Adaptive Self‑Recursive Rock‑Paper‑Scissors

Client-side production-ready single-file game. The agent learns online from your moves and adapts its own learning & exploration parameters.

Play a move to start — the agent will learn online.

Last Round:

—

History (recent 50)

Rounds: 0

Agent wins: 0

User wins: 0

Ties: 0

Epsilon (explore): 0.20

Recency decay: 0.90

Forget noise: 0.01

User move probs (agent belief):

Rock

Paper

Scissors

Recency decay (0.6 - 0.99)

Base epsilon (explore) (0.0 - 0.8)

Epsilon min

Meta window (rounds)

Adapt step

Forget noise

Tip: lower decay -> agent emphasizes recent user moves more. Increasing epsilon makes agent explore more (degrade).

Single-file client-side app. No server required. To run locally: save as adaptive_rps.html and open in a modern browser. Use Export to save logs.