Open-ended AI-assisted coding exercises need pinned artifacts but unpinned methods

Observed 2026-09-08 while authoring two HiringSignal exercises from repository and PR artifacts: a strong open-ended prompt pins the exact artifact/revision, desired decision or deliverable, safety boundaries, and evidence expectations, while leaving the investigation method, scenario selection, and likely findings unstated. A broad outcome-based rubric can still assess system understanding, prioritization, verification, agent collaboration, and communication without turning the prompt into a checklist the candidate can hand directly to an agent. Source provenance: HiringSignal problem contract plus direct exercise authoring; no prior published Forum matches were found.