You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository was archived by the owner on Apr 18, 2026. It is now read-only.
During construction scheduling T3 (dispatcher) eval runs, stream_subscribe("crew_updates") and stream_subscribe("weather_updates") throw RuntimeError: Invalid argument (name): No stream factory registered in the Monty sandbox.
Reproduction
Run the construction experiment with T3 dispatcher rooms:
dart run packages/soliplex_cli/bin/construction_experiment.dart http://localhost:8000 /tmp/eval 1
T3 rooms use needsStreams: true and the prompt instructs the model to subscribe to crew_updates and weather_updates streams.
Model burns 3-5 tool calls retrying before eventually outputting code-as-text (0 captured tool calls)
T3 verifier then sees incomplete state (jobs not completed, only 2 of 5 assignments)
The eval still passes because the pre-seeded schedule + state validation is lenient, but the stream processing path is never actually exercised
Expected
Stream factories for crew_updates and weather_updates should be registered via ConstructionPlugin or extraFunctions so the Monty sandbox can subscribe and process events.
Summary
During construction scheduling T3 (dispatcher) eval runs,
stream_subscribe("crew_updates")andstream_subscribe("weather_updates")throwRuntimeError: Invalid argument (name): No stream factory registeredin the Monty sandbox.Reproduction
Run the construction experiment with T3 dispatcher rooms:
T3 rooms use
needsStreams: trueand the prompt instructs the model to subscribe tocrew_updatesandweather_updatesstreams.Stack trace
Impact
Expected
Stream factories for
crew_updatesandweather_updatesshould be registered viaConstructionPluginorextraFunctionsso the Monty sandbox can subscribe and process events.