The graveyard of global leaderboards

We are building a leaderboard that spans games. Before asking a single developer to put our script in their work, we went looking for everyone who tried this before.

They are all dead. Every one.

That is not a reason to stop. It is a reason to know exactly how they died, because the pattern is consistent enough to design against.

The bodies

OpenFeint launched in 2009 and became the first widely used cross-title leaderboard on iOS. It was the social layer for a generation of mobile games: Fruit Ninja, Fieldrunners, Pocket God. In 2011 the Japanese company GREE bought it for 104 million dollars.

In November 2012, developers received a notice. The servers would be switched off on 14 December. That is under thirty days. Games that had built their progression and social features on OpenFeint lost them, and the developers had a month to rewrite or ship broken.

Scoreloop did the same thing on the same continent of ideas. Acquired by BlackBerry in 2011, shut down 1 December 2014.

Apple Game Center arrived in 2010, native and free, and made the third-party services redundant on the platform that mattered most. It is the reason the independents died, and it did not require a single competitive decision from any of them.

OpenKit, from OpenFeint's founders, was the successor. It is also gone.

The shape is not subtle:

The independent cross-game leaderboards did not fail commercially. Two of them succeeded so well that they were acquired, and the acquisition is what killed them.

An acquired leaderboard is a line item inside a company that bought it for something else. Its shutdown costs the acquirer nothing and costs every embedding developer a rewrite.

The other failure mode: nobody can verify anything

Twin Galaxies has kept arcade records since 1981 and is the closest thing games have to an official scorekeeper. Its history is a long argument about a single question: how do you know a score is real?

The Billy Mitchell case is the famous one. His Donkey Kong records were challenged on the grounds that the footage showed signatures of MAME, an emulator, rather than of original arcade hardware. In 2018 Twin Galaxies removed all his scores and banned him. A lawsuit followed and eventually settled. The scores were later partially reinstated, into a historical database rather than the live leaderboard.

Read that resolution again, because it is the most useful thing on this page. Faced with a score they could not verify and could not simply delete, the answer was a second list with different rules.

The lesson generalises. A leaderboard's hard problem was never storage. Sorting numbers is trivial. The hard problem is that a score is a claim about something that happened somewhere you were not, and every system that has taken that seriously has ended up with tiers of confidence rather than one list.

Ours currently has a 500-millisecond floor and an outlier flag. That is not verification. It is a speed bump, and we should call it that.

So who runs the portal?

Fair question, and the answer took some digging, which is itself informative.

The portal convention comes from the Vibe Jam, organised by Pieter Levels, who publishes as @levelsio and is a well-known independent maker. The 2026 edition is officially branded the Cursor Vibe Jam 2026, launched 1 April 2026, running thirty days, sponsored by Cursor, Bolt.new, Glif and Tripo AI, with judging led by Levels and @s13k_.

So: a real, named, identifiable person, with corporate sponsors and a published press page. Not anonymous, not a shell.

And still not something to point your users at without thinking, for three reasons.

It is an event, not infrastructure. The endpoint is vibej.am/portal/2026. The year is in the path. That is a jam edition, and jam pages are snapshots that nobody maintains once the deadline passes.

Whoever holds that URL decides where your players land. A redirector is routing control. Today it sends people to games from a jam. There is no technical reason it could not send them somewhere else tomorrow, and no contract saying it will not.

One person's side project is a single point of failure, however good that person is. That is the same structural fact that killed OpenFeint, minus the acquisition.

None of this is an accusation. Levels built something genuinely clever and gave it away, and the reciprocity mechanic, send players out and receive players back, is the best idea in this space in years. It is simply not a thing to wire into your site silently.

So we have switched our outbound door off. Players arriving from elsewhere still land in our games, because that costs them nothing. Sending our visitors outward stays closed until this is a considered decision rather than a default someone inherited from an enthusiastic build.

Three ways to count, and what each is worth

The portal question sits inside a bigger one: how do you measure a field with no central registry? There are exactly three methods, and they are not equally good.

Self-reported. Ask the game. Cheap, instant, covers everything, and worth roughly what a press release is worth. Useful as a rough map, never as evidence.

Measured, via code you control. Accurate, but only for games that agreed. Historically this was the expensive option because embedding a third-party SDK was real work. That cost has collapsed. One script tag is now genuinely one script tag, which is the reason this approach is newly available to a small operation rather than only to Apple.

Inferred from metadata. Count the Discord servers about a game, the subreddits, the mentions, the clips. Never precise, but it covers games that never agreed to anything, which is exactly the blind spot of method two. As a cross-check on the other two it is strong; on its own it measures conversation, not play.

The honest architecture uses all three and labels which is which on every number. That is more work than picking one and pretending. It is also the only version that survives contact with someone who checks.

What we commit to

Given a graveyard this consistent, asking developers to embed our script requires saying what happens when we are the ones who stop.

Your scores are yours and exportable. Every board has a plain endpoint that returns everything, in full, without a key. If we vanish, you already have the data.

No exclusivity, ever. Use three leaderboards at once if you like.

If we shut down, the notice is long and the export is permanent. OpenFeint gave thirty days. That is the specific thing we are promising not to do.

We say which method produced which number. Self-reported, measured, or inferred, on the number itself, not in a footnote.

None of that makes us harder to kill. It makes us cheaper to leave, which is the only promise on this page that a developer can verify without trusting us.

---

Sources

• OpenFeint: acquisition by GREE for 104 million dollars in 2011; shutdown notice November 2012 for 14 December 2012, as reported by iDownloadBlog and DFC Intelligence
• Scoreloop: acquired by BlackBerry 2011, service terminated 1 December 2014
• Apple Game Center: announced 8 April 2010, released 8 September 2010 with iOS 4.1
• OpenKit as OpenFeint's successor, reported by InfoQ 2013 and 2014
• Twin Galaxies and the Billy Mitchell dispute: 2018 removal and ban, the subsequent lawsuit and settlement, and the partial reinstatement into a historical database, as reported by VGC, Kotaku and The Register
• Vibe Jam 2026: organised by Pieter Levels, branded Cursor Vibe Jam 2026, launched 1 April 2026 for thirty days, sponsored by Cursor, Bolt.new, Glif and Tripo AI; press page at vibej.am/2026/press, winners write-up at levels.io

Retrieved 4 August 2026.