How we know it works
The evidence behind the claims.
Everywhere else on this site tells you what Cubelings does. This is where we show our working: the tests, the measurements and the mistakes that shaped how it plays. Written for anyone curious enough to check.
-
2026-08-21
A datapack that installs perfectly and does nothing
One selector had two bracket groups in it. The pack loaded, enabled, reported success, and then answered "Unknown function" for its own tick — silently, forever. Plus: why asking a model to rewrite a document quietly loses a third of it.
Read the post → -
2026-08-19
The six redstone wire rules nobody reasons their way to
Build order is load-bearing, a repeater faces its input not its output, and a closed loop of repeaters latches itself on. Six rules measured against a real server, not derived on paper.
Read the post → -
2026-08-12
Why a survival benchmark needs a fixture world
The first trials got worse every run, and it looked like the agent was failing more often. It was actually chopping down the forest it was being tested in — oak logs in range fell from 110 to 50 over eight runs.
Read the post → -
2026-08-05
Digging is not mining
Breaking a block drops an item on the ground, not into an inventory. Stand three blocks back while you dig and a technically perfect run scores zero.
Read the post →
Be in the first worlds.
We are opening a small number to begin with. Put your name down and we will tell you when yours is ready.