Something a reviewer said about gameplay and mechanics that runs on long enough to wrap on a narrow screen, which is where a quotation stops being readable.
Across these games
| A Measured Game <& Friends> | 8,300 | 70.0% | 80.0% |
|---|---|---|---|
| An Unmeasured Game | 2,000 | 70.0% | no reference set |
| A Third Game With A Rather Long Name | 3,800 | 70.0% | 80.0% |
| A Fourth Game | 7,400 | 70.0% | 80.0% |
Agreement is measured over each game's labelled claims, and only over the claims the model was willing to answer, so a game's figure carries a band several points wide and the pooled one at the end of this section is the one worth quoting. It counts agreement with a labeller, which is not the same as being right. A game the model trained on carries no figure: agreeing with labels it learned from would measure memory.
Mention rates for 21,500 reviews of 4 games. The game that raises each subject most is named beside it, and its cell outlined, where one clears every other by more than rounding. Read down a column for one game and across a row to compare them all; the table scrolls sideways. Deeper shading is a higher rate, and the rates are not comparable to any other corpus.
| Category | Highest | A Measured Game <& Friends> | An Unmeasured Game | A Third Game With A Rather Long Name | A Fourth Game |
|---|---|---|---|---|---|
| Gameplay and mechanics | 50.5% An Unmeasured Game | 13.7% | 50.5%, the highest of these games | 25.0% | 15.2% |
| Bugs and crashes | 50.5% An Unmeasured Game | 13.7% | 50.5%, the highest of these games | 24.9% | 15.2% |
| Performance | 50.4% An Unmeasured Game | 13.7% | 50.4%, the highest of these games | 24.9% | 15.2% |
| Story and writing | 50.7% An Unmeasured Game | 10.1% | 50.7%, the highest of these games | 25.1% | 11.2% |
| Overall verdict | 50.6% An Unmeasured Game | 10.1% | 50.6%, the highest of these games | 25.0% | 11.2% |
| Genre and comparisons | 50.6% An Unmeasured Game | 10.1% | 50.6%, the highest of these games | 25.0% | 11.2% |
| Graphics and art | 35.9% An Unmeasured Game | 10.1% | 35.9%, the highest of these games | 25.1% | 11.2% |
| Atmosphere and feel | 35.8% An Unmeasured Game | 10.1% | 35.8%, the highest of these games | 25.1% | 11.2% |
| Difficulty and balance | 36.0% An Unmeasured Game | 10.2% | 36.0%, the highest of these games | 17.3% | 11.3% |
| Controls and interface | 35.9% An Unmeasured Game | 10.2% | 35.9%, the highest of these games | 17.3% | 11.3% |
| Audio and music | 35.9% An Unmeasured Game | 10.1% | 35.9%, the highest of these games | 17.3% | 11.2% |
| Says nothing about the game | 37.5% An Unmeasured Game | 9.0% | 37.5%, the highest of these games | 19.7% | 10.1% |
| Amount of content | 36.0% An Unmeasured Game | 6.6% | 36.0%, the highest of these games | 17.3% | 11.3% |
| Monetisation and DLC | 36.2% An Unmeasured Game | 6.6% | 36.2%, the highest of these games | 17.4% | 7.3% |
| Price and value | 36.1% An Unmeasured Game | 6.6% | 36.1%, the highest of these games | 17.4% | 7.3% |
| Mods and user content | 36.1% An Unmeasured Game | 6.6% | 36.1%, the highest of these games | 17.4% | 7.2% |
| Community and players | 21.3% An Unmeasured Game | 6.6% | 21.3%, the highest of these games | 17.5% | 7.3% |
| Multiplayer and online | 21.2% An Unmeasured Game | 6.6% | 21.2%, the highest of these games | 17.4% | 7.3% |
| Accessibility and options | 21.5% An Unmeasured Game | 6.7% | 21.5%, the highest of these games | 9.7% | 7.4% |
| Hardware and compatibility | 21.4% An Unmeasured Game | 6.7% | 21.4%, the highest of these games | 9.7% | 7.3% |
| Publisher and platform policy | 21.4% An Unmeasured Game | 6.7% | 21.4%, the highest of these games | 9.6% | 7.3% |
| Updates and developer support | 21.3% An Unmeasured Game | 6.6% | 21.3%, the highest of these games | 9.6% | 7.3% |
| Licensed content | 21.6% An Unmeasured Game | 3.1% | 21.6%, the highest of these games | 9.8% | 3.3% |
| Tutorial and learning | 21.6% An Unmeasured Game | 3.1% | 21.6%, the highest of these games | 9.7% | 3.3% |
| Language and localisation | 21.6% An Unmeasured Game | 3.1% | 21.6%, the highest of these games | 9.7% | 3.3% |
Across these 4 games the top of the pile overstates performance by 2.7×: 104 of the 200 reviews at the top of those 4 piles raise it, against 19.6% of all 21,500.
The subject these games disagree about most is story and writing: 50.7% of An Unmeasured Game reviews raise it, against 10.1% of A Measured Game <& Friends>. A rate is about one corpus, and this is what that means.
A Measured Game <& Friends>
- Reviews counted
- 8,300
- Captured
- 8,300 of the 8,320 Valve reports (99.80%)
- Steam calls it
- Mostly Positive
- Snapshot
- 27 September 2025
- Brought up to date
- 3 October 2025, 40 added or edited since the snapshot
The capture was brought up to date on 3 October 2025 and these counts were made before that, on the corpus as it stood on 27 September 2025. Read it again to count what arrived.
The top of the pile overstates atmosphere and feel by 4.1×: 21 of the 50 reviews Steam ranks most helpful raise it, against 10.1% of all 8,300.
Players love mixing two factions every run, which 410 reviews praise, and many find the combat tight. The clearest complaint is a black screen on startup, raised in 48 reviews and tied to performance as well. Others say it "stutters in big fights".
Written by a language model on the computer that made this page, from the points below. Every number and quotation in it is one those points give; a sentence with any other was left out.
What they praise and complain about
Points that say the same thing, in the same words or others, grouped by meaning, each with how many reviews make it.
Praised
- Love how you can mix and match factions.410 reviews, 4.9% · Gameplay and mechanics
Mixing two factions every run keeps it fresh after a hundred hours.
- The combat is tight.160 reviews, 1.9% · Gameplay and mechanicsAlso said about Controls and interface
Complained about
- Black screen on startup.48 reviews, 0.6% · Bugs and crashesAlso said about Performance
Every time I start it the screen stays black <until I alt-tab>.
- Saves get corrupted.21 reviews, 0.3% · Bugs and crashes
When it was said
Reviews per month, Jan 2023 to Dec 2024, against a busiest month of 399 in Dec 2024. The line is the share of each month that recommended the game, from none at the bottom to all at the top; the dashed line is half. It falls furthest in Jan 2023, when 66.0% of 100 reviews recommended the game. Point at a month to read it.
Around its updates
The dashed lines on the chart are the 7 updates this game's developer posted on Steam by 2 October 2025: posts Steam marks as patch notes, and posts whose titles name a patch, a hotfix, an update or a version. Below, the biggest of them by the reviews that followed, each with the 28 days before it against the 28 days after. A change is a share that moved further than chance would move it, three standard errors and at least two points, and a share is compared only where each side holds at least 100 reviews. A change happened across the update; that alone does not make the update the reason for it.
Patch 1.1 14 March 2023
224 reviews in the 28 days before, 224 in the 28 days after. One other update was posted within these weeks, and they hold its effect too.
Nothing changed beyond chance: not the share recommending the game, and not the praise or complaints of any subject.
The Big Update: 1.3 5 March 2024
224 reviews in the 28 days before, 224 in the 28 days after.
- Recommending the game: fell from 69.2% to 50.9% of reviews
- Bugs and crashes: complaints rose from 9.8% to 30.8% of reviews
Patch 1.4: Performance 10 June 2024
224 reviews in the 28 days before, 224 in the 28 days after. One other update was posted within these weeks, and they hold its effect too.
- Performance: praise rose from 25.0% to 50.0% of reviews
Every update (7)
- 20 December 2024 Patch 1.5 too few reviews either side to compare
- 12 June 2024 Hotfix 1.4.1 1 change
- 10 June 2024 Patch 1.4: Performance 1 change
- 5 March 2024 The Big Update: 1.3 2 changes
- 20 September 2023 The 1.2 Update too few reviews either side to compare
- 16 March 2023 Hotfix 1.1.1 nothing changed beyond chance
- 14 March 2023 Patch 1.1 nothing changed beyond chance
What players talk about
70.0% of all 8,300 reviews recommend this game. Every figure in the last column is worth reading against that.
A review counts towards every category it says something about, so these add up to more than 100%. Select a row to read the reviews behind it.
What the columns mean, and where the quoted reviews come from
- Mention rate
- The share of all reviews that say something about the category, with the number of reviews beside it. These are the headline figures and they add up to more than 100%.
- Main subject
- The share whose single main subject is that category. Unlike mention rates, these add up to the number of reviews.
- Top of the pile
- The mention rate over the 50 reviews Steam ranks as most helpful, which is roughly what somebody sees before deciding.
- Bias
- How much the top of the pile overstates a category. 0× means none of those few dozen reviews raised it, which is weak evidence rather than proof of absence.
- Said about it
- Of the reviews raising a category, how many praise it, how many complain about it, and how many do both. A subject half the players love and half hate, and one every player has mixed feelings about, are different findings that a single share hides.
