Skip to content

orchestration: recovered-node completion lacks direct assertion in build_task_prompt/metrics paths (SC-4/SC-9) #6301

Description

@bug-ops

Description

During implementation review of #6243 (orchestration node control parity, spec-075), the impl-critic and tester independently flagged that two spec-075 §7 success criteria are only transitively covered by existing tests, not directly asserted:

  • SC-4: a Mode-1-recovered node's synthetic TaskResult value reaching build_task_prompt() for downstream dependents — covered indirectly through dependency-injection tests, no direct assertion that the injected state_injection string is what a dependent's prompt actually receives.
  • SC-9: a recovered node (via try_recover) is counted in tasks_completed through finalize_plan_completed — the counting mechanism is generic/status-derived, but no test explicitly asserts a __recovery__-derived Completed task increments this count.

Expected Behavior

Both criteria should have a direct, dedicated test rather than relying on transitive coverage through other mechanisms.

Actual Behavior

Only indirect/transitive coverage exists today. Risk assessed as low (mechanisms are generic, not recovery-specific), deliberately deferred out of PR #6243/#6265's scope by team-lead ruling to avoid scope creep.

Environment

Related

Metadata

Metadata

Assignees

Labels

P3Research — medium-high complexityorchestrationTask orchestration / DAG schedulingtech-debtTechnical debt

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions