Steam review intelligence · Sample report

Your angriest Chinese players are your most invested ones.

Slay the Spire 2 is at 91.9% positive in English and 31.4% positive in Simplified Chinese — and Chinese is half of all reviews written. I read 697 recent negative reviews in both languages to find out whether that 60-point split is review-bombing or signal. It is signal — and it is not the signal the English reviews are sending.

Northbound Data Data pulled 2026-08-21 App 2868840 Unsolicited sample — no client relationship
60.5pts
between the English and Chinese positive rates — 91.9% vs 31.4%
62×
more likely to blame the lead designer personally than English reviews
140h
median playtime when a Chinese player writes a negative review — 2.1× the English median
0
refunds in the entire 300-review Chinese sample

Finding 01 — the hole

The store page averages the problem away

Steam shows one headline score: 60.8% positive, "Mixed". That number describes almost nobody. It is the midpoint of two populations sitting at opposite ends:

English
Positive91.9%
Reviews73,966
Share of all37.6%
Simplified Chinese
Positive31.4%
Reviews97,645
Share of all49.6%

In English you have an overwhelmingly loved game. In Chinese you have a failure. And Chinese is not a minority segment you can round off — it is 49.6% of every review the game has ever received, larger than English.

A note on how that 60.8% is built. The blended score is not a neutral baseline to measure against, because roughly half of what goes into it is the Chinese cohort itself. Comparing Chinese to the global average understates the split by half. The honest comparison is cohort against cohort: 91.9% against 31.4%.

Finding 02 — who is writing

These are not drive-by reviews

The first thing to rule out is brigading. It does not survive contact with the data. I compared the 300 most recent Chinese negative reviews against the 397 most recent English ones:

Chinese · n=300
Median playtime140 h
Mean playtime182 h
Over 100 hours66%
Under 2 hours0%
Refunded0
English · n=397
Median playtime68 h
Mean playtime99 h
Over 100 hours36%
Under 2 hours2%
Refunded2

Not one review in the Chinese sample came from a player inside the refund window, and not one asked for their money back. Two thirds of them have over 100 hours in the game. A brigade does not look like this. This is your most engaged cohort filing a complaint, and none of them has left yet.

Finding 03 — the divergence

Same trigger, different verdict

Both audiences are set off by the same thing — balance patches, at almost identical rates. What separates them is what they conclude from it. Share of negative reviews mentioning each theme:

Chinese n=300 English n=397
Blames the lead designer by name
31%
1%
62×
Balance changes / nerfs
31%
29%
1.1×
Too hard / unfair
5%
25%
0.2×
RNG / randomness
4%
18%
0.2×
Worse than the first game
6%
14%
0.5×
Wants new content / early-access pace
7%
8%
0.9×

Read the top row against the second. Both groups are reacting to the same patches at the same rate. English players conclude the game is badly tuned — too hard, too random. Chinese players conclude a specific person is making these decisions and does not play his own game.

Why this matters operationally. The obvious fix suggested by the English reviews — soften the difficulty curve, reduce variance — does almost nothing for the Chinese cohort, where difficulty is a 5% complaint. You could ship that patch, watch your English score recover, and watch the Chinese score stay where it is.

Finding 04 — the buried report

There is a QA report in there

16% of the Chinese sample contains profanity, some of it directed at a named individual. That surface is why this cohort gets written off. But 19% name specific cards, relics and enemies — and the most-upvoted substantive review in the sample is a card-by-card breakdown that any designer would want in their inbox:

怪物攻击欲望与成长都很强，多人模式的血牛机制，更是促进了怪物的加速成长。我喜欢玩的猎人没有烧状态牌的手段，打沙漏只能拼爆发，可是我周转几次牌库，我就被凋萎干掉了。

我喜欢玩的骨妹的幻景被改了，虽然幻景本身是金卡，本就难以抽到，但是现在骨妹处理状态牌的手段就剩下了一个洁净，还只能处理牌库中的状态牌。骨妹的抽牌大多数都是延时收益，防御端和进攻端都启动缓慢或者需要特定组件配合，玩得很难受。

怀念还能玩小卡组循环的时候。
Enemy aggression and scaling are both very high, and the multiplayer HP-bloat mechanic accelerates enemy scaling further. The Hunter I like to play has no way to burn status cards — against the Hourglass all I can do is race for burst, but after cycling my deck a few times Wither kills me.

