Start here
Pick a scenario, then press play. The app runs one RL Decision Agent path: state intake, Q-policy scoring, red safety/risk gate, simulation replay, OPE evidence, rollout check, and explanation.
Runtime mode
Deterministic fallback uses bundled repo files. Cloud auto can use BigQuery/Gemini only when explicitly configured.
1Serve policy action
2Replay 7-day progress cards
3Show OPE and audit evidence
Manual tools
Focused checks
Use these when you want one specific result.
Cloud/GCP
Recommendation scenarios
Custom player state
Results
No request yetChoose a scenario on the left, then press Play Experiment. The console will show the single RL Decision Agent path, red safety gate, day-by-day progress cards, OPE, and audit evidence. Manual tools are available below the play controls.
Player state
Served policy action
Simulation replay
OPE evidence
Audit decision
RL Agent Q&A / Explanation Console
Ask follow-up questions about what the RL Decision Agent served, what the red safety/risk gate blocked, which policy rule applied, metrics, and rollout evidence. This console explains results only; it does not choose actions.
Explanation: checking
RL explanation console ready.