Where Minecraft mod downloads fail, country by country
A mod download either arrives or it doesn't, and until recently we could only count the failures — not the attempts. That sounds like a small gap. It isn't: without knowing how many people opened a mod in the first place, a country with more users always looks like a country with more problems. We fixed the denominator this month. The picture that came out is lopsided in a way we did not expect.
The numbers
These are failures per 1,000 mod opens, measured on 2 August 2026, for every country with at least 5,000 opens that day. A "failure" here means the app tried to fetch the file, could not, and fell back to asking our server for help.
| Country | Mod opens | Failures | Per 1,000 |
|---|---|---|---|
| Russia | 35,020 | 7,825 | 223.4 |
| United States | 32,370 | 1,292 | 39.9 |
| Algeria | 6,303 | 228 | 36.2 |
| Ukraine | 7,037 | 203 | 28.8 |
| Thailand | 6,720 | 192 | 28.6 |
| Germany | 7,266 | 203 | 27.9 |
| Philippines | 27,643 | 697 | 25.2 |
| Malaysia | 8,084 | 180 | 22.3 |
| Chile | 7,362 | 154 | 20.9 |
| Colombia | 7,640 | 124 | 16.2 |
Russia is fourteen times worse than Colombia. Not fourteen percent — fourteen times. And it is not a small country skewing a small sample: Russia and the United States had almost identical traffic that day, 35,020 opens against 32,370, and one of them failed six times as often.
It is not the mods. It is the road.
The obvious explanation would be that Russian players download different, more broken mods. They don't. The failures are spread across 482 different mods, and the single worst one accounts for 9% of them. That is the signature of something wrong with the delivery, not with the files.
So we split the failures by which host serves the file. Most mods are fetched straight from the original site; some we mirror ourselves so they stay reachable. Over four days:
| Date | Russia, our mirror | Russia, original CDN | US, our mirror | US, original CDN |
|---|---|---|---|---|
| 30 July | 3,343 | 325 | 397 | 629 |
| 31 July | 4,831 | 686 | 578 | 808 |
| 1 August | 5,668 | 877 | 304 | 856 |
| 2 August | 7,069 | 751 | 437 | 844 |
Failures on our own mirror in Russia more than doubled in four days. The same Russian users' downloads from the original CDN stayed flat. In the United States neither column moved. One host, in one country, getting steadily worse while everything else holds still — that is a domain being filtered somewhere on the route, not a file server having a bad week.
Knowing that changes what we can do about it. A degrading server would need more capacity. A filtered domain needs a different domain, which is a thing we can test in a day and undo just as fast.
Three ways this number lies, and how we caught them
We publish these figures with the traps attached, because we walked into all three.
1. A count without a denominator is a popularity chart
Our own daily report used to rank countries by raw failure count. Russia sat at the top, which was correct, and the United States sat second, which was not — the US was second because the US is big. Once divided by opens, the US drops to a middle-of-the-table 39.9 and countries nobody was watching move up.
2. Small countries produce fake emergencies daily
On the morning we started checking, the alarm flagged the Dominican Republic at 3.3× its usual level and Bolivia at 5.6×. Both looked like something had broken nationwide. Neither had. The Dominican spike was 477 failures spread over ten mods, with 288 of them — 60% — on a single one. Bolivia's was 55 failures across three mods, 95% on one. That is one broken file, or one determined person, not a country.
3. One person can look like a crowd
There is no "give up" state in the download button. If a file will not come, people tap again. We measured one player making 17 attempts on the same mod in 26 minutes. In a country with a few thousand daily opens, one such session is a visible spike. Any per-country number built from raw attempts is an upper bound, and we say so rather than quietly presenting attempts as people.
What we do with this
Three things follow directly. Countries where the rate is high but the failures sit on one or two mods get the mods fixed, not the infrastructure. Countries where the rate is high and spread wide get the route investigated — which is how the Russian mirror problem surfaced at all. And countries that look bad only because they are big get left alone, which saves the time that used to go into them.
None of this is visible from inside a single download. It only appears when you count the attempts against the opens, keep the two windows aligned, and refuse to believe a ratio built on eleven events. We got that wrong for a while. The numbers above are what it looks like after.