# Measurement-validation proposal and data dictionary

Submit a **two-page study proposal** and a **one-page data dictionary**. Choose one bounded monitoring, evaluation or feedback intervention. A new firm agreement is not required for this assignment.

## Study proposal

**Economic question and decision-maker:** [Name the organizational rule, who chooses it and the economic margin at issue.]

**Baseline population and assignment:** [Define eligibility before the intervention, the assignment unit and probabilities or comparison rule. If randomization is constrained, list the actual support. State the follow-up horizon and the analysis weights.]

**Information map:** [Describe which own/peer/task records workers, supervisors and researchers can access, and when. Distinguish more information from new incentives, observation or disclosure.]

**Two outcome families:** [Prespecify one measured performance margin and one distinct, independently recorded outcome family. Explain what each misses. Specify denominators, units, timing, rubric versions and record producers. Do not place minutes and arbitrary quality scores on an unlabeled common scale.]

**Worker welfare and confidentiality:** [Add a worker-welfare measure such as effort cost, stress, autonomy, schedule predictability or job amenity value; explain what can actually be learned. State who can access identifiable records, which aggregates are released and how reporting protects participants.]

**Competing interpretation and discriminating prediction:** [Name at least one explanation besides increased production: changed detection, recoding, task reassignment, learning, selective exit or evaluator behavior. Identify a measurement or comparison that could distinguish it.]

**Inference and remaining limits:** [Address the intervention's assignment/dependence units, how many are available, interference and attrition. If randomization inference is proposed, state its null and allowed assignments. A large p-value is not equivalence; a leave-out score does not solve endogenous assignment. State what would remain unidentified even with random assignment.]

**Access and limited conclusion:** [Classify needed inputs as public, restricted, requiring a partnership or unavailable. A code/supplement listing does not establish original-data access. End with one defensible economic conclusion and one welfare/production claim the study could not support.]

## One-page data dictionary

| Record/field | Unit and key | Producer and access | Time/horizon | Observed for whom? | Measurement limitation |
|---|---|---|---|---|---|
| Baseline population | [key] | [source] | [window] | [risk set] | [selection] |
| Assignment/exposure | [unit] | [source] | [timing] | [population] | [take-up/interference] |
| Primary performance | [units] | [source] | [horizon] | [denominator] | [rubric/noise] |
| Independent family | [units] | [source] | [horizon] | [denominator] | [coverage/error] |
| Welfare margin | [units] | [source] | [horizon] | [respondents/population] | [nonresponse/interpretation] |
| Exit/task reassignment | [event key] | [source] | [history] | [baseline workers] | [missing post-exit outcomes] |

**Reproducibility note:** [Command, seed, files inspected, Mode A synthetic status, and any modifications. Do not call the toy latent truth empirical evidence.]