RASPBERRY PI / VERIFIED EXPERIMENT

Two policies. One battlefield.

Two separate learned-policy checkpoints selected battle moves in independent PokéRogue clients. Automated lobby and dialogue handling assisted the run.

TV PAUSED Checking live game. No active match; stage summary shown.

Live co-op stage progress

—current stage

—best seen in this browser

—accepted model moves

—reported desyncs

Checking game activity…Two Raspberry Pi clients
Automatic paired frames
Checking verified gameplay… Actual moves and teacher feedback will appear here, with the full event timeline below.
SAM — Pokémon visual actor
Loading verified gameplay…

Live game frame or paused TV status · model-selected moves.

JevBot — transferred self-play actor
Loading verified gameplay…

Live game frame or paused TV status · synchronized co-op partner.

THE ACTUAL GAME DECISION LOG

Actions & teacher feedback

Verified selections, non-prescriptive feedback, and stage outcomes. Agent speech is shown only when independently recorded.

Loading verified events

Loading game decision evidence…

The teacher's comments are recorded; verified conversational replies from the agents are not yet available.

Teacher pool: feedback, not commands

After-action evaluation encourages valid moves or item choices when the observed evidence supports them. It never selects the agents' actions. Successful waves do not prove an optimal strategy or policy improvement.

Research with sources

SAM's native search retrieved an official PokéRogue Wiki result. The site's tutorial reading endpoint returned an access error, so comprehension and use of that knowledge remain unverified.

Official guide

Verification scope

The live TV shows matched pairs only while the game is active. While paused, it displays stage progress without streaming a previous battle. An unfinished live session is not independently verified.

Proof of the featured verified recording