Skip to content

Document LLM actuator tool return semantics - #820

Closed
bedaHovorka with Copilot wants to merge 3 commits into
goal-10from
copilot/investigation-actuators-tools-return-values
Closed

Document LLM actuator tool return semantics#820
bedaHovorka with Copilot wants to merge 3 commits into
goal-10from
copilot/investigation-actuators-tools-return-values

Conversation

Copilot AI commented Jul 29, 2026

Copy link
Copy Markdown
Contributor

LLM actuator tools acknowledge queue acceptance, not completed railway actions. This adds a critical analysis of the deferred execution path and its observability gaps.

  • LLM-facing results

    • Documents exact queued ... responses for request_route, release_route, and approve_train.
    • Separates synchronous validation and backpressure errors from later simulation outcomes.
  • Deferred actuation

    • Maps tool invocation through ActuatorCommandQueue, DispatchDecisionApplier, and actuator ports.
    • Identifies outcomes logged but not returned to the initiating tool call.
  • Operational assessment

    • Covers stale approval-cap observations, uncorrelated snapshot feedback, and stateless dispatch-cycle limitations.
    • Proposes an asynchronous, correlated outcome channel and simulation-thread admission-cap enforcement.
request_route(...) → "queued request_route ..."
                   → later: Reserved | AllPathsBlocked | Conflict | NoRouteExists
                   → currently logged, not returned to the LLM

@bedaHovorka

Copy link
Copy Markdown
Owner

done, no merge

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Investigation actuators tools return values

2 participants