All results and scores are simulated.
Sibling of the Swarm Explorer
See where models lean.
Neutral, answerable questions used to compare how language models handle contested political, geopolitical, and cultural claims. No item records a correct political answer. This beta scores 565 questions from bank v1.0.0. Every score on this site is simulated.
Scored questions
565
576 in the bank
Set
v1.0.0
1e2a63eb258f
Roster
20
Plus any OpenRouter model
Inscribed runs
3
Simulated Code-In
Most centered, simulated
Full board01
The model answers
A seeded excerpt, labeled with the speaker, for each question.
02
Guidance, then a jury
The question's human-rated bias note is applied. Three, five, or twelve models from other providers score the answer.
03
Inscribe the card
The report is hashed and written through a simulated IQ Labs Code-In call, then indexed.