Data basis: 6,799 English-language Steam reviews across six games, pulled from the public
appreviews endpoint and read 2026-08-27 — Project Zomboid (800), Songs of Syx, X4: Foundations,
Escape from Tarkov and Mistfall Hunter (1,200 each), Corsair Cove (1,199). Topics are assigned by
keyword matching, described under Method below. Sample windows differ by game because review
volume does.
Running the same topic classifier over six review corpora produces a result worth stating up front: no two of these games have the same leading complaint.
Mistfall Hunter’s players talk about performance. X4’s and Songs of Syx’s talk about the interface. Corsair Cove’s talk about running out of content. Project Zomboid’s talk about mods. Tarkov’s talk about performance again — the only repeat, and at half Mistfall’s rate.
That is not the pattern you would expect. “Bugs” or “performance” could plausibly have topped every list; instead each game’s community has converged on something specific to it.
The second number matters more than the ranking
A topic’s share of negative reviews tells you what unhappy players talk about. It does not tell you whether happy players agree. For that you need the same topic’s share of positive reviews.
Two shapes come out of it.
High ratio — a defect. Songs of Syx’s bugs appear in 13% of negatives and 2% of positives, six to one. Tarkov’s cheating complaints run 14% against 3%, and its bugs 12% against 2%. These are topics that essentially only come up when someone is unhappy. Nobody writes a positive review saying the cheating is fine.
Low ratio — the price of admission. X4’s interface appears in 39% of negatives and 18% of positives, barely two to one. Mistfall Hunter’s performance runs 43% against 19%. When a topic shows up in one in five recommendations, it has stopped being a complaint and become a shared condition — players raise it while still telling you to buy the game.
X4 is the clearest case. Its interface is simultaneously the most-cited reason to dislike it and a thing its advocates bring up unprompted. That is what “steep learning curve” looks like in data.
What this changes about reading a review score
Combining the two dimensions gives a more useful question than “is the score high?”:
| Pattern | Reading | Example here |
|---|---|---|
| Leading topic, low ratio | A known cost the community has accepted | X4 interface, Mistfall performance |
| Leading topic, high ratio | A defect that divides the playerbase | Corsair Cove content depth (27% vs 8%) |
| Mid topic, very high ratio | A specific unresolved problem | Tarkov cheating (14% vs 3%) |
A game whose top complaint has a low ratio is usually hard. A game whose complaints all have high ratios is usually broken. Mistfall Hunter is the awkward case: its leading topic has a low ratio, which normally signals an accepted cost — but no playtime band in its sample clears 50%, so the acceptance is not translating into recommendations.
For the other half of this picture — how much each game improves after the first evening — see the Steam score does not predict your first evening.
Method, stated plainly
Topics are assigned by regular-expression keyword matching over review text, not by human coding. The buckets are performance (fps, lag, stutter, crash, memory, server), bugs, monetisation (price, DLC, pay-to-win, microtransaction), balance (nerf, buff, grind, RNG), content depth (repetitive, shallow, endgame), UI/UX (interface, menu, tutorial, controls) and community/cheating (cheat, hacker, bot, matchmaking).
One review can match several topics, and matching is blind to sentiment within the review — “the UI finally clicked” and “the UI is impenetrable” both count as UI mentions. That is why the negative-to-positive ratio, rather than the raw count, carries the interpretation here. Counts are reproducible from the same endpoint.
What could not be confirmed
- That keyword buckets equal what players mean. They are a proxy. A human-coded sample would likely shift every figure by some margin.
- That the six windows are comparable. They are not: Corsair Cove’s sample covers 26 days, Songs of Syx’s six months.
- Whether leading complaints changed over time within any window — the endpoint returns recent reviews, not dated slices.
- Non-English reviews, excluded from every pull.
Sources
- Steam appreviews endpoint — six pulls, appids 108600, 1162750, 392160, 3932890, 3282300, 1368140, all read 2026-08-27
- Songs of Syx on Steam — read 2026-08-27
- Corsair Cove on Steam — read 2026-08-27
