Run your first published task

One local Moving AI task, a checked result and an export.

This walkthrough creates one new offline route for bundled movingai:0. It uses the platform service, not the historical dashboard. No model key, GPU, device, HDF5 download or paid API is needed.

1. Use an installed package or an authorized checkout

Use Python 3.11+ in your existing isolated environment. For a supplied wheel, compare its SHA-256 with the operator's build receipt before installing the exact file:

python -m pip install /path/to/worldfuzz-0.4.0rc6-py3-none-any.whl

This is a local file example, not a command to install an advertised PyPI release. From a checkout, python -m pip install -e . uses the same platform. Optional Can/LeRobot runtimes are unnecessary for this task.

2. Start the common platform service

Work in a new directory where workspace/ does not exist. Keep this terminal open:

python -m worldfuzz platform --root workspace/results --store workspace/catalog.sqlite3 serve

Open the local workspace. Documentation is at this quickstart. The listener is loopback-only. If port 8766 is occupied, choose a free serve --port and use that URL. Stop only the server you started, using Ctrl+C when finished.

3. Execute one task

In a second terminal, use the same environment and directory. These explicit root/store paths address the same private workspace:

python -m worldfuzz platform --root workspace/results --store workspace/catalog.sqlite3 run navigation --resource movingai:0 --case-index 0 --key quickstart-nav-0 --max-seconds 30 --cpu-seconds 20 --memory-mib 256 --artifact-mib 1 --wait-seconds 30 --export workspace/export-nav-0

The typed CLI calls run_goal, which compiles, prepares, executes and reads through the shared dispatcher. The adapter operation is solve_navigation under movingai.octile.v1. It checks the actual zero-based case index, map cells, adjacency, endpoints and the published optimal length.

For the browser alternative, select a published navigation resource, choose Use this task, review the limits, press Check compatibility, rights and limits, then Run this plan. Choose one approach for this walkthrough; each new submission is new work.

4. Read the result

A terminal response includes a run ID and a short report. On success, workspace/export-nav-0/ contains summary.json, junit.xml, result.json and the versioned JSON artifact view in exact-json/. Check execution status, task outcome and verification level separately. No-violation-found applies to the checked grid task, not robot safety.

If waiting expires, keep the returned run ID and repeat the same goal and key to read/resume the same logical job. A waiting response is not task failure and does not prove export completion. The CLI will not overwrite an existing export directory; when inspecting a completed repeat, omit --export or select a new output directory.

This direct local CLI is for the trusted filesystem owner. For organization workspaces, connect with an existing scoped token using the authenticated placement instructions; do not open their catalog as an unauthenticated local owner.

Next: compare results, export original artifacts or datasets, or resolve a refusal.