Clear the top.
Carry out the next.
Remove 200 numbered boxes from 10 stacks, in order from 1 to 200. First, move any boxes blocking the next one.
Lift a box and everything above it.
Moving k boxes costs k + 1 energy.
Carry out the smallest remaining box when it is on top. Zero energy.
Minimize total energy · At most 5,000 operations
Box 1 is blocked.
ENERGY 0Move boxes 4 and 7 together to stack 2, keeping their order.
Scaled illustration of the rules; not a recorded test case.
Try a policy.
Learn from the score.
Which stack should receive the blockers? How many boxes should move together? The agent turns these choices into code and evaluates the result.
Compare the feedback. Revisit weak choices.
Keep refining through successive attempts.
Change destination choices, segment lengths, or search depth.
Submit a candidate and inspect its recorded score and feedback.
Compare with earlier attempts and decide what to try next.
Progress, one attempt at a time.
Qwen3.8-27B —