Psychscanner Features Demonstrated
- 1: A custom multi-module LangGraph agent (MAP) vs. two comparison agents
- Task: Prospect Theory risky-choice gambles (
examples/tasks/prospect_theory_demo.tcard.psyscan,
generated per musslick/experiment_factor_discovery’s benchmark spec — see
examples/tasks/generators/generate_prospect_theory.py), run through 3
agent architectures from src/psychscanner/agents/:
make_map_agent (new — Webb, Mondal & Momennejad 2025’s Modular Agentic
Planner, https://doi.org/10.1038/s41467-025-63804-5), a conversation-memory
baseline (memory="Convo"), and make_lats_agent (Monte Carlo tree
search, pre-existing). Real Groq (llama-3.1-8b-instant) calls throughout.
See simulation/, data/, analysis/, reports/report.md.