autoresearch

To stop unreliable self-claims in research automation, you enforce an explicit completion gate requiring measurable validator evidence or architect approval plus a concrete output artifact before the loop terminates.

Artifact-gated completion is a validation mechanism that keeps iterative research loops running until a required output artifact passes a validator regime, ensuring progress is repeatable and reviewable across turns.

Start a validator-gated run by issuing a start command with your mission task, chosen validation mode, and required completion artifact path, which persists state inputs in a JSON file to survive across turns.

Mission-driven research automation supports exactly two validation modes: a mission-validator-script writing passing result JSON, or a prompt-architect-artifact including an architect verdict and output artifact path.

Yes, stateful research workflows require persisted state files to store mode selection, validator inputs, and completion artifact paths in JSON format so the process survives across multiple turns.

Strict artifact gating is not suited for open-ended engineering discovery workflows lacking a concrete mission-validation contract or measurable output artifacts, as it requires strict completion criteria to function properly.