experimentlisted

ultrathink # Experiment Lifecycle Orchestrator Use when designing, running, tracking, or evolving structured experiments. Enforces hypothesis validation, reproducibility, and evolution tracking. Domain-agnostic — works for LLM evals, HPC benchmarks, statistical analyses, anything structured. **Trigger:** When user types `/experiment` with a subcommand. ## Iron Law ``` NO EXPERIMENT WITHOUT A HYPOTHESIS. NO RUN WITHOUT A PLAN. NO EVOLUTION WITHOUT A RATIONALE. ``` ## Arguments - `$ARGUMENTS` — subcommand and parameters: - `init <name>` — create new experiment with directory structure + spec - `plan <name>` — mandatory pre-flight + batch planning (stops for approval) - `run <name>` — execute approved plan with reproducibility enforcement - `evolve <name>` — update spec with rationale + version bump - No argument or `help` — display usage summary ## Anti-Rationalization Table | Excuse | Reality | |--------|---------| | "Just run it quick" | Pre-flight exists because poorly-designed experiments waste more time than the 5 minutes of validation | | "I'll document the changes later" | Later never comes. The version history IS the documentation | | "The results speak for themselves" | Results without a prior hypothesis are unfalsifiable post-hoc stories | | "I just need to tweak one parameter" | That's an evolution. Record what changed and why, or in 3 months you won't know which version produced which results | | "I'll add the config later" | The hook will block y