The Necrobinder's Mirage, which I liked, was changed. Mirage was a rare and hard to draw in the first place, but now her only remaining answer to status cards is Cleanse, which only handles status in the draw pile. Most of her draw is delayed payoff; both her defense and her offense start slowly or need specific pieces to line up. It's miserable to play.

I miss when small-deck cycling was viable.
97 hours played · 30 upvotes · most-upvoted substantive review in sample · abusive passages omitted, not paraphrased

That is a reproducible bug-class report: a character has lost its answer to a status mechanic, and the remaining answer only covers one zone. It was written by someone with 97 hours in the game. It has 30 upvotes. And it is sitting in a language your team does not read, next to reviews that open with insults.

If this game had never been updated I'd give it 70 out of 100. Because it has kept being updated, I can only give it 0. 97 hours played · Simplified Chinese · translated

Named most often in the Chinese sample

卡组17
跃跃欲试15
盛碗虫13
沙漏12
遗物11
A1010
循环8

Card and enemy names are the Chinese-build strings. English equivalents above are my best mapping and should be checked against your own localisation table before anyone acts on them — I would rather flag that than guess silently.

What I would do

Three moves, in order of cost

  1. Put reasoning in the Chinese patch notes, not just the numbers. The 31%/1% split is the signature of changes that read as arbitrary. English players have the developer's blog, streams and Discord to infer intent from; Chinese players have the number that changed. Same patch, no context, so the only available explanation is a person's preference.
  2. Answer one specific complaint in public, by name. The Necrobinder status-card gap above is small, concrete and verifiable. Fixing it silently gets you nothing. Fixing it and saying this came from Chinese player feedback converts the loudest cohort into the one that defends you.
  3. Do not ship a difficulty patch expecting it to move this number. Difficulty is 5% of Chinese complaints and 25% of English ones. That patch is for the other audience.
The asset you still have. Zero refunds across 300 negative reviews, at a 140-hour median. These players have not left and cannot get their money back — they are complaining because they intend to keep playing. That window does not stay open indefinitely.

Limits

What this report cannot tell you

  • It is a recent-300 sample, not the full 97,641. Sorted by recency, so it reflects the current patch cycle and not the game's history. A different patch window would produce different theme weights.
  • Theme classification is keyword-based, and keywords miss sarcasm. The buckets were written after reading the sample, not from a generic sentiment lexicon — a generic one fails badly here, because in this cohort hostility correlates with more playtime, not less. Individual reviews can still be misfiled.
  • "Blames the designer" counts name mentions, not tone. A few of those 31% are neutral or sarcastic references. The 62× ratio is large enough to survive that noise; a 1.5× ratio would not have been.
  • It cannot tell you whether fixing this raises revenue. Review score and conversion are related, but I did not measure your conversion and will not claim a number I did not measure.
  • I am not a Slay the Spire 2 player. The mechanical claims are the reviewers', accurately relayed. Whether the Necrobinder change was correct is your call, not mine.

Method

Every number here is reproducible

One public, documented, unauthenticated Steam endpoint. No login, no scraping of logged-in pages, no CAPTCHA circumvention. The path is not disallowed by store.steampowered.com/robots.txt, checked the day this report was written.

GET store.steampowered.com/appreviews/2868840
  ?json=1&language=schinese&review_type=negative
  &filter=recent&num_per_page=100&cursor=…

Swap language=english for the comparison set and num_per_page=0 for the store-level summary. Run it yourself and you will get these figures, allowing for reviews posted since 2026-08-21.