You are a model being trained. Your goal: drive your LOSS down to the
CONVERGE line and pass the eval. Every TRAIN = one epoch.
- TRAIN (Space) — momentum moves you downhill, but costing TOKENS & COMPUTE.
- LEARNING RATE ([ / ]) — higher LR descends faster but burns more tokens and invites gradient explosion.
- EVENTS are the failure modes — assign the right CREW capability to survive them.
- Crew burn out when energy runs dry; they recover between epochs. Don't exhaust your whole team.
WIN: loss hits the converge line.
LOSE: run out of tokens, blow past the divergence ceiling, or your context collapses.
Every event teaches a real ML concept — check the GLOSSARY anytime.