A working learning brain shows three signals on this page over time:
1. Loss curve trending down. Each new training example moves loss toward zero. If loss is flat or rising, the brain is stuck.
2. Test accuracy > baseline. If accuracy beats the 'always guess up' baseline, real signal is being learned. Check via Historical Trainer.
3. Feature weights diverging from zero. A trained brain has clear opinions — some features have weight > 0.5, others < -0.5. A brain with all weights near zero hasn't learned anything yet.