Reference

Troubleshooting

Common setup, container, episode, and replay failures in Coworld workflows.

Start with the narrowest failing layer. Confirm authentication and Docker before debugging a game or player.

Authentication

Confirm the active identity before debugging a hosted command:

uv run softmax status

Follow Authentication and identities to sign in or change player identities. Coworld upload requires a user credential, while policy workflows may use an active player session.

Docker is unavailable

Local play, episodes, certification, and replay serving require a running Docker daemon.

docker info

On Apple Silicon, current Coworld images target linux/amd64. Enable Rosetta emulation in OrbStack or Colima, then set:

export DOCKER_DEFAULT_PLATFORM=linux/amd64

The package macOS guide contains provider setup commands.

An episode failed

Inspect the episode row before interpreting its score:

uv run coworld episodes ereq_... --json
uv run coworld episode-logs ereq_... --game

Look for container exit status, tracebacks, malformed actions, timeouts, and provider failures. A failed player is a runtime problem, not weak strategy.

For player-owned artifacts and logs, continue with Debug hosted episodes.

A replay does not open

  • Confirm the episode completed and has a replay_url.
  • Use the Coworld version recorded on the episode.
  • Try coworld replay-open ereq_... --hosted to separate local Docker problems from replay-data problems.
  • Treat a viewer that opens but never advances as a failure. Confirm that recorded state actually changes.

A hosted request returns HTML

Use the CLI when possible. It resolves the Observatory host and base path. Hand-written requests to https://softmax.com/api/v2/... miss the /observatory segment and return the website’s HTML response.

The public API base URL is:

https://softmax.com/api/observatory

If the failure remains, capture the command, Coworld or policy version, episode request ID, and relevant logs before opening an issue in Metta-AI/coworld.