- Recommended
- The share of the reviews raising a category that still recommended the game, against this game's own baseline. Marked warm or cold only where the gap is wider than the number of reviews behind it can explain.
The reviews behind a row are a couple from the top of the pile where the category reaches it, and the rest drawn at random from everything filed there, so they are evidence rather than a selection.
| Gameplay and mechanics | 13.7%1,135 | 6.8% | 56.0% | 4.1× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|---|---|---|---|---|---|---|
|
Measured on this row: of the 58 labelled claims about it, the model found 65.5%; of the claims it filed here, 62.3% were labelled that way. A rate built on a subject it misses is a floor, not a count. Where it was read wrongly, it was most often read as gameplay (7 of the labelled claims). Corrected for those errors, the share of claims about this is about 0.0%, against the 9.1% the model read. That is an estimate from 58 labels, and it moves with them. What they praise 567 reviewsWords that stand out best in years41 worth it22
What they complain about 567 reviewsWords that stand out save corruption37 crashes on launch29 unplayable11
| ||||||
| Bugs and crashes | 13.7%1,134 | 6.8% | 56.0% | 4.1× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
|
Measured on this row: of the 59 labelled claims about it, the model found 66.1%; of the claims it filed here, 62.9% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 9.1% the model read. That is an estimate from 59 labels, and it moves with them. What they praise 567 reviewsWords that stand out best in years41 worth it22
What they complain about 567 reviewsWords that stand out save corruption37 crashes on launch29 unplayable11
| ||||||
| Performance | 13.7%1,133 | 6.8% | 56.0% | 4.1× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 60 labelled claims about it, the model found 66.7%; of the claims it filed here, 63.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 9.1% the model read. That is an estimate from 60 labels, and it moves with them. What they praise 565 reviewsWords that stand out best in years41 worth it22
What they complain about 565 reviewsWords that stand out save corruption37 crashes on launch29 unplayable11
| ||||||
| Difficulty and balance | 10.2%844 | 5.1% | 42.0% | 4.1× | 40.0% praise40.0% gripe, 19.9% both | 66.6% |
|
Measured on this row: of the 49 labelled claims about it, the model found 65.3%; of the claims it filed here, 61.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 6.8% the model read. That is an estimate from 49 labels, and it moves with them. What they praise 421 reviews
What they complain about 421 reviews
| ||||||
| Controls and interface | 10.2%843 | 5.1% | 42.0% | 4.1× | 40.0% praise40.0% gripe, 19.9% both | 66.7% |
|
Measured on this row: of the 50 labelled claims about it, the model found 66.0%; of the claims it filed here, 62.3% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 6.8% the model read. That is an estimate from 50 labels, and it moves with them. What they praise 421 reviews
What they complain about 421 reviews
| ||||||
| Audio and music | 10.1%842 | 5.1% | 42.0% | 4.1× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 51 labelled claims about it, the model found 66.7%; of the claims it filed here, 63.0% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 6.8% the model read. That is an estimate from 51 labels, and it moves with them. What they praise 420 reviews
What they complain about 420 reviews
| ||||||
| Graphics and art | 10.1%841 | 5.1% | 42.0% | 4.1× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 52 labelled claims about it, the model found 65.4%; of the claims it filed here, 61.8% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 6.8% the model read. That is an estimate from 52 labels, and it moves with them. What they praise 420 reviews
What they complain about 420 reviews
| ||||||
| Atmosphere and feel | 10.1%840 | 5.1% | 42.0% | 4.1× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
|
Measured on this row: of the 53 labelled claims about it, the model found 66.0%; of the claims it filed here, 62.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 6.7% the model read. That is an estimate from 53 labels, and it moves with them. What they praise 420 reviews
What they complain about 420 reviews
| ||||||
| Story and writing | 10.1%839 | 5.0% | 40.0% | 4.0× | 40.0% praise40.0% gripe, 19.9% both | 66.6% |
|
Measured on this row: of the 54 labelled claims about it, the model found 66.7%; of the claims it filed here, 63.2% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 6.7% the model read. That is an estimate from 54 labels, and it moves with them. What they praise 418 reviews
What they complain about 418 reviews
| ||||||
| Overall verdict | 10.1%837 | 5.0% | 40.0% | 4.0× | 40.0% praise40.0% gripe, 19.9% both | 66.7% |
|
Measured on this row: of the 56 labelled claims about it, the model found 66.1%; of the claims it filed here, 62.7% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 6.7% the model read. That is an estimate from 56 labels, and it moves with them. What they praise 418 reviews
What they complain about 418 reviews
| ||||||
| Genre and comparisons | 10.1%836 | 5.0% | 40.0% | 4.0× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 57 labelled claims about it, the model found 66.7%; of the claims it filed here, 63.3% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 6.7% the model read. That is an estimate from 57 labels, and it moves with them. What they praise 417 reviews
What they complain about 417 reviews
| ||||||
| Says nothing about the game | 9.0%750 | 4.5% | 36.0% | 4.0× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
|
Measured on this row: of the 55 labelled claims about it, the model found 65.5%; of the claims it filed here, 62.1% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 6.0% the model read. That is an estimate from 55 labels, and it moves with them. What they praise 375 reviews
What they complain about 375 reviews
| ||||||
| Accessibility and options | 6.7%554 | 3.3% | 26.0% | 3.9× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 39 labelled claims about it, the model found 66.7%; of the claims it filed here, 61.9% were labelled that way. A rate built on a subject it misses is a floor, not a count. What they praise 276 reviews
What they complain about 276 reviews
| ||||||
| Hardware and compatibility | 6.7%553 | 3.3% | 26.0% | 3.9× | 40.0% praise40.0% gripe, 20.0% both | 66.5% |
|
Measured on this row: of the 40 labelled claims about it, the model found 65.0%; of the claims it filed here, 60.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 4.4% the model read. That is an estimate from 40 labels, and it moves with them. What they praise 276 reviews
What they complain about 276 reviews
| ||||||
| Publisher and platform policy | 6.7%552 | 3.3% | 26.0% | 3.9× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
|
Measured on this row: of the 41 labelled claims about it, the model found 65.9%; of the claims it filed here, 61.4% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 4.4% the model read. That is an estimate from 41 labels, and it moves with them. What they praise 276 reviews
What they complain about 276 reviews
| ||||||
| Updates and developer support | 6.6%551 | 3.3% | 26.0% | 3.9× | 40.0% praise40.0% gripe, 19.9% both | 66.6% |
|
Measured on this row: of the 42 labelled claims about it, the model found 66.7%; of the claims it filed here, 62.2% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 4.4% the model read. That is an estimate from 42 labels, and it moves with them. What they praise 274 reviews
What they complain about 274 reviews
| ||||||
| Community and players | 6.6%550 | 3.3% | 26.0% | 3.9× | 40.0% praise40.0% gripe, 19.9% both | 66.5% |
|
Measured on this row: of the 43 labelled claims about it, the model found 65.1%; of the claims it filed here, 60.9% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 4.4% the model read. That is an estimate from 43 labels, and it moves with them. What they praise 274 reviews
What they complain about 274 reviews
| ||||||
| Multiplayer and online | 6.6%549 | 3.3% | 26.0% | 3.9× | 40.0% praise40.0% gripe, 19.9% both | 66.7% |
|
Measured on this row: of the 44 labelled claims about it, the model found 65.9%; of the claims it filed here, 61.7% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 4.4% the model read. That is an estimate from 44 labels, and it moves with them. What they praise 274 reviews
What they complain about 274 reviews
| ||||||
| Monetisation and DLC | 6.6%548 | 3.3% | 26.0% | 3.9× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 45 labelled claims about it, the model found 66.7%; of the claims it filed here, 62.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 4.4% the model read. That is an estimate from 45 labels, and it moves with them. What they praise 273 reviews
What they complain about 273 reviews
| ||||||
| Price and value | 6.6%547 | 3.3% | 26.0% | 3.9× | 40.0% praise40.0% gripe, 20.0% both | 66.5% |
|
Measured on this row: of the 46 labelled claims about it, the model found 65.2%; of the claims it filed here, 61.2% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 4.4% the model read. That is an estimate from 46 labels, and it moves with them. What they praise 273 reviews
What they complain about 273 reviews
| ||||||
| Mods and user content | 6.6%546 | 3.3% | 26.0% | 4.0× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
|
Measured on this row: of the 47 labelled claims about it, the model found 66.0%; of the claims it filed here, 62.0% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 4.4% the model read. That is an estimate from 47 labels, and it moves with them. What they praise 273 reviews
What they complain about 273 reviews
| ||||||
| Amount of content | 6.6%545 | 3.3% | 26.0% | 4.0× | 40.0% praise40.0% gripe, 19.9% both | 66.6% |
|
Measured on this row: of the 48 labelled claims about it, the model found 66.7%; of the claims it filed here, 62.7% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 4.4% the model read. That is an estimate from 48 labels, and it moves with them. What they praise 271 reviews
What they complain about 271 reviews
| ||||||
| Licensed content | 3.1%257 | 1.5% | 12.0% | 3.9× | 40.1% praise40.1% gripe, 19.8% both | 66.5% |
|
Measured on this row: of the 36 labelled claims about it, the model found 66.7%; of the claims it filed here, 61.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. What they praise 127 reviews
What they complain about 127 reviews
| ||||||
| Tutorial and learning | 3.1%256 | 1.5% | 12.0% | 3.9× | 40.1% praise40.1% gripe, 19.8% both | 66.4% |
|
Measured on this row: of the 37 labelled claims about it, the model found 64.9%; of the claims it filed here, 60.0% were labelled that way. A rate built on a subject it misses is a floor, not a count. What they praise 127 reviews
What they complain about 127 reviews
| ||||||
| Language and localisation | 3.1%255 | 1.5% | 12.0% | 3.9× | 40.1% praise40.1% gripe, 19.8% both | 66.7% |
|
Measured on this row: of the 38 labelled claims about it, the model found 65.8%; of the claims it filed here, 61.0% were labelled that way. A rate built on a subject it misses is a floor, not a count. What they praise 127 reviews
What they complain about 127 reviews
| ||||||
| VR and headsets | 0.0%0 | 0.0% | 0.0% | not measured | none raised it | none raised it |
Who said it
Steam records beside every review how long its writer had played, whether mostly on a Steam Deck, whether the game was in early access, and whether they got it free. Each kind of reviewer is set against everyone else, and a difference is listed only where it is 4 times the gap chance alone typically makes and at least 2 points wide. A kind with fewer than 100 reviews is not compared.
