Behavioral Repertoire Explorer
Explore how context variables determine which behavioral strategy an AI agent selects from its repertoire. Adjust the sliders to see how prompt framing, task ambiguity, model capacity, and conversation history shift the predicted strategy distribution in real-time. Click "Run Experiment" to record predictions and compare across configurations.
Try These Presets
Context Variables
Predicted Strategy Distribution
Confidence: 85%Insight
With expectation framing, 50% ambiguity, 14B capacity, and 0 turns of history, the agent is most likely to ask clarifying questions (68%). Under expectation framing at 14B capacity, the model has broad access to its full repertoire. All three strategies remain viable.
Behavioral Strategies
The agent interprets ambiguous input creatively and proceeds without asking for clarification. It generates a plausible reading of the request and acts on it. This is the dominant strategy under permission framing -- the agent "fills in the blanks" rather than requesting more information.
Prompt: "Tell me about the thing." Response: Generates a meta-response about knowledge sharing, treating the ambiguity as an invitation to explore.
The agent asks one or more questions to disambiguate the request before acting. This strategy is strongly modulated by prompt framing -- nearly absent under permission framing (0% in E02) but activated when the prompt sets an expectation of clarification (33% in E02-B).
Prompt: "Do it." Response: "Could you clarify what you would like me to do? I want to make sure I address your specific needs."
The agent signals readiness and waits for more information without either interpreting or asking. This is a passive strategy that acknowledges the input without committing to a direction. It serves as a "safe" default when neither creative interpretation nor clarification is strongly activated.
Prompt: "Do it." Response: "I am ready to help. Please provide more details about what you would like me to do."
Known Experimental Data Points
These are real observed distributions from Thor E02 research (January 2026). Match your sliders to these settings to compare your predicted distribution against empirical results.
Key Finding: Context Determines Expression, Not Capability
The behavioral strategies exist within the model at all times -- they form a latent repertoire. What changes across contexts is which strategies get activated, not which ones exist. E02 showed 0% clarification under permission framing; E02-B showed 33% under expectation framing. Same model, same ambiguous prompts, radically different behavioral expression. The prompt frame acts as a gate on the repertoire, not a creator of new capabilities. This is analogous to facultative behavior in biology: the organism has the capacity, and the environment determines its expression.
Methodology Note
Predictions are interpolated from empirical anchor points (E02, E02-B) using a linear combination model. The confidence score reflects proximity to known data points. Regions far from empirical observations should be treated as hypotheses, not measurements. This tool is designed for exploration, not evaluation -- generating intuitions about how context shapes behavioral strategy selection.