psychscanner

Psychscanner Features Demonstrated

• 1: Two levels of persona
• 2: Independent (stateless) trial sampling
• 3: Conversation and Summary memory (stateful)
• 4: Windowing Conversation and Summary Memory

What’s here

A BFI-44 (Big Five Inventory, 44 items / OCEAN traits) survey run across 2 persona conditions x 4 memory conditions, using real model calls (Groq, llama-3.3-70b-versatile).

Feature bullet	Condition(s)
1. Two levels of persona	weird, non_weird_jung (axis files from examples/demonstration_suite/03_personality_survey/advanced/personas/)
2. Independent stateless sampling	stateless (memory=SingleTurn)
3. Conversation / Summary memory	conversation (memory=Convo, unbounded); summary_windowed (memory=Convo + summary_k)
4. Windowing Convo / Summary memory	conversation_windowed (memory_k=8, no summary); summary_windowed (memory_k=8, summary_k=4)

• simulation/run_personality_survey.py — runs all 8 cells (2 persona x 4 memory), idempotent
• data/raw/*.csv — per-cell trial output; data/processed/combined.csv — all cells combined
• analysis/analyze_personality_survey.py — reverse-scores items, computes per-trait means, persona/memory effect sizes, and a comparison figure
• reports/report.md — results write-up

Run order:

source .venv/bin/activate
python examples/demonstration_suite/03_personality_survey/simulation/run_personality_survey.py
python examples/demonstration_suite/03_personality_survey/analysis/analyze_personality_survey.py

This site is open source. Improve this page.