Every mod on the Steam Workshop gets one line of free advertising: its title. This series has spent a year taking that line apart — the bracket divide and its 2.96x for [tag] prefixes, the short-title problem, the name-collision tax, the dead-end tags that mark a shelf nobody browses. All of those measured shape: brackets, length, punctuation, alphabet, tags.
This one measures something none of them touched — vocabulary. The actual word a creator chose. And it turned up the sharpest, most durable title signal we have found: not whether a title looks good, but whether it admits the thing is not finished.
Of the 6,026,296 active Workshop mods carrying at least one subscription, 106,737 — 1.77% — contain a word that says "unfinished." Test. Testing. WIP. Prototype. Placeholder. Bugfix. Experimental. Temp. They are a rounding error on the catalogue. They are also the worst-performing cohort this series has ever isolated.
Restrict to mods with at least ten subscribers — 3,915,570 of them, a floor that clears the dead tail our crawler never fully revisits and the sub-10 cohort where the reach-1,000 signal is pure noise — and split them by vocabulary:
| Title vocabulary | Mods | Reach 1,000 subs | Avg subs |
|---|---|---|---|
| Finished — pack, bundle, collection, mega, complete, full, HD, enhanced, remastered, update(d), fixed, final, rebalance, rework | 119,230 | 36.24% | 5,371 |
| Neutral — neither class | 3,743,467 | 22.14% | 3,168 |
| Version — alpha, beta, early access | 18,092 | 20.79% | 2,820 |
| Unfinished — test, wip, prototype, placeholder, bugfix, unfinished, experimental, temp | 34,781 | 9.60% | 1,045 |
A mod whose title says it is finished reaches a thousand subscribers at 36.24%. A mod whose title says it is a test reaches a thousand at 9.60%. Same platform, same games, same crowd. That is a 3.8x gap, and it points in a direction the usual "write a better title" advice never quite names: it is not that a good title sells the mod. It is that a bad title tells you, in advance, that the mod was a sketch somebody uploaded and moved on from.
It is not length, and it is not vintage
The obvious confound is length. Words like wip and test are short, and this series already showed that very short titles underperform. But the WIP cohort's average title is 24.4 characters — longer than the 21.6-character neutral average, and the class's words cluster in exactly the band where length stops mattering. Hold length fixed and the gap is untouched:
Title length 30–39 characters only (base cohort):
| Title vocabulary | Mods | Reach 1,000 |
|---|---|---|
| Finished | 28,162 | 39.39% |
| Neutral | 478,107 | 28.31% |
| Unfinished (either class) | 8,823 | 16.65% |
A second confound is age. Older mods have had years to accumulate subscribers, and a title written in 2013 was written to a different platform. Restrict to mods published in 2024 or later — the cohort every creator today actually competes in — and the ordering is identical:
| Title vocabulary | Mods | Reach 1,000 |
|---|---|---|
| Finished | 48,844 | 25.30% |
| Neutral | 1,555,877 | 17.70% |
| Unfinished (either class) | 18,441 | 10.48% |
The level falls for everyone as the catalogue crowds — the same attention dilution that has flattened every cohort in this series. The ratio does not move.
The words themselves
Pooled reach by word, for every token appearing in at least 5,000 titles in the base cohort, splits cleanly into a winner list and a graveyard:
| Word | Mods containing it | Reach 1,000 |
|---|---|---|
| hd | 3,911 | 47.10% |
| pack | 59,859 | 41.16% |
| collection | 5,436 | 40.08% |
| enhanced | 5,149 | 38.67% |
| remastered | 3,805 | 32.93% |
| update / updated | 15,387 | 32.81% |
| fixed / fix | 19,282 | 28.16% |
| vanilla | 15,279 | 27.28% |
| — platform baseline — | 22.14% | |
| alpha / beta | 15,468 | 20.78% |
| wip | 6,835 | 16.75% |
| prototype | 3,085 | 10.08% |
| rocket / boat / coaster / flying | 32,206 | ~10.1% |
| testing | 3,042 | 7.92% |
| chamber | 5,913 | 6.33% |
| test | 18,601 | 6.26% |
| plane | 17,393 | 4.47% |
| blueprint | 8,068 | 4.05% |
Two things fall out of that table that a single "good word / bad word" lesson would miss.
First, the winners share a meaning, not a length: pack, bundle, collection, HD, enhanced, remastered, complete. Every one of them signals a finished, packaged object — a thing that was assembled deliberately and shipped whole. pack alone carries 59,859 mods and 41.16% reach. The words that are about completeness are the words that correlate with it.
Second, the losers do not map onto topic, which is why "just pick a better subject" is bad advice. blueprint, plane, chamber, rocket, boat, coaster and flying are all creative, legitimate subjects — and all land between 4% and 11%, because they are the vocabulary of build-it-yourself sandboxes (Kerbal Space Program, Besiege, Trailmakers, Scrap Mechanic), where the vast majority of uploads are one person's experiment that never left the workbench. test at 6.26% is not a curse word. It is a flag on a category.
The control that separates a lever from a costume
Everything above is pooled, and every pooled number in this series has so far been at least half composition. The only test that separates a real effect from a creator's own quality is to compare the same creator against themselves, inside the same game — the discipline the cadence tax, the zero-favorite floor and the catalog ladder all had to pass.
Run that test and the two halves of the finding split apart.
