Pilot experiment execution — read Pilot Spec YAML, write pilot code, run experiment(Confirm with the user before operation and require the applicant to conduct manual inspection), return results. Called by /ideate Phase 5. Does NOT modify wiki pages or judge pass/fail.

Pilot code generation from a YAML spec reads fields like setup, metrics, and hypothesis to create runnable scripts under experiments/pilot/code/{slug}/. It requires explicit user confirmation before executing the diagnostic experiment to produce raw metrics.

Remote SSH pilot execution supports running experiments via tools/remote.py, which includes connectivity and GPU availability checks. You can execute diagnostic runs on remote machines while maintaining human-in-the-loop safety confirmations before launch.

Pilot experimentation produces raw diagnostic metrics for validating research ideas without judging pass/fail or modifying wiki pages. It focuses on baseline comparison and multi-metric reporting to assess hypotheses quickly before committing to full evaluations.

Pilot execution saves per-seed JSON results to provide granular diagnostic data for baseline comparison. It outputs raw metrics without producing pass/fail verdicts, ensuring researchers can manually inspect the outcomes for idea validation.

Yes, manual inspection of the generated pilot code and configurations is required before execution. The workflow enforces human-in-the-loop safety by demanding explicit user confirmation to launch local GPU or remote SSH training runs.

Automated pilot experimentation should not be used when you need automated pass/fail verdicting or direct wiki page modifications. It is designed strictly for generating raw metrics and baseline comparisons requiring manual review.