experiment-design

This skill automates experiment config generation by reading formalization.md, benchmarks.md, and source.md, then outputting a runnable experiments/<exp_id>/config.yaml without manual drafting.

Pre-registration in experiment design involves defining hypotheses and failure modes before running trials. This skill pre-registers hypotheses, failure modes, baselines, and venue trial counts directly into the generated config.yaml.

Yes, you can generate experiment configs for reproduction, diagnostic, and approach modes. The skill validates mode alignment with your project stage and preserves design provenance during config generation.

No, the experiment config generator does not launch the experiment. It only generates the config.yaml file and enforces design checks against doctrine guidelines and review_plan requirements during generation.

You need to place formalization.md, benchmarks.md, and source.md in your project directory. These files provide the formalization, benchmark definitions, and current state required to generate the config.yaml.

Design checks are enforced during experiment config generation by validating the design against doctrine section 8 and review_plan section 1.6. This ensures the generated config.yaml aligns with doctrinal guidelines.