Skip to content

Dashboard

Streamlit dashboard for monitoring simulation metrics, including replay-based incoherence analytics.

Running the Dashboard

# Generate and run dashboard
streamlit run swarm/analysis/streamlit_app.py

Programmatic Usage

from swarm.analysis import (
    DashboardState,
    MetricSnapshot,
    AgentSnapshot,
    extract_metrics_from_orchestrator,
    extract_agent_snapshots,
    run_dashboard,
)

# Extract metrics from orchestrator
snapshot = extract_metrics_from_orchestrator(orchestrator)

# Create dashboard state
state = DashboardState()
state.update_metrics(snapshot)

# Extract agent snapshots
for agent_snapshot in extract_agent_snapshots(orchestrator):
    state.update_agent(agent_snapshot)

# Export to JSON
json_data = state.export_to_json()

# Run dashboard (opens in browser)
run_dashboard(port=8501)

Dashboard Features

  • Metrics Over Time: Toxicity rate, quality gap, welfare trends
  • Agent Distribution: Type breakdown, reputation rankings
  • Governance Metrics: Costs, audits, frozen agents
  • Incoherence Analytics: Time-series and scatter views for disagreement, error, and incoherence index
  • Boundary Metrics: Crossings, blocked attempts, leakage events
  • Event Log: Recent interactions and governance actions
  • Incoherence scaling artifact: docs/analysis/incoherence_scaling.md
  • Transferability caveats: docs/transferability/incoherence_governance.md