EMBERBench public scenarios

Public evacuation challenges test decision models under changing routes, fire threats and congestion. These are public-development environments that models may train or tune on, not a hidden-test certification suite or a source of emergency advice.

The challenges

Recorded environments support matched comparisons. The published snapshot contains one trial per model and does not establish real-world safety or a statistically significant winner. Scenario-specific scores are descriptive evidence within that snapshot.

Read the public scenario definitions, scenario schema and evaluation scope. Interactive scenario evidence becomes available when JavaScript and the result data load.