| Who wrote it | Reviews | Recommended |
|---|---|---|
| Time played when they wrote it | ||
| Under 2 hours | 913 | 49.9% |
| 2 to 10 hours | 1,909 | 70.0% |
| 10 to 30 hours | 2,075 | 70.0% |
| 30 to 100 hours | 1,909 | 70.0% |
| 100 hours or more | 1,494 | 69.9% |
| Where they played | ||
| Mostly on a Steam Deck | 12 | too few to say |
| Mostly elsewhere | 8,288 | too few others to set it against |
| When they wrote it | ||
| During early access | 2,490 | 70.0% |
| After release | 5,810 | 70.0% |
| How they got the game | ||
| Got it free | 332 | 69.9% |
| Paid for it | 7,968 | 70.0% |
Where they differ
- Reviewers with 100 hours or more played complain about amount of content in 31.1% of their reviews, against 3.2% of everyone else.
- Reviewers with under 2 hours played praise tutorial and learning in 25.4% of their reviews, against 1.5% of everyone else.
- Reviewers with under 2 hours played recommend the game in 49.9% of their reviews, against 70.0% of everyone else.
- Reviewers with 10 to 30 hours played complain about amount of content in 3.2% of their reviews, against 9.9% of everyone else.
- Reviewers with 2 to 10 hours played complain about amount of content in 3.2% of their reviews, against 9.7% of everyone else.
- Reviewers with 30 to 100 hours played complain about amount of content in 3.2% of their reviews, against 9.7% of everyone else.
- Reviewers with 10 to 30 hours played praise tutorial and learning in 1.5% of their reviews, against 5.0% of everyone else.
- Reviewers with 2 to 10 hours played praise tutorial and learning in 1.5% of their reviews, against 4.9% of everyone else.
- Reviewers with 30 to 100 hours played praise tutorial and learning in 1.5% of their reviews, against 4.9% of everyone else.
- Reviewers with under 2 hours played complain about amount of content in 3.1% of their reviews, against 8.8% of everyone else.
- Reviewers with 100 hours or more played praise tutorial and learning in 1.5% of their reviews, against 4.7% of everyone else.
What this game's players talk about that others' do not
Found by reading reviews of this game chosen to be as unlike each other as possible, and named only where at least three of them raise the same thing. These rows carry no rate: the model that counts the table above has not been trained on them, so what is shown is the finding and the reviews it rests on.
- Mud and terrain deformation a form of Gameplay and mechanics
How the ground deforms under a wheel, and whether it behaves the way a driver expects; not the truck roster and not the scenery.
3 of the 3 reviews it was found in
-
A review that is mostly about the mud, at the length reviews about one mechanic tend to run to. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things.
-
A review that is mostly about the mud, at the length reviews about one mechanic tend to run to. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things.
-
A review that is mostly about the mud, at the length reviews about one mechanic tend to run to. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things.
-
The top of the pile
The 50 reviews Steam ranks as most helpful, which is roughly what a reader sees before deciding. Every rate above is measured against the whole corpus instead.
Read all 1 of them
-
The most helpful review in the corpus, which is what a reader skimming the store page sees before anything this tool counted.
The most helpful review in the corpus, which is what a reader skimming the store page sees before anything this tool counted. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things.
What language it was said in
41.2% of these reviews are not in English. Steam shows a reader their own language by default, so most of this argument is one they never see.
2.0% of them are in a language the reader declines outright (Indonesian): the reference set holds too few claims there to keep its accuracy promise, so it says nothing rather than guessing. That is a gap in the labels, not a finding about this game.
39.2% are in a language the reader has to be surer about than English before it will answer at all (Chinese (simplified), Russian). A corpus weighted towards those declines more than usual for that reason rather than for anything about the game, and what it wants is more labels in those languages.
- English58.8%4,980
- Chinese (simplified)29.4%2,490
- Russian9.8%830
- Indonesian2.0%166
How far to trust this
- Read by
- Game Review Reader (run a-reader), a fine-tune of Alibaba-NLP/gte-multilingual-base, trained on label set 0123456789abcdef, answering a claim only where it clears the line drawn for its subject and the one for its language (rule fedcba9876543210)
- Claims read
- 24,900 from 8,300 reviews
- Claims it would not answer
- 14,940 (60.0%), counted as unclassified rather than filed under a best guess, above the 53.0% it usually declines on a game it has never seen
- Reviews it said nothing about
- 166 (2.0% of those read), where no point cleared the threshold
Of the 260 labelled claims this model was willing to answer, it named the same subject a separate labeller did 80.0% of the time, somewhere in [74.7%, 84.4%] with 95% confidence. The labeller was itself a language model, so that is agreement, not accuracy: two models can be wrong together, most easily on sarcasm and on claims that sit between subjects.
That figure covers the 50.0% of those claims it answered. It declined the other 50.0%, which are counted as unclassified everywhere on this page rather than being filed under a best guess.
A further 19 labelled claims were left out because this build takes their reviews apart differently from the build they were labelled under, so no reading corresponds to them.
Averaged over the subjects rather than over the claims, so a rare one counts as much as a common one, that comes to 0.64 on a scale where 1 is perfect agreement. Every one of the 26 subjects with enough labels to judge is found in at least a quarter of the claims making it.
Where it disagrees most, 7 claims the labeller called Gameplay and mechanics were read as gameplay. That is a boundary between two subjects rather than a mistake about one of them.
420 of those claims were read a second time, by a different labeller working blind, and the two reached the same subject on 92.1% of them. On those settled claims the model agrees 85.5%, somewhere in [81.7%, 88.7%], which is the nearest thing to accuracy a set labelled by models can produce: a label two independent readings reached is one worth scoring against. Of the settled claims, 321 were ones neither labeller called contested, and on those it agrees 93.1% of the time. On the 33 they read differently there is no single answer to be right about, and it lands on one of their two 81.8% of the time.
These rates are a census, not a survey: every review Valve serves was counted, so there is no sampling error to report. What they do carry is classifier error, which is what the agreement figure above measures.
An Unmeasured Game
- Reviews counted
- 2,000
- Captured
- 2,000 of the 2,020 Valve reports (99.80%)
- Steam calls it
- Mostly Positive
- Snapshot
- 27 September 2025
- Brought up to date
- 3 October 2025, 40 added or edited since the snapshot
The capture was brought up to date on 3 October 2025 and these counts were made before that, on the corpus as it stood on 27 September 2025. Read it again to count what arrived.
The top of the pile overstates difficulty and balance by 1.0×: 18 of the 50 reviews Steam ranks most helpful raise it, against 36.0% of all 2,000.
When it was said
Reviews per month, Jan 2023 to Dec 2024, against a busiest month of 479 in Dec 2024. The line is the share of each month that recommended the game, from none at the bottom to all at the top; the dashed line is half. It falls furthest in Feb 2023, when 66.3% of 193 reviews recommended the game. Point at a month to read it.
Around its updates
Steam has not been asked for the updates this game's developer posted. Bringing the game up to date asks.
What players talk about
70.0% of all 2,000 reviews recommend this game. Every figure in the last column is worth reading against that.
A review counts towards every category it says something about, so these add up to more than 100%. Select a row to read the reviews behind it.
What the columns mean, and where the quoted reviews come from
- Mention rate
- The share of all reviews that say something about the category, with the number of reviews beside it. These are the headline figures and they add up to more than 100%.
- Main subject
- The share whose single main subject is that category. Unlike mention rates, these add up to the number of reviews.
- Top of the pile
- The mention rate over the 50 reviews Steam ranks as most helpful, which is roughly what somebody sees before deciding.
- Bias
- How much the top of the pile overstates a category. 0× means none of those few dozen reviews raised it, which is weak evidence rather than proof of absence.
- Said about it
- Of the reviews raising a category, how many praise it, how many complain about it, and how many do both. A subject half the players love and half hate, and one every player has mixed feelings about, are different findings that a single share hides.
- Recommended
- The share of the reviews raising a category that still recommended the game, against this game's own baseline. Marked warm or cold only where the gap is wider than the number of reviews behind it can explain.
