Efficient frontier fit, hidden

The prompt

Build the same frontier-fitting game as a visible scatter of portfolios, but hide the scatter. I can see the individual assets and the range of risk I am supposed to cover. I cannot see the cloud of mixed portfolios, and I cannot see the upper edge I am trying to match. I still submit a smooth curve, a few tries, and after each try you tell me how far off I was, without drawing the answer.

A close curve scores well. A wild one scores poorly. When the round ends, still do not reveal the cloud. The point is to build the curve from the assets and the error messages, not from tracing a picture.

Several markets. The same market should behave the same way every time.

# AI : shared Action contract — the page posts {"move": "..."}; keep this shape.
from pydantic import BaseModel


class Action(BaseModel):
    move: str