# Proposal review rubric: evidence before sophistication

Use **clear**, **needs clarification**, or **not yet supported** for each dimension. Do not total a spurious numerical score or rank proposals by institution, method, data size, or journal aspirations.

| Dimension | Evidence the reviewer should see | Useful reviewer question |
|---|---|---|
| Economic contribution | Decision, labor margin, two competing explanations | What would we learn beyond this organization's report? |
| Estimand and population | Initial eligibility, treatment/control, horizon, weights | Which workers or sites are included after an exit or failed application? |
| Identification | Actual variation and explicit remaining threat | What would change without the reform? |
| Measurement | Metric definitions plus independent validation | Could the recording system create the result? |
| Inference | Independent units, dependence, test null, uncertainty | Does the method address this assignment mechanism? |
| Portability | Named target, modifiers, overlap, version | What assumption moves the estimate out of the source? |
| Welfare | Worker and firm outcomes with a consistent boundary | Which transfers have been double counted? |
| Feasibility | Access status, minimal schema, fallback | What can be done now, not only after a hypothetical partnership? |
| Reproducibility and care | Permissions, privacy, versioning, documented restricted inputs | What can another researcher actually rerun? |

**Constructive verdicts:** proceed with the scoped design; narrow the question/population; or obtain a missing comparison or measurement before the claimed inference. A null result, one-firm setting, or familiar estimator is not a reason to reject a well-posed design. An experiment, prominent partner, or large dataset is not a reason to ignore missing assumptions.

The final oral defense should include: a strongest-claim sentence, an unsupported-claim sentence, the most valuable next observation, and a response to one concrete criticism. The course ends with a defensible research plan, not a publication guarantee.