Skip to content

Local interface

From the repository root:

poetry install --extras ui
poetry run interop-trace ui

Open http://127.0.0.1:8080/ in your browser. The interface runs on your computer. For a different checkout or port, use --directory /path/to/checkout --port 8090.

Run and inspect

  1. The table lists experiments from root-level YAML files in experiments/, with their IDs, source files and component choices. Loading it downloads nothing.
  2. Check the experiments to include, across any number of files. Nothing is selected initially; the header checkbox selects all. Click Run to execute the selection in table order, using the shared cache and a new results/ui-…/ directory.
  3. Watch status, stages and elapsed time. Updates arrive about once per second.
  4. Select a saved run and expand Files and provenance. File group selects Run-level files (configuration and console log) or one experiment's files. Use the single preview/download control for outputs and logs; a selected log updates live. Within an experiment, benchmark reports are under evaluation/; within run-level files, configuration.json contains the run configuration snapshot. File names are relative to the selected group.

If an experiment fails, later experiments still run. The batch stays running until execution finishes, then shows failed if any experiment failed. Run-level summary.json lists completed IDs and failures; each experiment retains its stage, error and partial outputs.

The first experiment is a small, verified starting point. ProteoBench configurations also need its extra: poetry install --extras "ui proteobench".

Behavior and limits

One run can be active per UI server. Each run can contain multiple sequential experiments. Closing or refreshing the browser tab leaves execution running; keep the terminal server running until the experiment finishes. Stop the server with Ctrl+C after completion. Stopping it during execution may interrupt the experiment.

Completed and failed CLI runs also appear in history. After restarting the server, unfinished runs are shown as unknown because the interface cannot confirm their process state. Inspect their records and logs before starting more work.

The interface shows stages rather than estimated percentages or completion times. The shared preview shows the first 32 KiB of UTF-8 text, or the last 32 KiB for logs. The complete console log is saved beside the run directory as ui-….console.log. Native workflow logs remain inside the run. Binary files remain downloadable. JSON, YAML, XML and TOML previews use built-in syntax highlighting, line numbers and folding. Other formats, including CSV/TSV, are shown as plain text. Execution completion and scientific quality are separate: interpret each benchmark's scope and metrics.

Experiment definitions remain YAML files. Edit them using the configuration guide, then reload the page to refresh the list.