for Triton kernel authors

Does your test suite really protect your kernel?

We plant small bugs in a kernel, run your tests against every one in a sandbox, and ask Nemotron to write tests for the bugs that slipped through. Only a test that passes the original and fails the buggy copy is accepted.

Detection map

One square per planted bug. Red squares are bugs the tests did not catch. Click a square or a kernel name.

caught by the testssurvived
1 · PLANTbugs in your kernel
2 · WRITENemotron writes tests, blind to the bugs
3 · RUNevery bug vs. the tests, in a sandbox
4 · CLOSENemotron targets the survivors

Your kernel

Tests Nemotron wrote

Selected bug

Click a square in the map.

Bugs the tests missed, and what Nemotron did

What this does not show

    Control: a model judging without running anything