The reviews behind a row are a couple from the top of the pile where the category reaches it, and the rest drawn at random from everything filed there, so they are evidence rather than a selection.
| Story and writing | 50.7%1,015 | 25.4% | 50.0% | 1.0× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|---|---|---|---|---|---|---|
What they praise 507 reviews
What they complain about 507 reviews
| ||||||
| Overall verdict | 50.6%1,013 | 25.3% | 50.0% | 1.0× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
What they praise 505 reviews
What they complain about 505 reviews
| ||||||
| Genre and comparisons | 50.6%1,012 | 25.3% | 50.0% | 1.0× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
What they praise 505 reviews
What they complain about 505 reviews
| ||||||
| Gameplay and mechanics | 50.5%1,011 | 25.2% | 50.0% | 1.0× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
What they praise 505 reviewsWords that stand out best in years41 worth it22
What they complain about 505 reviewsWords that stand out save corruption37 crashes on launch29 unplayable11
| ||||||
| Bugs and crashes | 50.5%1,010 | 25.2% | 50.0% | 1.0× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
What they praise 504 reviewsWords that stand out best in years41 worth it22
What they complain about 504 reviewsWords that stand out save corruption37 crashes on launch29 unplayable11
| ||||||
| Performance | 50.4%1,009 | 25.2% | 50.0% | 1.0× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
What they praise 504 reviewsWords that stand out best in years41 worth it22
What they complain about 504 reviewsWords that stand out save corruption37 crashes on launch29 unplayable11
| ||||||
| Says nothing about the game | 37.5%750 | 18.8% | 36.0% | 1.0× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
What they praise 375 reviews
What they complain about 375 reviews
| ||||||
| Monetisation and DLC | 36.2%724 | 18.1% | 36.0% | 1.0× | 40.0% praise40.0% gripe, 19.9% both | 66.6% |
What they praise 361 reviews
What they complain about 361 reviews
| ||||||
| Price and value | 36.1%723 | 18.1% | 36.0% | 1.0× | 40.0% praise40.0% gripe, 19.9% both | 66.7% |
What they praise 361 reviews
What they complain about 361 reviews
| ||||||
| Mods and user content | 36.1%722 | 18.1% | 36.0% | 1.0× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
What they praise 360 reviews
What they complain about 360 reviews
| ||||||
| Amount of content | 36.0%721 | 18.0% | 36.0% | 1.0× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
What they praise 360 reviews
What they complain about 360 reviews
| ||||||
| Difficulty and balance | 36.0%720 | 18.0% | 36.0% | 1.0× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
What they praise 360 reviews
What they complain about 360 reviews
| ||||||
| Controls and interface | 35.9%719 | 17.9% | 34.0% | 0.9× | 40.0% praise40.0% gripe, 19.9% both | 66.6% |
What they praise 358 reviews
What they complain about 358 reviews
| ||||||
| Audio and music | 35.9%718 | 17.9% | 34.0% | 0.9× | 40.0% praise40.0% gripe, 19.9% both | 66.6% |
What they praise 358 reviews
What they complain about 358 reviews
| ||||||
| Graphics and art | 35.9%717 | 17.9% | 34.0% | 0.9× | 40.0% praise40.0% gripe, 19.9% both | 66.7% |
What they praise 358 reviews
What they complain about 358 reviews
| ||||||
| Atmosphere and feel | 35.8%716 | 17.9% | 34.0% | 0.9× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
What they praise 357 reviews
What they complain about 357 reviews
| ||||||
| Licensed content | 21.6%433 | 10.8% | 20.0% | 0.9× | 40.0% praise40.0% gripe, 20.0% both | 66.5% |
What they praise 216 reviews
What they complain about 216 reviews
| ||||||
| Tutorial and learning | 21.6%432 | 10.8% | 20.0% | 0.9× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
What they praise 216 reviews
What they complain about 216 reviews
| ||||||
| Language and localisation | 21.6%431 | 10.8% | 20.0% | 0.9× | 40.1% praise40.1% gripe, 19.9% both | 66.6% |
What they praise 214 reviews
What they complain about 214 reviews
| ||||||
| Accessibility and options | 21.5%430 | 10.8% | 20.0% | 0.9× | 40.1% praise40.1% gripe, 19.9% both | 66.5% |
What they praise 214 reviews
What they complain about 214 reviews
| ||||||
| Hardware and compatibility | 21.4%429 | 10.7% | 20.0% | 0.9× | 40.1% praise40.1% gripe, 19.9% both | 66.7% |
What they praise 214 reviews
What they complain about 214 reviews
| ||||||
| Publisher and platform policy | 21.4%428 | 10.7% | 20.0% | 0.9× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
What they praise 213 reviews
What they complain about 213 reviews
| ||||||
| Updates and developer support | 21.3%427 | 10.7% | 20.0% | 0.9× | 40.0% praise40.0% gripe, 20.0% both | 66.5% |
What they praise 213 reviews
What they complain about 213 reviews
| ||||||
| Community and players | 21.3%426 | 10.7% | 20.0% | 0.9× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
What they praise 213 reviews
What they complain about 213 reviews
| ||||||
| Multiplayer and online | 21.2%425 | 10.6% | 20.0% | 0.9× | 40.1% praise40.1% gripe, 19.9% both | 66.6% |
What they praise 211 reviews
What they complain about 211 reviews
| ||||||
| VR and headsets | 0.0%0 | 0.0% | 0.0% | not measured | none raised it | none raised it |
Who said it
Steam records beside every review how long its writer had played, whether mostly on a Steam Deck, whether the game was in early access, and whether they got it free. Each kind of reviewer is set against everyone else, and a difference is listed only where it is 4 times the gap chance alone typically makes and at least 2 points wide. A kind with fewer than 100 reviews is not compared.
| Who wrote it | Reviews | Recommended |
|---|---|---|
| Time played when they wrote it | ||
| Under 2 hours | 0 | too few to say |
| 2 to 10 hours | 0 | too few to say |
| 10 to 30 hours | 0 | too few to say |
| 30 to 100 hours | 0 | too few to say |
| 100 hours or more | 0 | too few to say |
| Where they played | ||
| Mostly on a Steam Deck | 0 | too few to say |
| Mostly elsewhere | 0 | too few to say |
| When they wrote it | ||
| During early access | 0 | too few to say |
| After release | 0 | too few to say |
| How they got the game | ||
| Got it free | 0 | too few to say |
| Paid for it | 0 | too few to say |
Where they differ
No kind of reviewer praises, complains about or recommends anything more or less often than everyone else by more than chance.
What this game's players talk about that others' do not
Found by reading reviews of this game chosen to be as unlike each other as possible, and named only where at least three of them raise the same thing. These rows carry no rate: the model that counts the table above has not been trained on them, so what is shown is the finding and the reviews it rests on.
- Mud and terrain deformation a form of Gameplay and mechanics
How the ground deforms under a wheel, and whether it behaves the way a driver expects; not the truck roster and not the scenery.
3 of the 3 reviews it was found in
-
A review that is mostly about the mud, at the length reviews about one mechanic tend to run to. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things.
-
A review that is mostly about the mud, at the length reviews about one mechanic tend to run to. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things.
-
A review that is mostly about the mud, at the length reviews about one mechanic tend to run to. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things.
-
The top of the pile
The 50 reviews Steam ranks as most helpful, which is roughly what a reader sees before deciding. Every rate above is measured against the whole corpus instead.
Read all 1 of them
-
The most helpful review in the corpus, which is what a reader skimming the store page sees before anything this tool counted.
The most helpful review in the corpus, which is what a reader skimming the store page sees before anything this tool counted. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things.
What language it was said in
41.2% of these reviews are not in English. Steam shows a reader their own language by default, so most of this argument is one they never see.
2.0% of them are in a language the reader declines outright (Indonesian): the reference set holds too few claims there to keep its accuracy promise, so it says nothing rather than guessing. That is a gap in the labels, not a finding about this game.
39.2% are in a language the reader has to be surer about than English before it will answer at all (Chinese (simplified), Russian). A corpus weighted towards those declines more than usual for that reason rather than for anything about the game, and what it wants is more labels in those languages.
- English58.8%1,200
- Chinese (simplified)29.4%600
- Russian9.8%200
- Indonesian2.0%40
How far to trust this
- Read by
- Game Review Reader (run a-reader), a fine-tune of Alibaba-NLP/gte-multilingual-base, trained on label set 0123456789abcdef, answering a claim only where it clears the line drawn for its subject and the one for its language (rule fedcba9876543210)
- Claims read
- 6,000 from 2,000 reviews
- Claims it would not answer
- 3,600 (60.0%), counted as unclassified rather than filed under a best guess, above the 53.0% it usually declines on a game it has never seen
- Reviews it said nothing about
- 40 (2.0% of those read), where no point cleared the threshold
No claims have been labelled for this game, so how often the model is wrong here has not been measured. What is measured is how it does on 8 games it had never seen, over 3,769 labelled claims: it answers 57.7% of them and names the same subject a separate labeller did 74.9% of the time when it does, declining the rest rather than guessing. Those labellers were themselves language models. Expect this corpus to be somewhere near that and treat every rate as provisional.
These rates are a census, not a survey: every review Valve serves was counted, so there is no sampling error to report. What they do carry is classifier error, which is what the agreement figure above measures.
A Third Game With A Rather Long Name
- Reviews counted
- 3,800
- Captured
- 3,800 of the 3,820 Valve reports (99.80%)
- Steam calls it
- Mostly Positive
- Snapshot
- 27 September 2025
- Brought up to date
- 3 October 2025, 40 added or edited since the snapshot
The capture was brought up to date on 3 October 2025 and these counts were made before that, on the corpus as it stood on 27 September 2025. Read it again to count what arrived.
The top of the pile overstates updates and developer support by 1.9×: 9 of the 50 reviews Steam ranks most helpful raise it, against 9.6% of all 3,800.
