The Steam Workshop's Bracket Divide (2026): A Title That Opens With `[TFA]` Reaches 1,000 Subscribers 61.7% of the Time — and 14.3% When the Bracket Names a Language

This series has now measured what a Workshop mod contains, what it is called, how long its name is, when it ships, and how many mods the person who made it has already published. It has never measured the shape of the title itself — the punctuation that sits in front of the words, before any reader gets to the actual name. That turns out to be where some of the largest and most misleading gaps on the platform live.

Using ModSignal's live dataset — 5,999,860 active Workshop items that carry at least one subscription, of which 863,261 (14.39%) have reached 1,000 subscribers — we classified every title by which marker, if any, sits at the front of it, and measured the hit rate of each class. Then we broke each class down by what the marker actually says, and ran the composition controls this series always runs.

The four markers

Two of these are enormous. A title that opens with a bracketed tag is 2.96x more likely to have crossed a thousand subscribers than a plain title. A pipe separator gets you 2.62x. And ALL CAPS goes the other way — 9.11% against 13.76%, a 34% shortfall.

Before any of those numbers mean anything, they have to survive the control this series applies every time: the same creator, inside the same game.

What the bracket actually says

The first problem with "40.80%" is that [Anything] is not one thing. It is at least three different things stacked on top of each other, and they behave nothing alike.

Framework tags read 61.65%. Language tags read 14.27% — statistically level with the plain-title baseline of 13.76%, and 4.3x below the framework class. Same punctuation, same privileged position, same platform.

That is a warning, not a finding, and the rest of this article is about why. But it is also the most useful thing in the article for anyone naming a mod today, so it is worth stating plainly first: [DEUTSCH] and [TFA] are the same shape and opposite signals, and the platform treats them as if they were the same thing.

The composition controls

Per game. The bracket is wildly unevenly distributed across ecosystems, and each ecosystem has its own baseline hit rate. Take every game with at least 20,000 active subscribed items and compare tagged against plain inside that game:

Across the 28 games large enough to test with at least 20,000 active subscribed items, 27 favour the tagged title. The direction is real and near-universal — but look at the ratio column, not the sign. In Garry's Mod and Left 4 Dead 2, where bracketed titles are 7% to 14% of the catalogue, the advantage is 1.2x to 1.3x — barely there. In Trailmakers, Besiege and Kerbal Space Program, where bracketed titles are a rounding error (0.12%, 0.24%, 0.30% of items), the advantage is 5x to 16x. The marker is not doing work in the games that use it most. It is doing work in the games that almost never use it — which is what you would expect from a signal that marks rarity rather than quality.

The single reversal is Planet Zoo at 0.89x: 349 tagged items, 45 of which hit, against 17,685 plain hits.

Wallpaper Engine. The ecosystem that breaks every other rule in this dataset, so check it does not supply this one. ModSignal tracks 119,397 active subscribed Wallpaper Engine items, of which 111,650 have crossed a thousand subscribers. Its bracket rate is 9.76% and its hit rates are 93.48% tagged against 92.22% plain: 1.01x. Wallpaper Engine is a saturated ecosystem where almost everything crosses a thousand subscribers, so it contributes volume and no signal. Remove it and the pooled gap narrows — 27.0 points to 23.7 — because its own tagged-versus-plain difference is negligible.

Framework tags only, versus everything. Restrict the tagged class to the fifteen framework tags and compare against all non-framework items: 61.65% against 13.93% — a 4.4x pooled gap, larger than the raw bracket number. Controls do not touch it. Now the strictest test this series uses.

The control that breaks it

Everything above compares one creator's mods against other creators' mods, in the same game. The strictest test holds both at once: the same creator, inside the same game, with at least one bracketed and one plain title in that game. That yields 19,961 creator-and-game pairs.

The pooled 40.80%-against-13.76% gap — 27.0 points — lands at 6.1 points. The bracket divide is still there, and it is real, but it is roughly a quarter of the size it looked like, and it has stopped being an explanation for anything. It is the residue of one creator being better than another and one game being friendlier than another, wearing a punctuation mark as a costume.

Run the same paired test on every marker, because two of them do not survive the way the pooled data implies:

The parenthesis is the cleanest cautionary tale: a pooled 1.66x advantage over a plain title becomes 23.57% against 23.21% once the creator and the game are held fixed across 118,322 pairs. The (...) in a Workshop title carries no information whatsoever about whether the mod crosses a thousand subscribers. It looked like a 9-point lever. It is a rounding error.

ALL CAPS behaves the opposite way and is the only marker here that survives as a genuine negative: 11.43% against 17.04% for the same creator's other titles in the same game. Pooled it looks like a 34% shortfall; paired it is a 33% one. Unlike the bracket, there is nothing to rescue.

