the field finished below field average. And the Strength Index says, when asked to weigh all four signals at once, the field stratifies cleanly into six STRONGER, sixteen AVERAGE, six WEAKER — a 6-16-6 split that the Archive expects will firm or shuffle as the next several weeks of data come in.
We will keep publishing the layers and keep marking the difference.
— The Editor
Sources: Infinite Archive performance pull for ISO 2026-W23 (BCP, 5+ round events, snapshot 2 June 2026). W22 cross-source: Warpfriends 27 May, 40kstats 26 May. W23 cross-source: Warpfriends 3 June, 40kstats 3 June. Per-game VP data from the Archive’s W23 matchups audit. W22 Attention capture (25–31 May): Google Trends index + YouTube counts. Practice tier captures: Breaking Heads (15 May 2026, German Super Major anchor) and Veizla (March 2026). Theory: BSData v10.6.0 roster archetype scores per faction. Strength Index composite computed against the site’s default 25/40/25/10 weights. Methodology companions: The Big Soup Problem, Six Bins, Swiss Isn’t Random, What the Numbers Can Bear. Companion dispatches: Awakenings — The Baseline Week (W21), Awakenings — The Second Week (W22), The Defiler-Five, Read Twice.