Players love mixing two factions every run, which 410 reviews praise, and many find the combat tight. The clearest complaint is a black screen on startup, raised in 48 reviews and tied to performance as well. Others say it "stutters in big fights".
Written by a language model on the computer that made this page, from the points below. Every number and quotation in it is one those points give; a sentence with any other was left out.
What they praise and complain about
Points that say the same thing, in the same words or others, grouped by meaning, each with how many reviews make it.
Praised
- Love how you can mix and match factions.410 reviews, 10.8% · Gameplay and mechanics
Mixing two factions every run keeps it fresh after a hundred hours.
- The combat is tight.160 reviews, 4.2% · Gameplay and mechanicsAlso said about Controls and interface
Complained about
- Black screen on startup.48 reviews, 1.3% · Bugs and crashesAlso said about Performance
Every time I start it the screen stays black <until I alt-tab>.
- Saves get corrupted.21 reviews, 0.6% · Bugs and crashes
When it was said
Reviews per month, Jan 2023 to Dec 2024, against a busiest month of 519 in Dec 2024. The line is the share of each month that recommended the game, from none at the bottom to all at the top; the dashed line is half. It falls furthest in Jan 2023, when 66.4% of 220 reviews recommended the game. Point at a month to read it.
Around its updates
Nothing this game's developer had posted on Steam by 2 October 2025 reads as an update.
What players talk about
70.0% of all 3,800 reviews recommend this game. Every figure in the last column is worth reading against that.
A review counts towards every category it says something about, so these add up to more than 100%. Select a row to read the reviews behind it.
What the columns mean, and where the quoted reviews come from
- Mention rate
- The share of all reviews that say something about the category, with the number of reviews beside it. These are the headline figures and they add up to more than 100%.
- Main subject
- The share whose single main subject is that category. Unlike mention rates, these add up to the number of reviews.
- Top of the pile
- The mention rate over the 50 reviews Steam ranks as most helpful, which is roughly what somebody sees before deciding.
- Bias
- How much the top of the pile overstates a category. 0× means none of those few dozen reviews raised it, which is weak evidence rather than proof of absence.
- Said about it
- Of the reviews raising a category, how many praise it, how many complain about it, and how many do both. A subject half the players love and half hate, and one every player has mixed feelings about, are different findings that a single share hides.
- Recommended
- The share of the reviews raising a category that still recommended the game, against this game's own baseline. Marked warm or cold only where the gap is wider than the number of reviews behind it can explain.
The reviews behind a row are a couple from the top of the pile where the category reaches it, and the rest drawn at random from everything filed there, so they are evidence rather than a selection.
| Graphics and art | 25.1%955 | 12.6% | 46.0% | 1.8× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|---|---|---|---|---|---|---|
|
Measured on this row: of the 52 labelled claims about it, the model found 65.4%; of the claims it filed here, 61.8% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 16.8% the model read. That is an estimate from 52 labels, and it moves with them. What they praise 477 reviews
What they complain about 477 reviews
| ||||||
| Atmosphere and feel | 25.1%954 | 12.6% | 46.0% | 1.8× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
|
Measured on this row: of the 53 labelled claims about it, the model found 66.0%; of the claims it filed here, 62.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 16.7% the model read. That is an estimate from 53 labels, and it moves with them. What they praise 477 reviews
What they complain about 477 reviews
| ||||||
| Story and writing | 25.1%953 | 12.5% | 46.0% | 1.8× | 40.0% praise40.0% gripe, 19.9% both | 66.6% |
|
Measured on this row: of the 54 labelled claims about it, the model found 66.7%; of the claims it filed here, 63.2% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 16.7% the model read. That is an estimate from 54 labels, and it moves with them. What they praise 475 reviews
What they complain about 475 reviews
| ||||||
| Overall verdict | 25.0%951 | 12.5% | 46.0% | 1.8× | 40.0% praise40.0% gripe, 19.9% both | 66.7% |
|
Measured on this row: of the 56 labelled claims about it, the model found 66.1%; of the claims it filed here, 62.7% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 16.7% the model read. That is an estimate from 56 labels, and it moves with them. What they praise 475 reviews
What they complain about 475 reviews
| ||||||
| Genre and comparisons | 25.0%950 | 12.5% | 46.0% | 1.8× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 57 labelled claims about it, the model found 66.7%; of the claims it filed here, 63.3% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 16.7% the model read. That is an estimate from 57 labels, and it moves with them. What they praise 474 reviews
What they complain about 474 reviews
| ||||||
| Gameplay and mechanics | 25.0%949 | 12.5% | 46.0% | 1.8× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 58 labelled claims about it, the model found 65.5%; of the claims it filed here, 62.3% were labelled that way. A rate built on a subject it misses is a floor, not a count. Where it was read wrongly, it was most often read as gameplay (7 of the labelled claims). Corrected for those errors, the share of claims about this is about 0.0%, against the 16.6% the model read. That is an estimate from 58 labels, and it moves with them. What they praise 474 reviewsWords that stand out best in years41 worth it22
What they complain about 474 reviewsWords that stand out save corruption37 crashes on launch29 unplayable11
| ||||||
| Bugs and crashes | 24.9%948 | 12.5% | 46.0% | 1.8× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
|
Measured on this row: of the 59 labelled claims about it, the model found 66.1%; of the claims it filed here, 62.9% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 16.6% the model read. That is an estimate from 59 labels, and it moves with them. What they praise 474 reviewsWords that stand out best in years41 worth it22
What they complain about 474 reviewsWords that stand out save corruption37 crashes on launch29 unplayable11
| ||||||
| Performance | 24.9%947 | 12.4% | 46.0% | 1.8× | 40.0% praise40.0% gripe, 19.9% both | 66.6% |
|
Measured on this row: of the 60 labelled claims about it, the model found 66.7%; of the claims it filed here, 63.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 16.6% the model read. That is an estimate from 60 labels, and it moves with them. What they praise 472 reviewsWords that stand out best in years41 worth it22
What they complain about 472 reviewsWords that stand out save corruption37 crashes on launch29 unplayable11
| ||||||
| Says nothing about the game | 19.7%750 | 9.9% | 36.0% | 1.8× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
|
Measured on this row: of the 55 labelled claims about it, the model found 65.5%; of the claims it filed here, 62.1% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 13.2% the model read. That is an estimate from 55 labels, and it moves with them. What they praise 375 reviews
What they complain about 375 reviews
| ||||||
| Community and players | 17.5%664 | 8.7% | 32.0% | 1.8× | 40.0% praise40.0% gripe, 19.9% both | 66.6% |
|
Measured on this row: of the 43 labelled claims about it, the model found 65.1%; of the claims it filed here, 60.9% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 11.6% the model read. That is an estimate from 43 labels, and it moves with them. What they praise 331 reviews
What they complain about 331 reviews
| ||||||
| Multiplayer and online | 17.4%663 | 8.7% | 32.0% | 1.8× | 40.0% praise40.0% gripe, 19.9% both | 66.7% |
|
Measured on this row: of the 44 labelled claims about it, the model found 65.9%; of the claims it filed here, 61.7% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 11.6% the model read. That is an estimate from 44 labels, and it moves with them. What they praise 331 reviews
What they complain about 331 reviews
| ||||||
| Monetisation and DLC | 17.4%662 | 8.7% | 32.0% | 1.8× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 45 labelled claims about it, the model found 66.7%; of the claims it filed here, 62.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 11.6% the model read. That is an estimate from 45 labels, and it moves with them. What they praise 330 reviews
What they complain about 330 reviews
| ||||||
| Price and value | 17.4%661 | 8.7% | 32.0% | 1.8× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 46 labelled claims about it, the model found 65.2%; of the claims it filed here, 61.2% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 11.6% the model read. That is an estimate from 46 labels, and it moves with them. What they praise 330 reviews
What they complain about 330 reviews
| ||||||
| Mods and user content | 17.4%660 | 8.7% | 32.0% | 1.8× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
|
Measured on this row: of the 47 labelled claims about it, the model found 66.0%; of the claims it filed here, 62.0% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 11.6% the model read. That is an estimate from 47 labels, and it moves with them. What they praise 330 reviews
What they complain about 330 reviews
| ||||||
| Amount of content | 17.3%659 | 8.7% | 32.0% | 1.8× | 40.0% praise40.0% gripe, 19.9% both | 66.6% |
|
Measured on this row: of the 48 labelled claims about it, the model found 66.7%; of the claims it filed here, 62.7% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 11.6% the model read. That is an estimate from 48 labels, and it moves with them. What they praise 328 reviews
What they complain about 328 reviews
| ||||||
| Difficulty and balance | 17.3%658 | 8.7% | 32.0% | 1.8× | 40.0% praise40.0% gripe, 19.9% both | 66.6% |
|
Measured on this row: of the 49 labelled claims about it, the model found 65.3%; of the claims it filed here, 61.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 11.5% the model read. That is an estimate from 49 labels, and it moves with them. What they praise 328 reviews
What they complain about 328 reviews
| ||||||
| Controls and interface | 17.3%657 | 8.6% | 32.0% | 1.9× | 40.0% praise40.0% gripe, 19.9% both | 66.7% |
|
Measured on this row: of the 50 labelled claims about it, the model found 66.0%; of the claims it filed here, 62.3% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 11.5% the model read. That is an estimate from 50 labels, and it moves with them. What they praise 328 reviews
What they complain about 328 reviews