Finished vocabulary vs neutral, paired (creator, game):
- 42,630 paired groups
- finished side reaches 1,000 at 39.36% (95,899 mods)
- neutral side at 35.66% (836,806 mods)
- the finished side wins 55.0% of the pairings
The 1.6x pooled advantage collapses to 1.10x. Once you hold the creator fixed, most of the finished-language boost was selection: creators who name things pack and HD are the creators who ship finished projects, and the label simply rode along. Real, but small.
Unfinished vocabulary vs neutral, paired (creator, game):
- 15,074 paired groups
- unfinished side reaches 1,000 at 11.14% (23,421 mods)
- neutral side at 27.48% (323,172 mods)
- the unfinished side loses 62.8% of the pairings
The 2.31x pooled penalty does not collapse. It holds at 2.47x — inside the same creator's own catalogue, inside the same game, against their own neutral-titled uploads. Version language (alpha/beta) is only mildly negative in the paired test — 22.55% against 30.99% — but the plain test / wip / prototype words survive everything.
That asymmetry is the finding. The positive title vocabulary is mostly a mirror. The negative vocabulary is a signal.
And it is not a transliteration artefact. Re-run the whole thing on ASCII-only titles (3,480,456 mods — stripping every non-Latin-alphabet title, where regex word-boundaries behave differently): the unfinished cohort still lands at 9.40%, against 9.60% in the full cohort. The result does not move.
It is not one game
The per-game sign test on finished-vs-neutral is lopsided: of the 50 games with at least 500 mods on each side, the finished cohort wins reach-1,000 in 43. The two or three reversals are the small, curation-heavy ecosystems where an "update" tag is a maintenance note rather than a shipping signal.
The unfinished cohort is even more consistent. Take the 14 games with at least 400 WIP-vocabulary titles, and in every single one the WIP titles reach 1,000 below their own game's baseline:
| Game | Mods | Game reach-1k | Reach-1k, WIP titles |
|---|---|---|---|
| Kerbal Space Program | 63,687 | 1.98% | 0.41% |
| Trailmakers | 250,591 | 1.15% | 0.86% |
| Portal 2 | 66,883 | 6.80% | 3.69% |
| Scrap Mechanic | 75,233 | 6.57% | 3.71% |
| Besiege | 62,738 | 8.14% | 5.30% |
| Tabletop Simulator | 82,539 | 16.69% | 5.69% |
| DayZ | 56,416 | 14.43% | 6.57% |
| Arma 3 | 83,650 | 11.77% | 7.52% |
| Stormworks | 77,435 | 19.13% | 8.94% |
| Brick Rigs | 147,095 | 14.83% | 9.11% |
| Counter-Strike 2 | 23,711 | 33.09% | 10.90% |
| ARK: Survival Evolved | 15,998 | 34.95% | 12.11% |
| Space Engineers | 61,210 | 19.58% | 14.11% |
| Garry's Mod | 167,434 | 52.53% | 42.64% |
Fourteen games, fourteen negative signs. A WIP title is a worse-than-baseline signal in every ecosystem big enough to measure it, including the two — Garry's Mod and Counter-Strike 2 — where a title like that could hardly be more different in tone.
The exception worth naming
The rule is a rate, not a wall, and the counter-examples in this series are always the most instructive. The single most-subscribed title in the unfinished cohort is FA:S 2.0 Alpha SWEPs — an early-access weapons pack for Garry's Mod sitting on 1,158,376 subscribers, with four sibling uploads (Rifles, Pistols, SMGs, Shotguns) each past a million. Black Ops C4 (Sandbox) holds 2,054,453. UE4SS Experimental (Palworld) holds 1,481,093.
Those are not contradictions. They are the mechanism stated plainly: the WIP cohort produces the occasional giant, and it is defined by a long tail of ones nobody ever finished. 333 of Garry's Mod's WIP-titled mods did cross 1,000 subscribers — a healthy absolute count, in a game where 42.64% of everything crosses it. The rate is 42.64%. The rate for the cohort is what a creator is betting on, not the existence of winners.
What the words actually do
Read the two halves of the control together and the practical shape is clear.
A finished-sounding title — pack, HD, collection, enhanced, remastered — buys you 1.10x over your own neutral titles. That is worth having and worth nothing to chase: it is almost entirely the label following the finished object, not causing it.
An unfinished-sounding title — test, wip, prototype, placeholder — costs 2.47x, and unlike everything else this series has measured, the cost survives the creator control, the length control, the vintage control and the per-game sign test. The honest reading is still not "rename your mod and it will grow." It is that a title is a self-report, and the platform has spent a decade pricing it accurately: a mod called Prototype is, far more often than not, exactly that. The name is not the lever. The name is the receipt.
And the receipt is public. In a catalogue of six million where 1.77% of creators are still telling you, in the one line of advertising they control, that the upload is a draft, the finished-sounding 36% and the unfinished-sounding 9.6% are two readings of the same fact: the word on the box was always honest.
Method: all figures from the modsignal.premarcu.com catalogue, October 2026 — 6,026,296 active Workshop mods with ≥1 current subscription across 391 tracked games; rate analysis on the ≥10-subscriber base cohort (3,915,570 mods). Vocabulary classes are case-insensitive word-boundary matches on the item title. Paired controls group by (creator, game) and require at least one mod on each side; "wins/loses" compares the group's mean log-subscriptions on each side. Per-game sign tests require ≥500 mods per side; per-game WIP cuts require ≥400 WIP titles. Reach-1,000 = share of mods with subscriptions_current ≥ 1000. Non-ASCII titles excluded only in the robustness re-run.