[Anonymized] Code and data for "Aligning With Whom? The Cultural Limits of Constitutional AI"

Anonymized code, prompts, raw model responses, and analysis scripts supporting the manuscript "Aligning With Whom? The Cultural Limits of Constitutional AI," currently under double-blind peer review.

**Version 2**: Adds `scripts/additional_analyses.py` (bootstrap CIs on Inglehart-Welzel dimensions and item-level binomial sensitivity test for directional steerability) and minor figure-resolution updates (300 DPI for fig1_cultural_map.png and steerability_heatmap.png). All other analyses, raw data, and processed outputs are unchanged from version 1.

Contents:
- src/: Python source for item selection, prompt generation, API calls, response parsing, statistical analyses, and figure generation
- data/: World Values Survey items selected for the study (from the GlobalOpinionQA dataset) and country-level response means for 90 nations
- prompts/: All 770 generated prompts (55 Format A direct survey + 715 Format B advice-seeking with cultural context)
- results/: Raw API responses, parsed Format A data, LLM-coded Format B data, generated figures, statistical tables, and human inter-rater validation sample
- scripts/: Validation spreadsheet generator (create_validation_spreadsheet.py) and additional reviewer-requested analyses (additional_analyses.py: bootstrap IW dimensions + item-level binomial)

This deposit is anonymized for double-blind peer review. Author identity, institutional affiliation, and the public source code repository will be linked upon publication acceptance.

Models tested: claude-sonnet-4-20250514 (primary), claude-haiku-4-5-20251001 (cross-model replication).