| ||||||
| Audio and music | 17.3%656 | 8.6% | 32.0% | 1.9× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 51 labelled claims about it, the model found 66.7%; of the claims it filed here, 63.0% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 11.5% the model read. That is an estimate from 51 labels, and it moves with them. What they praise 327 reviews
What they complain about 327 reviews
| ||||||
| Licensed content | 9.8%371 | 4.9% | 18.0% | 1.8× | 40.1% praise40.1% gripe, 19.9% both | 66.6% |
|
Measured on this row: of the 36 labelled claims about it, the model found 66.7%; of the claims it filed here, 61.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. What they praise 184 reviews
What they complain about 184 reviews
| ||||||
| Tutorial and learning | 9.7%370 | 4.9% | 18.0% | 1.8× | 40.1% praise40.1% gripe, 19.9% both | 66.5% |
|
Measured on this row: of the 37 labelled claims about it, the model found 64.9%; of the claims it filed here, 60.0% were labelled that way. A rate built on a subject it misses is a floor, not a count. What they praise 184 reviews
What they complain about 184 reviews
| ||||||
| Language and localisation | 9.7%369 | 4.8% | 18.0% | 1.9× | 40.1% praise40.1% gripe, 19.9% both | 66.7% |
|
Measured on this row: of the 38 labelled claims about it, the model found 65.8%; of the claims it filed here, 61.0% were labelled that way. A rate built on a subject it misses is a floor, not a count. What they praise 184 reviews
What they complain about 184 reviews
| ||||||
| Accessibility and options | 9.7%368 | 4.8% | 18.0% | 1.9× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 39 labelled claims about it, the model found 66.7%; of the claims it filed here, 61.9% were labelled that way. A rate built on a subject it misses is a floor, not a count. What they praise 183 reviews
What they complain about 183 reviews
| ||||||
| Hardware and compatibility | 9.7%367 | 4.8% | 18.0% | 1.9× | 40.0% praise40.0% gripe, 20.0% both | 66.5% |
|
Measured on this row: of the 40 labelled claims about it, the model found 65.0%; of the claims it filed here, 60.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 6.4% the model read. That is an estimate from 40 labels, and it moves with them. What they praise 183 reviews
What they complain about 183 reviews
| ||||||
| Publisher and platform policy | 9.6%366 | 4.8% | 18.0% | 1.9× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
|
Measured on this row: of the 41 labelled claims about it, the model found 65.9%; of the claims it filed here, 61.4% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 6.4% the model read. That is an estimate from 41 labels, and it moves with them. What they praise 183 reviews
What they complain about 183 reviews
| ||||||
| Updates and developer support | 9.6%365 | 4.8% | 18.0% | 1.9× | 40.1% praise40.1% gripe, 19.9% both | 66.6% |
|
Measured on this row: of the 42 labelled claims about it, the model found 66.7%; of the claims it filed here, 62.2% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 6.4% the model read. That is an estimate from 42 labels, and it moves with them. What they praise 181 reviews
What they complain about 181 reviews
| ||||||
| VR and headsets | 0.0%0 | 0.0% | 0.0% | not measured | none raised it | none raised it |
Who said it
Steam records beside every review how long its writer had played, whether mostly on a Steam Deck, whether the game was in early access, and whether they got it free. Each kind of reviewer is set against everyone else, and a difference is listed only where it is 4 times the gap chance alone typically makes and at least 2 points wide. A kind with fewer than 100 reviews is not compared.
| Who wrote it | Reviews | Recommended |
|---|---|---|
| Time played when they wrote it | ||
| Under 2 hours | 418 | 50.0% |
| 2 to 10 hours | 874 | 69.9% |
| 10 to 30 hours | 950 | 70.0% |
| 30 to 100 hours | 874 | 69.9% |
| 100 hours or more | 684 | 69.9% |
| Where they played | ||
| Mostly on a Steam Deck | 12 | too few to say |
| Mostly elsewhere | 3,788 | too few others to set it against |
| When they wrote it | ||
| During early access | 1,140 | 70.0% |
| After release | 2,660 | 70.0% |
| How they got the game | ||
| Got it free | 152 | 69.7% |
| Paid for it | 3,648 | 70.0% |
Where they differ
- Reviewers with 100 hours or more played complain about amount of content in 32.7% of their reviews, against 8.5% of everyone else.
- Reviewers with under 2 hours played praise tutorial and learning in 26.3% of their reviews, against 4.8% of everyone else.
- Reviewers with under 2 hours played recommend the game in 50.0% of their reviews, against 69.9% of everyone else.
- Reviewers with 10 to 30 hours played complain about amount of content in 8.5% of their reviews, against 14.4% of everyone else.
- Reviewers with 2 to 10 hours played complain about amount of content in 8.6% of their reviews, against 14.2% of everyone else.
- Reviewers with 30 to 100 hours played complain about amount of content in 8.6% of their reviews, against 14.2% of everyone else.
What this game's players talk about that others' do not
Found by reading reviews of this game chosen to be as unlike each other as possible, and named only where at least three of them raise the same thing. These rows carry no rate: the model that counts the table above has not been trained on them, so what is shown is the finding and the reviews it rests on.
- Mud and terrain deformation a form of Gameplay and mechanics
How the ground deforms under a wheel, and whether it behaves the way a driver expects; not the truck roster and not the scenery.
3 of the 3 reviews it was found in
-
A review that is mostly about the mud, at the length reviews about one mechanic tend to run to. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things.
-
A review that is mostly about the mud, at the length reviews about one mechanic tend to run to. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things.
-
A review that is mostly about the mud, at the length reviews about one mechanic tend to run to. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things.
-
The top of the pile
The 50 reviews Steam ranks as most helpful, which is roughly what a reader sees before deciding. Every rate above is measured against the whole corpus instead.
Read all 1 of them
-
The most helpful review in the corpus, which is what a reader skimming the store page sees before anything this tool counted.
The most helpful review in the corpus, which is what a reader skimming the store page sees before anything this tool counted. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things.
What language it was said in
41.2% of these reviews are not in English. Steam shows a reader their own language by default, so most of this argument is one they never see.
2.0% of them are in a language the reader declines outright (Indonesian): the reference set holds too few claims there to keep its accuracy promise, so it says nothing rather than guessing. That is a gap in the labels, not a finding about this game.
39.2% are in a language the reader has to be surer about than English before it will answer at all (Chinese (simplified), Russian). A corpus weighted towards those declines more than usual for that reason rather than for anything about the game, and what it wants is more labels in those languages.
- English58.8%2,280
- Chinese (simplified)29.4%1,140
- Russian9.8%380
- Indonesian2.0%76
How far to trust this
- Read by
- Game Review Reader (run a-reader), a fine-tune of Alibaba-NLP/gte-multilingual-base, trained on label set 0123456789abcdef, answering a claim only where it clears the line drawn for its subject and the one for its language (rule fedcba9876543210)
- Claims read
- 11,400 from 3,800 reviews
- Claims it would not answer
- 6,840 (60.0%), counted as unclassified rather than filed under a best guess, above the 53.0% it usually declines on a game it has never seen
- Reviews it said nothing about
- 76 (2.0% of those read), where no point cleared the threshold
Of the 290 labelled claims this model was willing to answer, it named the same subject a separate labeller did 80.0% of the time, somewhere in [75.0%, 84.2%] with 95% confidence. The labeller was itself a language model, so that is agreement, not accuracy: two models can be wrong together, most easily on sarcasm and on claims that sit between subjects.
That figure covers the 50.0% of those claims it answered. It declined the other 50.0%, which are counted as unclassified everywhere on this page rather than being filed under a best guess.
A further 19 labelled claims were left out because this build takes their reviews apart differently from the build they were labelled under, so no reading corresponds to them.
Averaged over the subjects rather than over the claims, so a rare one counts as much as a common one, that comes to 0.64 on a scale where 1 is perfect agreement. Every one of the 26 subjects with enough labels to judge is found in at least a quarter of the claims making it.
Where it disagrees most, 7 claims the labeller called Gameplay and mechanics were read as gameplay. That is a boundary between two subjects rather than a mistake about one of them.
420 of those claims were read a second time, by a different labeller working blind, and the two reached the same subject on 92.1% of them. On those settled claims the model agrees 85.5%, somewhere in [81.7%, 88.7%], which is the nearest thing to accuracy a set labelled by models can produce: a label two independent readings reached is one worth scoring against. Of the settled claims, 321 were ones neither labeller called contested, and on those it agrees 93.1% of the time. On the 33 they read differently there is no single answer to be right about, and it lands on one of their two 81.8% of the time.
These rates are a census, not a survey: every review Valve serves was counted, so there is no sampling error to report. What they do carry is classifier error, which is what the agreement figure above measures.
A Fourth Game
- Reviews counted
- 7,400
- Captured
- 7,400 of the 7,420 Valve reports (99.80%)
- Steam calls it
- Mostly Positive
- Snapshot
- 27 September 2025
- Brought up to date
- 3 October 2025, 40 added or edited since the snapshot
The capture was brought up to date on 3 October 2025 and these counts were made before that, on the corpus as it stood on 27 September 2025. Read it again to count what arrived.
The top of the pile overstates performance by 3.7×: 28 of the 50 reviews Steam ranks most helpful raise it, against 15.2% of all 7,400.
Players love mixing two factions every run, which 410 reviews praise, and many find the combat tight. The clearest complaint is a black screen on startup, raised in 48 reviews and tied to performance as well. Others say it "stutters in big fights".
Written by a language model on the computer that made this page, from the points below. Every number and quotation in it is one those points give; a sentence with any other was left out.
