Skip to content

Streamlit Dashboard Guide

This page covers running, customizing, and deploying the NHRA Strategic Scenario Dashboard.

Quick Start

Running Locally

# Activate environment
poetry shell

# Run the dashboard
streamlit run scripts/dashboard.py

The dashboard will open at http://localhost:8501.

Dashboard Features

Tab Overview

Tab Purpose
📈 Scenario Analysis Trajectory plots, executive summary, intervention ranking
🕸️ Strategic Map D3 interactive network of game relationships
🌲 Game Tree Explorer Extensive form decision trees for each subgame
🏥 LHN Variance Intra-state local hospital network distribution
🧬 Data Lineage Parameter evidence traceability
⚖️ Validation Scorecard Backtest performance metrics and ghost overlays
🔬 Technical Analytics Stability regions and global sensitivity analysis
🛡️ Evidence Manager Evidence registry and conflict resolution
🔍 Forensic Audit Raw state inspection and parity checking
Category Parameters
Funding & Valuation Nominal Cth Share, NEP Growth
Operational Capacity Bed Capacity Index, Discharge Delay
Policy & Behaviour Political Salience, Audit Pressure
Clinical & Workforce Rurality Weight, Cost-Shifting Intensity
Patient Choice GP Out-of-Pocket, Time Value

Expert Mode

Toggle "🧠 Expert Strategic Mode" to manually override game equilibria:

  • Definition: R (Realism) / E (Strict)
  • Bargaining: A (Agree) / D (Defer)
  • Cost-Shifting: I (Invest) / S (Shift)
  • Discharge: C (Coordinate) / F (Fragment)
  • Governance: I (Integrate) / S (Separate)
  • Compliance: T (Tight) / L (Light)

Deploying to Streamlit Cloud

Prerequisites

  1. GitHub Repository: The app must be in a public GitHub repo (or private with Streamlit Cloud Pro)
  2. Streamlit Account: Sign up at share.streamlit.io

Step 1: Create requirements.txt

Streamlit Cloud uses pip, not Poetry. Create a requirements file:

poetry export -f requirements.txt --without-hashes > requirements.txt

Step 2: Create .streamlit/config.toml

[theme]
primaryColor = "#008080"
backgroundColor = "#FFFFFF"
secondaryBackgroundColor = "#F0F2F6"
textColor = "#262730"
font = "sans serif"

[server]
headless = true
port = 8501

[browser]
gatherUsageStats = false

Step 3: Create streamlit_app.py

Streamlit Cloud looks for streamlit_app.py by default. Create a symlink or entrypoint:

# streamlit_app.py
import subprocess
import sys

# Ensure src is in path
sys.path.insert(0, "src")

# Import and run the main dashboard
from scripts.dashboard import main

if __name__ == "__main__":
    main()

Step 4: Deploy

  1. Go to share.streamlit.io
  2. Click "New app"
  3. Select your repository: edithatogo/nhra_game
  4. Set branch: main
  5. Set main file path: streamlit_app.py
  6. Click "Deploy"

Step 5: Configure Secrets (Optional)

For API keys or sensitive config:

  1. Go to app settings → Secrets
  2. Add secrets in TOML format:
[api_keys]
logfire = "your-api-key"

Access in code:

import streamlit as st
api_key = st.secrets["api_keys"]["logfire"]

Performance Optimization

JAX Caching

The dashboard uses JAX with persistent disk caching:

# Automatic XLA compilation cache
# Located at ~/.cache/jax_cache

Streamlit Caching

Key functions use @st.cache_data for responsiveness:

@st.cache_data
def cached_run_model(p: Params, years: list[int], n_mc: int = 50):
    return run_hybrid(years, p, seed=42, n_mc=n_mc)

Monte Carlo Samples

  • Lite Mode (default): 50 samples — fast, lower confidence
  • SOTA Mode: 1000 samples — slower, high confidence

Toggle via sidebar "🚀 Boost to SOTA Accuracy" button.

Customization

Theming

Modify apply_custom_theme() in scripts/dashboard.py:

def apply_custom_theme():
    st.markdown("""
        <style>
        :root {
            --primary-color: #008080;  /* Teal */
        }
        .stMetric {
            border-left: 5px solid #008080;
        }
        </style>
    """, unsafe_allow_html=True)

Adding New Tabs

  1. Add to the tabs list:
tab1, tab2, ..., new_tab = st.tabs([..., "🆕 New Tab"])
  1. Implement tab content:
with new_tab:
    st.markdown("### New Feature")
    # Your content here

Adding New Interventions

Update the rank_interventions function's intervention list:

interventions = [
    "Pooled Funding",
    "UCC Integration",
    "New Intervention",  # Add here
    ...
]

And update apply_intervention() in src/nhra_gt/engine.py to handle the new option.

Troubleshooting

Common Issues

Issue Solution
Import errors Add PYTHONPATH=src before running
D3 network missing Run python scripts/interactive/make_d3_network_v26.py
Validation data missing Run python scripts/validation/recursive_backtest.py
GSA data missing Run python scripts/run_gsa.py
Slow startup First run compiles JAX kernels; subsequent runs are cached

Debug Mode

NHRA_DEBUG=true streamlit run scripts/dashboard.py

See Also