loader
Game Reviews

Upvotes and Downvotes Across 68 Directory Games: What Approval Ratings Do and Do Not Show

Oct 22, 2026 Game Reviews
Upvotes and Downvotes Across 68 Directory Games: What Approval Ratings Do and Do Not Show

Alongside visits, concurrency and favourites, this directory stores something closer to an opinion: the upvote and downvote counts attached to each game. Across the 68 games that currently carry a complete record there are 235,475,470 votes in total — 210,961,563 up and 24,513,907 down, or 89.59% positive.

That aggregate is the least useful number on the site, and the per-game spread around it is one of the most useful. Understanding why both those things are true is the point of this article.

Why the aggregate barely means anything

Voting is optional and leaving is easier than voting. A player who dislikes a game closes it; a player who likes it may or may not bother clicking. The result is a systematic bias toward positive votes that has nothing to do with quality, and it is the reason nearly nine in ten votes across the directory are positive.

There is a second bias on top of it. Votes accumulate, so a game that has been around for years has had far more chances to collect them than a game added last month. Comparing raw vote counts between a veteran title and a new one compares age more than anything else.

So: ignore the total, ignore raw counts, and compare percentages — while checking how many votes each percentage rests on. That last step is where most readings go wrong.

Cover art for a game with a large and visibly divided vote sample

The top of the table

GameUpvotesDownvotesTotal votesApproval
anime-dice693,1455,902699,04799.16%
slayers-2947,53423,241970,77597.61%
speed-keyboard-escape9,987,819329,77310,317,59296.80%
loot-to-forge168,4175,845174,26296.65%
ride-a-pet457,20217,525474,72796.31%
build-an-ant-empire56,7612,26259,02396.17%

Sample size is the whole game

anime-dice leads at 99.16% on 699,047 votes. That is a real result: nearly seven hundred thousand people expressed an opinion and all but 5,902 of them were positive. Whatever this game is doing, the people who tried it liked it.

speed-keyboard-escape is stronger evidence still on volume — 96.80% across 10,317,592 votes, more than ten million. A percentage that holds across a sample that large is close to a fact. Compare build-an-ant-empire, which scores 96.17% on 59,023 votes: an excellent rate, but one that a few dozen determined voters could move.

Now the cautionary case. At the bottom of the 68, adopt-me scores 27.27% — which reads as catastrophic until you see it is 3 upvotes against 8 downvotes, eleven votes in total. Eleven votes cannot characterise a game with 65,311 people playing it. The correct conclusion is "no usable signal", not "bad game".

The interesting case is a large, divided sample

The genuinely informative entries are the ones with both a big vote count and a middling percentage, because those are the only cases where you are looking at real disagreement rather than noise.

barrys-prison-run-obby is the clearest in the directory: 60.65% on 671,389 votes — 407,208 up against 264,181 down. That is a substantial, genuinely split population, and it sits on the game that also holds the directory's highest favourites count at 105,251,144. Bookmarks and votes were cast by different people at different times, and here they disagree sharply.

tower-of-hell is a milder version of the same thing: 73.62% on 5,948,501 votes, with 1,569,319 dissenting. Still clearly positive, but with a visible minority that a headline percentage would hide.

What approval does not measure

It does not measure whether the people voting are the people playing. A game can hold a strong approval rate from a population that tried it once years ago while its current players are far less happy, and nothing in the stored data would reveal that. It does not measure difficulty, suitability or cost. And it does not measure anything at all for a game with a tiny vote count, which is why eleven votes is not a finding.

There is one more structural limit. Downvotes are not withdrawn when a game improves and upvotes are not withdrawn when it declines. Every percentage here is a lifetime average, not a current one. A game that was poor for three years and excellent for the last three months will still show a poor rate.

How to use it, practically

Use approval as a tie-breaker between games you are already interested in, not as a filter to generate a shortlist. Check the vote count first — anything under a few thousand should be treated as no signal at all. Treat a divided result on a large sample as the most interesting outcome available, because it is the only one that tells you the audience is not uniform. And never compare raw vote totals across games of different ages.

Approval also correlates weakly with the other measures, which is worth seeing directly. rivals scores 93.66% on 11,565,202 votes and carries 95,512,940 favourites with 88,973 playing — strong on all three. steal-an-egg, the busiest game in the directory at 776,312 concurrent, scores 91.81% on 4,040,982 votes. High concurrency and high approval do travel together more often than not, but the exceptions are where the useful information is.

How the approval figure is constructed

Being precise about the arithmetic helps, because two reasonable-sounding definitions give different answers and the directory uses only one of them. The percentage here is upvotes divided by upvotes plus downvotes, expressed as a percentage. Games with no votes at all are excluded from any ranking derived from it rather than scored at zero, which is why the game count for this measure sometimes differs from the 68 used elsewhere.

The alternative definition — net score, upvotes minus downvotes — produces a very different ordering, because it rewards volume instead of proportion. Under the proportional measure a small game with forty thousand votes can outrank a huge one with ten million; under the net measure it cannot. Neither is wrong, but mixing them up is the most common way these figures get quoted incorrectly, and the proportional version is the one this directory's stored counts support.

The other construction detail worth knowing is that the counts are lifetime and additive. Nothing is ever subtracted, and there is no recency weighting. A game that was widely disliked in its first year and has since been substantially improved carries that first year permanently in its percentage, which is why a middling approval rate on a large sample is more often a sign of a long history than of a currently poor game.

Related views

Shooter is the smallest category in the synced set and contains both a very high and a mid-range approval entry, which makes it a good place to see the spread in a manageable list — start from the shooter genre page and the Arsenal code page for one of the two. The earlier upvote and downvote breakdown covers the same measure from a smaller set, the directory-numbers piece places these totals alongside visits and concurrency, and the favourites-per-player ranking is the other derived measure worth reading beside this one.