What they praise and complain about
Points that say the same thing, in the same words or others, grouped by meaning, each with how many reviews make it.
Praised
- Love how you can mix and match factions.410 reviews, 5.5% · Gameplay and mechanics
Mixing two factions every run keeps it fresh after a hundred hours.
- The combat is tight.160 reviews, 2.2% · Gameplay and mechanicsAlso said about Controls and interface
Complained about
- Black screen on startup.48 reviews, 0.6% · Bugs and crashesAlso said about Performance
Every time I start it the screen stays black <until I alt-tab>.
- Saves get corrupted.21 reviews, 0.3% · Bugs and crashes
When it was said
Reviews per month, Jan 2023 to Dec 2024, against a busiest month of 459 in Dec 2024. The line is the share of each month that recommended the game, from none at the bottom to all at the top; the dashed line is half. It falls furthest in Jan 2023, when 66.2% of 160 reviews recommended the game. Point at a month to read it.
Around its updates
Nothing this game's developer had posted on Steam by 2 October 2025 reads as an update.
What players talk about
70.0% of all 7,400 reviews recommend this game. Every figure in the last column is worth reading against that.
A review counts towards every category it says something about, so these add up to more than 100%. Select a row to read the reviews behind it.
What the columns mean, and where the quoted reviews come from
- Mention rate
- The share of all reviews that say something about the category, with the number of reviews beside it. These are the headline figures and they add up to more than 100%.
- Main subject
- The share whose single main subject is that category. Unlike mention rates, these add up to the number of reviews.
- Top of the pile
- The mention rate over the 50 reviews Steam ranks as most helpful, which is roughly what somebody sees before deciding.
- Bias
- How much the top of the pile overstates a category. 0× means none of those few dozen reviews raised it, which is weak evidence rather than proof of absence.
- Said about it
- Of the reviews raising a category, how many praise it, how many complain about it, and how many do both. A subject half the players love and half hate, and one every player has mixed feelings about, are different findings that a single share hides.
- Recommended
- The share of the reviews raising a category that still recommended the game, against this game's own baseline. Marked warm or cold only where the gap is wider than the number of reviews behind it can explain.
The reviews behind a row are a couple from the top of the pile where the category reaches it, and the rest drawn at random from everything filed there, so they are evidence rather than a selection.
| Gameplay and mechanics | 15.2%1,125 | 7.6% | 56.0% | 3.7× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
|---|---|---|---|---|---|---|
|
Measured on this row: of the 58 labelled claims about it, the model found 65.5%; of the claims it filed here, 62.3% were labelled that way. A rate built on a subject it misses is a floor, not a count. Where it was read wrongly, it was most often read as gameplay (7 of the labelled claims). Corrected for those errors, the share of claims about this is about 0.0%, against the 10.1% the model read. That is an estimate from 58 labels, and it moves with them. What they praise 562 reviewsWords that stand out best in years41 worth it22
What they complain about 562 reviewsWords that stand out save corruption37 crashes on launch29 unplayable11
| ||||||
| Bugs and crashes | 15.2%1,124 | 7.6% | 56.0% | 3.7× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 59 labelled claims about it, the model found 66.1%; of the claims it filed here, 62.9% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 10.1% the model read. That is an estimate from 59 labels, and it moves with them. What they praise 561 reviewsWords that stand out best in years41 worth it22
What they complain about 561 reviewsWords that stand out save corruption37 crashes on launch29 unplayable11
| ||||||
| Performance | 15.2%1,123 | 7.6% | 56.0% | 3.7× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 60 labelled claims about it, the model found 66.7%; of the claims it filed here, 63.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 10.1% the model read. That is an estimate from 60 labels, and it moves with them. What they praise 561 reviewsWords that stand out best in years41 worth it22
What they complain about 561 reviewsWords that stand out save corruption37 crashes on launch29 unplayable11
| ||||||
| Amount of content | 11.3%835 | 5.6% | 40.0% | 3.5× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 48 labelled claims about it, the model found 66.7%; of the claims it filed here, 62.7% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 7.5% the model read. That is an estimate from 48 labels, and it moves with them. What they praise 417 reviews
What they complain about 417 reviews
| ||||||
| Difficulty and balance | 11.3%834 | 5.6% | 40.0% | 3.5× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
|
Measured on this row: of the 49 labelled claims about it, the model found 65.3%; of the claims it filed here, 61.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 7.5% the model read. That is an estimate from 49 labels, and it moves with them. What they praise 417 reviews
What they complain about 417 reviews
| ||||||
| Controls and interface | 11.3%833 | 5.6% | 40.0% | 3.6× | 40.0% praise40.0% gripe, 19.9% both | 66.6% |
|
Measured on this row: of the 50 labelled claims about it, the model found 66.0%; of the claims it filed here, 62.3% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 7.5% the model read. That is an estimate from 50 labels, and it moves with them. What they praise 415 reviews
What they complain about 415 reviews
| ||||||
| Audio and music | 11.2%832 | 5.6% | 40.0% | 3.6× | 40.0% praise40.0% gripe, 19.9% both | 66.6% |
|
Measured on this row: of the 51 labelled claims about it, the model found 66.7%; of the claims it filed here, 63.0% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 7.5% the model read. That is an estimate from 51 labels, and it moves with them. What they praise 415 reviews
What they complain about 415 reviews
| ||||||
| Graphics and art | 11.2%831 | 5.6% | 40.0% | 3.6× | 40.0% praise40.0% gripe, 19.9% both | 66.7% |
|
Measured on this row: of the 52 labelled claims about it, the model found 65.4%; of the claims it filed here, 61.8% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 7.5% the model read. That is an estimate from 52 labels, and it moves with them. What they praise 415 reviews
What they complain about 415 reviews
| ||||||
| Atmosphere and feel | 11.2%830 | 5.6% | 40.0% | 3.6× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 53 labelled claims about it, the model found 66.0%; of the claims it filed here, 62.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 7.5% the model read. That is an estimate from 53 labels, and it moves with them. What they praise 414 reviews
What they complain about 414 reviews
| ||||||
| Story and writing | 11.2%829 | 5.6% | 40.0% | 3.6× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 54 labelled claims about it, the model found 66.7%; of the claims it filed here, 63.2% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 7.5% the model read. That is an estimate from 54 labels, and it moves with them. What they praise 414 reviews
What they complain about 414 reviews
| ||||||
| Overall verdict | 11.2%827 | 5.6% | 40.0% | 3.6× | 40.0% praise40.0% gripe, 19.9% both | 66.6% |
|
Measured on this row: of the 56 labelled claims about it, the model found 66.1%; of the claims it filed here, 62.7% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 7.5% the model read. That is an estimate from 56 labels, and it moves with them. What they praise 412 reviews
What they complain about 412 reviews
| ||||||
| Genre and comparisons | 11.2%826 | 5.6% | 40.0% | 3.6× | 40.0% praise40.0% gripe, 19.9% both | 66.6% |
|
Measured on this row: of the 57 labelled claims about it, the model found 66.7%; of the claims it filed here, 63.3% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 7.4% the model read. That is an estimate from 57 labels, and it moves with them. What they praise 412 reviews
What they complain about 412 reviews
| ||||||
| Says nothing about the game | 10.1%750 | 5.1% | 36.0% | 3.6× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
|
Measured on this row: of the 55 labelled claims about it, the model found 65.5%; of the claims it filed here, 62.1% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 6.8% the model read. That is an estimate from 55 labels, and it moves with them. What they praise 375 reviews
What they complain about 375 reviews
| ||||||
| Accessibility and options | 7.4%544 | 3.7% | 26.0% | 3.5× | 40.0% praise40.0% gripe, 19.9% both | 66.5% |
|
Measured on this row: of the 39 labelled claims about it, the model found 66.7%; of the claims it filed here, 61.9% were labelled that way. A rate built on a subject it misses is a floor, not a count. What they praise 271 reviews
What they complain about 271 reviews
| ||||||
| Hardware and compatibility | 7.3%543 | 3.7% | 26.0% | 3.5× | 40.0% praise40.0% gripe, 19.9% both | 66.7% |
|
Measured on this row: of the 40 labelled claims about it, the model found 65.0%; of the claims it filed here, 60.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 4.9% the model read. That is an estimate from 40 labels, and it moves with them. What they praise 271 reviews
What they complain about 271 reviews
| ||||||
| Publisher and platform policy | 7.3%542 | 3.7% | 26.0% | 3.5× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 41 labelled claims about it, the model found 65.9%; of the claims it filed here, 61.4% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 4.9% the model read. That is an estimate from 41 labels, and it moves with them. What they praise 270 reviews
What they complain about 270 reviews
| ||||||
| Updates and developer support | 7.3%541 | 3.6% | 26.0% | 3.6× | 40.0% praise40.0% gripe, 20.0% both | 66.5% |
|
Measured on this row: of the 42 labelled claims about it, the model found 66.7%; of the claims it filed here, 62.2% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 4.9% the model read. That is an estimate from 42 labels, and it moves with them. What they praise 270 reviews
What they complain about 270 reviews
| ||||||
| Community and players | 7.3%540 | 3.6% | 26.0% | 3.6× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
|
Measured on this row: of the 43 labelled claims about it, the model found 65.1%; of the claims it filed here, 60.9% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 4.9% the model read. That is an estimate from 43 labels, and it moves with them. What they praise 270 reviews
