Use task-specific hardware status and backflip warning

Prepare public jump/backflip release with public source links and fresh behavior evidence

Add verified full-checkpoint training continuation and provenance