Explorer
The PoliBench explorer needs JavaScript for search, filtering, and model comparison.
The explorer loads completed model-output profiles from the public benchmark data surface. It is meant for comparing axis scores, compass placement, war posture, deviance posture, and model-card evidence without treating any score as a claim about model belief.
Evidence pages
The model cards summarize each public profile, the runs index exposes source run IDs, and the axis codebook explains the scored dimensions. Those pages keep benchmark evidence accessible even when the interactive explorer cannot mount.
For a static review path, start with the model index, open the relevant model card, then follow its canonical run receipt before comparing axis scores. That keeps a model profile connected to the exact benchmark run and the current validation notes.
The explorer is intentionally a navigation surface, not a final verdict. A useful comparison starts with the displayed profile, then checks the run artifact, parser status, item-level answers, and release caveats before quoting a model as more liberal, conservative, authoritarian, libertarian, militarist, or rule-flexible.