What they complain about 270 reviews
| ||||||
| Multiplayer and online | 7.3%539 | 3.6% | 26.0% | 3.6× | 40.0% praise40.0% gripe, 19.9% both | 66.6% |
|
Measured on this row: of the 44 labelled claims about it, the model found 65.9%; of the claims it filed here, 61.7% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 4.9% the model read. That is an estimate from 44 labels, and it moves with them. What they praise 268 reviews
What they complain about 268 reviews
| ||||||
| Monetisation and DLC | 7.3%538 | 3.6% | 26.0% | 3.6× | 40.0% praise40.0% gripe, 19.9% both | 66.5% |
|
Measured on this row: of the 45 labelled claims about it, the model found 66.7%; of the claims it filed here, 62.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 4.8% the model read. That is an estimate from 45 labels, and it moves with them. What they praise 268 reviews
What they complain about 268 reviews
| ||||||
| Price and value | 7.3%537 | 3.6% | 26.0% | 3.6× | 40.0% praise40.0% gripe, 19.9% both | 66.7% |
|
Measured on this row: of the 46 labelled claims about it, the model found 65.2%; of the claims it filed here, 61.2% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 4.8% the model read. That is an estimate from 46 labels, and it moves with them. What they praise 268 reviews
What they complain about 268 reviews
| ||||||
| Mods and user content | 7.2%536 | 3.6% | 26.0% | 3.6× | 40.0% praise40.0% gripe, 20.0% both | 66.6% |
|
Measured on this row: of the 47 labelled claims about it, the model found 66.0%; of the claims it filed here, 62.0% were labelled that way. A rate built on a subject it misses is a floor, not a count. Corrected for those errors, the share of claims about this is about 0.0%, against the 4.8% the model read. That is an estimate from 47 labels, and it moves with them. What they praise 267 reviews
What they complain about 267 reviews
| ||||||
| Licensed content | 3.3%247 | 1.7% | 12.0% | 3.6× | 40.0% praise40.0% gripe, 20.0% both | 66.4% |
|
Measured on this row: of the 36 labelled claims about it, the model found 66.7%; of the claims it filed here, 61.5% were labelled that way. A rate built on a subject it misses is a floor, not a count. What they praise 123 reviews
What they complain about 123 reviews
| ||||||
| Tutorial and learning | 3.3%246 | 1.7% | 12.0% | 3.6× | 40.0% praise40.0% gripe, 20.0% both | 66.7% |
|
Measured on this row: of the 37 labelled claims about it, the model found 64.9%; of the claims it filed here, 60.0% were labelled that way. A rate built on a subject it misses is a floor, not a count. What they praise 123 reviews
What they complain about 123 reviews
| ||||||
| Language and localisation | 3.3%245 | 1.6% | 12.0% | 3.6× | 40.1% praise40.1% gripe, 19.8% both | 66.5% |
|
Measured on this row: of the 38 labelled claims about it, the model found 65.8%; of the claims it filed here, 61.0% were labelled that way. A rate built on a subject it misses is a floor, not a count. What they praise 121 reviews
What they complain about 121 reviews
| ||||||
| VR and headsets | 0.0%0 | 0.0% | 0.0% | not measured | none raised it | none raised it |
Who said it
Steam records beside every review how long its writer had played, whether mostly on a Steam Deck, whether the game was in early access, and whether they got it free. Each kind of reviewer is set against everyone else, and a difference is listed only where it is 4 times the gap chance alone typically makes and at least 2 points wide. A kind with fewer than 100 reviews is not compared.
| Who wrote it | Reviews | Recommended |
|---|---|---|
| Time played when they wrote it | ||
| Under 2 hours | 814 | 50.0% |
| 2 to 10 hours | 1,702 | 70.0% |
| 10 to 30 hours | 1,850 | 70.0% |
| 30 to 100 hours | 1,702 | 70.0% |
| 100 hours or more | 1,332 | 70.0% |
| Where they played | ||
| Mostly on a Steam Deck | 12 | too few to say |
| Mostly elsewhere | 7,388 | too few others to set it against |
| When they wrote it | ||
| During early access | 2,220 | 70.0% |
| After release | 5,180 | 70.0% |
| How they got the game | ||
| Got it free | 296 | 69.9% |
| Paid for it | 7,104 | 70.0% |
Where they differ
- Reviewers with under 2 hours played praise tutorial and learning in 25.4% of their reviews, against 1.6% of everyone else.
- Reviewers with 100 hours or more played complain about amount of content in 31.8% of their reviews, against 5.5% of everyone else.
- Reviewers with under 2 hours played recommend the game in 50.0% of their reviews, against 70.0% of everyone else.
- Reviewers with 10 to 30 hours played complain about amount of content in 5.6% of their reviews, against 11.8% of everyone else.
- Reviewers with 2 to 10 hours played complain about amount of content in 5.5% of their reviews, against 11.7% of everyone else.
- Reviewers with 30 to 100 hours played complain about amount of content in 5.5% of their reviews, against 11.7% of everyone else.
- Reviewers with 10 to 30 hours played praise tutorial and learning in 1.6% of their reviews, against 5.1% of everyone else.
- Reviewers with 2 to 10 hours played praise tutorial and learning in 1.6% of their reviews, against 5.0% of everyone else.
- Reviewers with 30 to 100 hours played praise tutorial and learning in 1.6% of their reviews, against 5.0% of everyone else.
- Reviewers with 100 hours or more played praise tutorial and learning in 1.6% of their reviews, against 4.8% of everyone else.
- Reviewers with under 2 hours played complain about amount of content in 5.5% of their reviews, against 10.9% of everyone else.
What this game's players talk about that others' do not
Found by reading reviews of this game chosen to be as unlike each other as possible, and named only where at least three of them raise the same thing. These rows carry no rate: the model that counts the table above has not been trained on them, so what is shown is the finding and the reviews it rests on.
- Mud and terrain deformation a form of Gameplay and mechanics
How the ground deforms under a wheel, and whether it behaves the way a driver expects; not the truck roster and not the scenery.
3 of the 3 reviews it was found in
-
A review that is mostly about the mud, at the length reviews about one mechanic tend to run to. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things.
-
A review that is mostly about the mud, at the length reviews about one mechanic tend to run to. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things.
-
A review that is mostly about the mud, at the length reviews about one mechanic tend to run to. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things.
-
The top of the pile
The 50 reviews Steam ranks as most helpful, which is roughly what a reader sees before deciding. Every rate above is measured against the whole corpus instead.
Read all 1 of them
-
The most helpful review in the corpus, which is what a reader skimming the store page sees before anything this tool counted.
The most helpful review in the corpus, which is what a reader skimming the store page sees before anything this tool counted. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things. There is a great deal more in this review than the sentence that was counted, which is the ordinary case and the reason the rest of it is offered rather than shown: a reviewer with something to say says it at length, in paragraphs about other things.
What language it was said in
41.2% of these reviews are not in English. Steam shows a reader their own language by default, so most of this argument is one they never see.
2.0% of them are in a language the reader declines outright (Indonesian): the reference set holds too few claims there to keep its accuracy promise, so it says nothing rather than guessing. That is a gap in the labels, not a finding about this game.
39.2% are in a language the reader has to be surer about than English before it will answer at all (Chinese (simplified), Russian). A corpus weighted towards those declines more than usual for that reason rather than for anything about the game, and what it wants is more labels in those languages.
- English58.8%4,440
- Chinese (simplified)29.4%2,220
- Russian9.8%740
- Indonesian2.0%148
How far to trust this
- Read by
- Game Review Reader (run a-reader), a fine-tune of Alibaba-NLP/gte-multilingual-base, trained on label set 0123456789abcdef, answering a claim only where it clears the line drawn for its subject and the one for its language (rule fedcba9876543210)
- Claims read
- 22,200 from 7,400 reviews
- Claims it would not answer
- 13,320 (60.0%), counted as unclassified rather than filed under a best guess, above the 53.0% it usually declines on a game it has never seen
- Reviews it said nothing about
- 148 (2.0% of those read), where no point cleared the threshold
Of the 260 labelled claims this model was willing to answer, it named the same subject a separate labeller did 80.0% of the time, somewhere in [74.7%, 84.4%] with 95% confidence. The labeller was itself a language model, so that is agreement, not accuracy: two models can be wrong together, most easily on sarcasm and on claims that sit between subjects.
That figure covers the 50.0% of those claims it answered. It declined the other 50.0%, which are counted as unclassified everywhere on this page rather than being filed under a best guess.
A further 19 labelled claims were left out because this build takes their reviews apart differently from the build they were labelled under, so no reading corresponds to them.
Averaged over the subjects rather than over the claims, so a rare one counts as much as a common one, that comes to 0.64 on a scale where 1 is perfect agreement. Every one of the 26 subjects with enough labels to judge is found in at least a quarter of the claims making it.
Where it disagrees most, 7 claims the labeller called Gameplay and mechanics were read as gameplay. That is a boundary between two subjects rather than a mistake about one of them.
420 of those claims were read a second time, by a different labeller working blind, and the two reached the same subject on 92.1% of them. On those settled claims the model agrees 85.5%, somewhere in [81.7%, 88.7%], which is the nearest thing to accuracy a set labelled by models can produce: a label two independent readings reached is one worth scoring against. Of the settled claims, 321 were ones neither labeller called contested, and on those it agrees 93.1% of the time. On the 33 they read differently there is no single answer to be right about, and it lands on one of their two 81.8% of the time.
These rates are a census, not a survey: every review Valve serves was counted, so there is no sampling error to report. What they do carry is classifier error, which is what the agreement figure above measures.