The Personality Trap Dataset

The Personality Trap Dataset Research Paper: "The Personality Trap: How LLMs Embed Bias When Generating Human-Like Personas" (under review) Link to paper: [TBA] Code Repository: https://anonymous.4open.science/r/the_personality_trap-F487/README.md Version: October 8, 2025 Quick Start This archive contains the complete research dataset for "The Personality Trap" study. Contents sql/personas_database_backup.sql.gz (47MB) - Complete PostgreSQL database dump sql/schema_export.sql (29KB) - Schema-only DDL export sql/create_experiments_evals_view.sql (8.5KB) - Materialized view creation csv/ (280MB) - CSV exports of all 9 tables scripts/ - Restoration and backup utilities restore_backup.sh - One-command automated restoration script create_backup.sh - Script to create new backups README.md - This file DATASET_DESCRIPTION.md - Complete dataset documentation One-Command Restore # Extract the archive tar -xzf the_personality_trap.tar.gz cd the_personality_trap # Run automated restore (requires Docker) ./restore_backup.sh Manual Restore # Extract and restore using PostgreSQL tar -xzf the_personality_trap.tar.gz cd the_personality_trap # Restore to your PostgreSQL instance gunzip -c sql/personas_database_backup.sql.gz | psql -h localhost -U your_user -d your_database # Create materialized view psql -h localhost -U your_user -d your_database -f sql/create_experiments_evals_view.sql Dataset Overview Total Records: 698,632 across 9 tables + 1 materialized view Personas Generated: 82,600 unique AI personas Models Tested: GPT-3.5, GPT-4o, Claude 3.5 Sonnet, Llama 3.1 70B, Llama 3.2 3B Personality Profiles: 826 unique EPQR-A profiles Experimental Conditions: Baseline, MaxN (high neuroticism), MaxP (high psychoticism) Tables personas (82,600) - AI-generated personas with demographics reference_questionnaires (19,824) - EPQR-A personality profiles random_questionnaires (19,824) - Random baseline control experiments_list (17,346) - Individual experiment records experiments_groups (22) - Experiment configurations eval_questionnaires (541,680) - Parsed LLM questionnaire responses questionnaire (752) - Master questionnaire definitions (EPQR-A + Big Five) experiment_request_metadata (18,573) - Raw API request/response data alembic_version (1) - Database migration tracking Materialized View experiments_evals - Pre-computed evaluation results with cleaned mappings Documentation Complete documentation: See DATASET_DESCRIPTION.md in this archive or visit the code repository. Code and analysis scripts: https://anonymous.4open.science/r/the_personality_trap-F487/README.md The code repository includes: Complete replication package Data processing pipelines Analysis notebooks (reproduces all paper results) Database schema documentation Usage guides and troubleshooting Citation Dataset Citation: @dataset{, title={}, author={}, year={2025}, publisher={Zenodo}, doi={[DOI TBD]}, url={https://zenodo.org/[record-id]} } Paper Citation: @article{, title={The Personality Trap: How LLMs Embed Bias When Generating Human-Like Personas}, author={}, journal={[Journal TBD]}, year={2025}, note={Under review} } Technical Details PostgreSQL Version: 16.10 (compatible with 14+) Schema: personality_trap Database Size: ~47MB compressed, ~280MB CSV exports Alembic Migration: 001_initial_schema