Every field on a Steam Workshop listing has now been measured by this series except one, and it is the field that sits at the top of every card, in the largest type, on every surface the platform has: the title. Not its length — that was the short-title problem — and not whether it is unique, which was the name collision. What nobody has asked is what it is written in.
The Workshop is a global catalogue with a nominally English interface, and its creators title their work in whatever script they think in. So we classified all 5,990,372 active, subscribed Workshop mods by the writing system of their name, and split the non-ASCII ones into four groups. They behave nothing like each other.
The four populations
| Script of the title | Mods | Share of catalog | Share of all subscriptions | Reached 1,000 subs | Median title length |
|---|---|---|---|---|---|
| ASCII only | 5,389,331 | 89.97% | 87.13% | 13.41% | 19.6 chars |
| CJK only | 235,989 | 3.94% | 4.93% | 21.50% | 8.7 chars |
| CJK + Latin | 142,721 | 2.38% | 5.89% | 41.14% | 26.7 chars |
| Latin extended (accents) | 101,195 | 1.69% | 1.74% | 15.48% | 27.5 chars |
| Cyrillic only | 83,378 | 1.39% | 0.10% | 2.44% | 14.6 chars |
| Cyrillic + Latin | 37,758 | 0.63% | 0.22% | 10.77% | 30.4 chars |
Read the two bold rows against each other. A title written only in Cyrillic reaches a thousand subscribers 2.44% of the time. Add a Latin version of the same name alongside it and the rate is 10.77% — 4.4x. Both rows are Russian-language mods by Russian-speaking creators in the same global catalogue. The only difference is whether the name carries a second script.
Now read the CJK rows, which look like the bigger story and are the more fragile one. A CJK-only title reaches 1,000 subscribers 21.50% of the time. A CJK title with a Latin twin reaches 41.14%. That is 1.91x, the largest pooled gap in the table — and most of it is not the script.
Why the CJK number is mostly an artifact
Two things inflate it, and both are already documented axes.
Title length. CJK-only titles are 8.7 characters at the median against 26.7 for CJK-plus-Latin ones, because a Chinese or Japanese title packs a complete name into a handful of glyphs — 迷你城市立交 is six characters and names an entire motorway interchange. This series already showed that title length is the strongest single predictor in the dataset (the short-title problem). A group of titles that is three times longer will outperform on that axis alone.
Restrict both groups to titles of 10 to 39 characters — the same width band — and the gap narrows:
| Mods | Reached 1,000 subs | |
|---|---|---|
| CJK + Latin, 10–39 chars | 98,450 | 41.14% |
| CJK only, 10–39 chars | 69,349 | 33.04% |
1.91x becomes 1.25x the moment the two groups are the same length.
Wallpaper Engine. It is the largest ecosystem on the Workshop (3.2 million items) and it breaks every rule in this dataset, because almost everything on it succeeds: its reach rate sits near the ceiling. Of its 108,146 mods, 58,990 are CJK-only titles and they reach 1,000 subscribers 93.31% of the time, against 17.02% for CJK-only titles everywhere else. One ecosystem supplies 18% of all CJK-only titles on the platform and a near-perfect hit rate inside them.
Remove Wallpaper Engine and the CJK picture changes shape entirely:
| All games | Excluding Wallpaper Engine | |
|---|---|---|
| CJK only | 21.50% | 11.44% |
| CJK + Latin | 41.14% | 27.38% |
The CJK-only row falls below the ASCII baseline once Wallpaper Engine is out. Give every game one vote instead of every mod, across the 76 ecosystems with at least 200 mods in each group, and the two means are 21.51% vs 15.43% — a 1.39x gap, with 65 games favouring the Latin twin and 40 favouring the pure CJK title. Still positive, no longer a headline.
The control that breaks it
The strongest control this series uses compares the same creator's own mods to each other. Across 2,907 creator-and-game pairs where one creator published at least three CJK-only titles and three CJK-plus-Latin titles in the same game:
| Mods | Reached 1,000 subs | |
|---|---|---|
| CJK + Latin | 47,331 | 51.60% |
| CJK only | 50,476 | 44.09% |
The aggregate still favours the Latin twin — but count the pairs where the two sides differ and 786 creators did better on their pure CJK titles against 754 on their bilingual ones. The paired test does not confirm the aggregate. It splits it.