The bracket is a status field, and the platform reads it

Here is the part that explains the whole thing. [DEPRECATED], [OUTDATED], [OBSOLETE], [LEGACY] — words that mean this mod is finished, abandoned, or superseded — sit on 3,540 items that read 42.71%, and on 1,512 that have crossed a thousand subscribers. The most-subscribed ones are not marginal:

A mod whose entire title is a single word of warning about its own deadness holds 226,825 subscribers. A mod explicitly labelled deprecated is the most-subscribed Portal Gun in the dataset, at 1.4 million. The status tags concentrate in the communities with enough catalogue depth that the audience has learned to read the label: Ravenfield has 294 of them (269 hit), Stormworks 215, Garry's Mod 200, Hearts of Iron IV 130, Project Zomboid 130.

This is the same finding as the disappearing mods, seen from the title instead of the roster: Workshop demand is extraordinarily sticky and largely indifferent to whether a mod still works. The bracket is not a promise that the mod is good. It is a field the ecosystem has repurposed as a shelf label, and the shelf is what gets shopped.

The four meanings of one bracket

Put the classes side by side and the mechanism is not mysterious:

The language class is the control that proves the bracket is not magic. It has the same punctuation, the same privileged position, and it earns nothing — 14.27% against a 13.76% plain-title baseline. If opening with a bracket were a discovery advantage, [RU] Advanced AI Command would get it too. It does not, because a bracket only helps when it tells a browser which shelf they are already shopping in.

Framework tags get the highest rate on the platform for a structural reason this series already documented in the one-subscriber floor and the tag cliff: a mod that plugs into [TFA] inherits [TFA]'s audience. Its addressable market is not "people who search for this mod", it is "everyone using the stack this mod extends" — and the paired control still leaves it 8 points ahead of the same creator's non-framework titles in the same game. A localisation tag does the exact opposite: it removes readers from the addressable market.

What it means

If you are naming a mod, the bracket is worth six points, not twenty-seven. The paired control is the number that survives, and it survives in the same direction for brackets, framework tags and pipes. Open with a framework tag if your mod genuinely belongs to a stack — 62.47% against 54.50% for the same creator's other titles in the same game, the largest surviving effect in this analysis. Use a pipe if you need to separate a name from its qualifier; it is worth about 3 points. Skip the parenthesis — 0.4 points, which is noise. And skip the ALL CAPS entirely: it is a genuine 33% penalty that survives every control, and the only marker here that costs you.

If you are reading a "tags help" statistic, check whether the tag is a shelf or a language. A pooled 40.80% against 13.76% is correct, survives per-game and excludes Wallpaper Engine, and still hides a 4.3x spread between [TFA] at 61.65% and [RU] at 14.27%. The pooled number is a mixture of a stack-grouping signal, a status-flagging signal and an audience-narrowing signal, and only the first two are positive. This series has now applied the same test eight times and the rule has not changed: a pooled gap that looks like a finding is usually composition wearing a costume. Here the costume is a square bracket, and what is underneath is smaller, sharper, and in one case pointing the other way.

Related reading: the short-title problem on the length of a name, the copy-paste descriptions on boilerplate blurbs, the script split on Cyrillic titles, the tag cliff on the 0-to-1 tag step, and the debut signal on why the first mod predicts the rest.

Method: 5,999,860 Steam Workshop items tracked by ModSignal as of 2 October 2026 across 392 games — every item that is is_active and carries at least one subscription, of which 863,261 have subscriptions_current >= 1000 (14.39%). "Reached 1,000 subscribers" is a point-in-time threshold, not a tracked trajectory. "Opens with a bracketed tag" is title LIKE '[%]%' — 139,292 items, 56,841 of them over the threshold. "Framework tag", "status tag" and "language tag" are curated lists of literal leading tokens, matched on the substring between the opening bracket and the next closing bracket; the framework list is TFA, SIMFPHYS, LVS, RICO, WOTC, COMM, B42, ARC9, DRGBASE, VJ, TFA-VOX, PM, XCOMM, TOUHOU and BLUE ARCHIVE, and the language list is the standard two- and three-letter codes plus DEUTSCH. The three curated classes are mutually exclusive and restricted to leading-bracket titles, so they sum to 139,290 rather than the full 139,292; the framework-versus-everything control deliberately uses the looser LIKE '[%]%' test, which is why its non-framework baseline (13.93%) differs slightly from the plain-title baseline (13.76%). ALL CAPS is title = upper(title) AND title <> lower(title) AND length(title) > 3; it is not mutually exclusive with the other markers. The paired control groups items by (creator_steam_id, game_id) and keeps only pairs containing at least one item in the class and one outside it; items with a null creator are excluded, which is why paired item counts (e.g. 101,772 tagged against 322,185 plain) differ from the platform totals. Per-game tables use games with at least 20,000 active subscribed items for the bracket comparison and at least 50,000 for the ALL-CAPS and pipe comparison. Wallpaper Engine is never excluded from the headline tables; its exclusion is reported as a control. Counts are from a live production crawl and drift by a few dozen rows between reads.