Track moved examples and training log

deployment changes

Add interactive frontend: radar demo + live API + results dashboard

Revert prompt.py to on-call engineer framing (undo ATC narrative)

ATC framing: update observation_to_prompt narrative (no logic changes)

Fix threshold to 0.45, R3 neutral to 0.0, rebalance weights

Tune goodness threshold to 0.68

Fix R3 neutral score, raise goodness threshold to 0.72

Raise goodness threshold to 0.72 for better training signal

initial commit