---
title: Fusion Lab
emoji: 🧬
colorFrom: green
colorTo: gray
sdk: gradio
sdk_version: 6.17.3
python_version: "3.11"
app_file: app.py
pinned: false
---

# Fusion Lab

A model-credibility explorer for a turn-level late-fusion earnings-call sentiment
model. Gated text+audio+interaction fusion scores **0.741 macro-F1** on 5-fold
walk-forward CV across 1,038 calls, beating a FinBERT text-only baseline (0.699) by
+5.1 points (95% CI excludes zero).

Explore the 5-rung benchmark ladder, drill into any of 675 out-of-fold calls with
turn-level attention weights over the real transcript, and read the full methodology
including an honest negative result on the volatility-regime target.

All data is static — precomputed by `src/evaluation/export_dashboard_data.py` in the
parent repo, no live inference.