Experiment Feasibility
Use this before implementation or long training.
Procedure
- Read the idea/proposal/spec and current repo constraints.
- Confirm backend mapping and baseline/copy strategy.
- Check data availability and path-binding assumptions.
- Check metric/evaluator availability and trust.
- Estimate compute/runtime burden at a practical level.
- Define the smallest smoke path that would reduce risk.
- Return one decision:
GO,CONDITIONAL, orNO_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.