Experiment Feasibility

Use when a proposed experiment needs a grounded GO, CONDITIONAL, or NO_GO decision covering backend fit, baseline preservation, data availability, compute burden, metric support, and validation path.

woosook0127 Updated

File contents

Experiment Feasibility

Use this before implementation or long training.

Procedure

  1. Read the idea/proposal/spec and current repo constraints.
  2. Confirm backend mapping and baseline/copy strategy.
  3. Check data availability and path-binding assumptions.
  4. Check metric/evaluator availability and trust.
  5. Estimate compute/runtime burden at a practical level.
  6. Define the smallest smoke path that would reduce risk.
  7. Return one decision: GO, CONDITIONAL, or NO_GO.

Output

  • feasibility summary
  • technical fit
  • operational risks
  • missing evidence
  • recommended next team/skill
  • exact condition for moving forward if conditional

Guardrails

  • Do not hide missing runtime support.
  • Do not collapse feasibility into implementation.
  • Long training, dataset binding, and external credentials remain human-supervised.

woosook0127/Codex_Skillset/tree/main/.agents/skills/experiment-feasibility commit a5bfade72c

Frequently asked questions

npx skillmds@latest add woosook0127/experiment-feasibility