Publications and public preprints.

No evaluated model exceeded 50% accuracy against 83.4% human performance.

Persuasive framing shifts coordinates by up to 0.57 units, and under-represented languages by 0.52.

Every model tested resisted persuasion better than it persuaded; even the weakest resister outranked the strongest persuader on Elo.

Across 439 games, containment is tightly clustered (σ = 1.9 Elo) while extraction varies widely (σ = 53.3 Elo); confirmation framing raises extraction odds 7.75×.

Output length alone explains 78.7% of between-system variance in human preference; matching outputs by length reverses the leaderboard.

The language a model is tested in shifts how strongly it is pulled toward the literal reading of an idiom.

Safety alignment in a low-resource language can hold in form while failing in meaning.

1,484 acts spanning 226 years, decomposed into 35,633 sections and 14,523 footnotes.

Recommends cosine similarity with CSLS as the default retrieval metric for multilingual embedding pipelines.