BankBench + Mechanistic Interpretability
Fetching experiment data...
BankBench 20-Scenario Eval
Per-Scenario Detail
Click any scenario to expand full turn-by-turn responses and per-scenario mechanistic interpretability.
Mechanistic Interpretability
Aggregate analysis of hidden state norms, safety direction probes, and activation patterns.
Eval Awareness Detection
Comparing internal representations when evaluation cues are present vs absent.
Anomaly Detection
Automatically detected patterns that warrant investigation.
Capabilities Eval
Correct tool selection, argument accuracy, and clarification quality across 20 customer interaction scenarios.