src.visualization builds and serves the static dashboard used to compare
baseline, treatment, and other AgentKernelArena experiment reports.
The module keeps three kinds of files separate:
- Python implementation and static frontend assets live in
src/visualization/. - Generated dashboard payloads live in
.visualization/dashboard/. - Optional manually collected report bundles live in
.visualization/reports/<report_name>/.
Normal run reports remain under:
workspace_<gpu>_<agent>/run_<timestamp>/reports/
Run these commands from the AgentKernelArena repository root.
Build a dashboard from normal workspace runs:
python3 -m src.visualization build --include-workspace-runsServe an already-built dashboard:
python3 -m src.visualization serve --host 127.0.0.1 --port 8080Build and serve in one command:
python3 -m src.visualization run \
--include-workspace-runs \
--host 127.0.0.1 \
--port 8080Equivalent Make targets are available from the repository root:
make visualization-build
make visualization-serve
make visualization-runWithout --include-workspace-runs, the builder scans only
.visualization/reports/<report_name>/. Each report directory must contain:
overall_summary.csvtask_type_breakdown.jsonoverall_report.txt
src/visualization/
├── __init__.py
├── __main__.py
├── build_data.py
├── paths.py
├── server.py
└── frontend/
├── index.html
└── dashboard/
├── app.js
└── styles.css
The server exposes only the dashboard assets plus .csv, .json, and .txt
report artifacts. Directory listings and path traversal are rejected.
For non-default layouts, set AKA_PROJECT_ROOT or
AKA_VISUALIZATION_RUNTIME_ROOT before invoking the module.