That is the tell, and the reason is visible in the titles. Tighten the definition of "Latin twin" to require real Latin words outside any bracket, and the effect shrinks again — because a large part of what the loose definition was counting is bracketed metadata, not a translation:
[4K]Audio Visualizer v0.6.6(音频可视化)·Reimu 灵梦（1080P 60FPS）重置版·熠烛 御剑驭龙-红鸾樱落 高度自定义Red Warbler-Sakura falls （Highly customizable）
[4K], (1080P 60FPS) and v0.6.6 are listing-quality markers — the same signal as the bracket effect in the short-title problem — not a language twin. Once they are excluded, the CJK bilingual advantage survives only as a residue of roughly 1.1x to 1.4x, and the paired test is a coin flip.
The Cyrillic result is the one that holds
Nothing about the Cyrillic gap is fragile, and it is measured four independent ways.
Length-matched. Cyrillic-only titles of 10–39 characters reach 1,000 subscribers 2.81% of the time. Cyrillic-plus-Latin titles of the same length reach 9.65%. A 3.4x gap with title length held constant.
Per game. Across the 63 ecosystems with enough of each, the mean reach rate is 12.89% for Cyrillic-plus-Latin against 8.50% for Cyrillic-only, and 45 games favour the Latin twin against 12 that favour the pure Cyrillic title. That is not a composition effect; it is a direction that repeats in three-quarters of the ecosystems where it can be tested.
Against the ASCII baseline, in the same length band, per game. This is the harshest test in the article. Across 59 games with at least 50 mods on each side, both restricted to titles of 10–39 characters:
| Mean of per-game reach rates | |
|---|---|
| Cyrillic only | 4.96% |
| ASCII only | 12.87% |
Fifty-seven of 59 games favour ASCII and two favour Cyrillic. A Russian-language title in the global catalogue, published by the same creators into the same ecosystems, at the same length, reaches a thousand subscribers at less than half the rate of an English one — and there is almost no game where that reverses.
Against a homoglyph control. This is the check that decides whether the finding is real or a detection artifact. Cyrillic and Latin share visually identical letters — А/A, В/B, Е/E, К/K, М/M, Н/H, О/O, Р/P, С/C, Т/T, Х/X — so a title written entirely in Latin lookalikes would be classified as Cyrillic by any script test. There are 3,609 such titles in the catalogue, and if the Cyrillic penalty were a detection error they would carry it too. They do not: homoglyph-only titles reach 1,000 subscribers 10.34% of the time, near the ASCII baseline of 13.41%, against 2.44% for genuine Cyrillic. The penalty tracks the language, not the character table.
Script is not the cause
The mechanism is reachability, and the proof is that the effect appears even where the audience reads the script.
MudRunner is a Russian-origin trucking simulator with a large Russian-speaking Workshop. Its Cyrillic-only titles are the best-performing in the entire dataset: 70.61% reach a thousand subscribers. Its ASCII-only titles: 81.44%. Even inside an ecosystem built by and for Russian speakers, the English title still wins by eleven points.
The dead zones are elsewhere entirely, and they are sandbox and vehicle games whose audiences are overwhelmingly English-speaking:
| Ecosystem | Cyrillic-only titles | Reached 1,000 subs | Its ASCII rate |
|---|---|---|---|
| Scrap Mechanic | 14,572 | 0.40% | 3.26% |
| Terraria | 7,070 | 0.17% | 6.84% |
| Space Engineers | 4,958 | 0.06% | 7.68% |
| Besiege | 2,644 | 0.04% | 3.14% |
| Trailmakers | 2,096 | 0.00% | 0.89% |
| Kerbal Space Program | 4,019 | 0.15% | 1.16% |
Some of these are not bad mods. Су-25 (Грач) — a Su-25 Frogfoot — has 4,033 subscribers in Scrap Mechanic. Пройденный мир Зенит, a completed Terraria world, has 7,345. Скорость света, a light-speed build for Space Engineers, has 3,038. They are functioning mods in ecosystems where nobody searches in the language they are named in.
