Geopolitical Passports: LLM Ideological Divergence Across 10 Institutional Documents

Replication data for a structural discourse audit of three 20B–30B parameter class large language models. The dataset includes a 10-document corpus, execution scripts, and a CSV file of comparative ideological analysis across models. Hakan Mehmetcik published the dataset on figshare in May 2026 under a CC-BY-4.0 license.

Use Cases

• Replicate a structural discourse audit based on the provided scripts and corpus.
• Compare threat framing and epistemology across different LLM alignments mentioned in the description.
• Analyze model responses to contested texts on democracy and sovereignty referenced in the study.

Strengths

• Includes deterministic inference setup with temperature 0.0 and num_ctx 4096 for reproducibility.
• Contains the full computational artifact set: 10 pre-processed documents, scripts, prompts, and results.
• Compares three distinct model alignments: Western Commercial, Western Uncensored, and Global South.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is small at 158.5 KB, indicating limited scope.

Provenance

Source

figshare

Collection Method

Generated via a controlled experiment using strict prompt engineering and JSON-enforced extraction on three open-weight LLMs.

Time Range

null

Freshness

Last updated 2026-05-07 17:42:49; freshness should be verified.

Geography

null