---
title: Steered-output explorer
emoji: 🧭
colorFrom: yellow
colorTo: indigo
sdk: streamlit
sdk_version: "1.39.0"
app_file: analysis/dashboard.py
pinned: false
short_description: Browse persona-vector steered outputs vs baseline
---

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.