Which is what the Latin twin fixes, and why the fix is mechanical rather than stylistic. A Cyrillic-only title has exactly one searchable form, and it is one that a large majority of the platform cannot type. A Cyrillic-plus-Latin title has two. This is the same discovery-limited mechanism the series documented in the discovery lottery and the name collision — the loss is at the top of the funnel, and no amount of listing quality recovers a mod nobody can find.
The axis that died
Emoji deserve a paragraph because they look like the strongest signal in the dataset and are the cleanest null result.
Titles containing an emoji reach 1,000 subscribers 63.34% of the time against 21.98% for titles without one — a 2.9x pooled gap, far larger than anything script produces. It is entirely a proxy for Wallpaper Engine, which is both the emoji-heaviest and the highest-performing ecosystem on the platform.
Restrict to well-presented listings and compare the same creator's own mods to each other, requiring at least two of each, and it collapses:
| Mods | Reached 1,000 subs | Creators favouring | |
|---|---|---|---|
| Emoji in title | 4,241 | 67.58% | 173 |
| No emoji | 34,791 | 63.10% | 151 |
| 1.07x | 492 pairs, essentially a tie |
An emoji in a Workshop title does nothing measurable. The axis is retired.
What it means
If you title a mod in Cyrillic, Japanese, Chinese, Korean or any non-Latin script, adding a Latin translation of the name beside it is the highest-return edit available to you. For Cyrillic the measured return is 2.44% to 10.77% pooled, 2.81% to 9.65% length-matched, 45 to 12 ecosystems per-game, and it is the only effect in this article that survives every control this series has. You are not writing a bilingual title for style. You are doubling the number of search terms your mod can be found by.
Do not over-read the CJK version of the same advice. The pooled 1.91x shrinks to about 1.25x once the two groups are the same length, to 1.39x when every game gets one vote, to roughly nothing once bracketed metadata is excluded and the paired test is applied — and the honest number for a Chinese or Japanese title with a real translation beside it is somewhere near 1.2x, not 2x. The advice is still worth taking; the headline is not.
If you are reading a Workshop statistic, be careful with any figure that pools a script against the whole catalogue. Wallpaper Engine alone moves the CJK-only reach rate from 11.44% to 21.50%, and it is a single game. The rule this series has now applied five times holds again: a pooled gap that looks like a finding is usually composition wearing a costume, and the real residue underneath is smaller, narrower, and more useful. Here the residue is Cyrillic, and it is large.
Related reading: the short-title problem on the title-length axis this one has to control for, the name collision on the 977,000 mods sharing a name, the copy-paste descriptions on the blurb field, the discovery lottery on the 94% of mods that get zero views in a week, and the thumbnail divide on the 37% of mods with no preview image.
Method: 5,990,372 Steam Workshop items tracked by ModSignal as of 30 September 2026 across 392 games — every item that is is_active and carries at least one subscription, holding 12.45 billion subscriptions. Script classes are mutually exclusive and assigned from the raw title in this order: CJK (any of U+4E00–9FFF, U+3400–4DBF, U+3040–30FF, U+AC00–D7A3) with a Latin letter present; CJK without; Cyrillic (U+0400–04FF) containing at least one Cyrillic character that is not a Latin homoglyph, with a Latin letter present; Cyrillic without; Cyrillic homoglyph-only; other non-ASCII; ASCII. "Latin twin" in the strict definition means two or more Latin letters surviving removal of bracketed and parenthesised spans ([...], (...), 【...】, （...）), which excludes version numbers, resolution tags and quality markers; the looser definition used in the pooled table counts any Latin letter at all, and the article reports where the two disagree. Titles containing only Cyrillic lookalikes (3,609 items) are folded into the Latin-extended row of the headline table because they cannot be distinguished from Latin by glyph; their reach rate (10.34%) is reported separately in the homoglyph control. "Reached 1,000 subscribers" means subscriptions_current >= 1000 at a single consistent read; it is a point-in-time threshold, not a tracked trajectory. The paired test compares creators who published at least three mods on each side within the same game and reports the aggregate rate on both sides with ties excluded from the win/loss tally. Per-game means are unweighted means of individual game rates, restricted to games with at least 200 (pooled) or 100 (sign test) mods in each compared group. Wallpaper Engine is excluded from one per-game test and flagged wherever it is included; it is never excluded from the pooled table. Counts are from a live production crawl and drift by a few dozen rows between reads.