VityaVitalich/steered-outputs-dashboard

Private dashboard for exploring steered-output CSVs from the project-2-evil-llama codebase. Walks data/model_responses/eval/ and surfaces every (eval_model, eval_revision, persona-vector trait, extract checkpoint, layer, coef) combination that has both a usable steered CSV and a matched baseline, so the sidebar only lists combinations that actually render.