# Data Sufficiency Checklist

Before choosing a method, check whether the dataset can support the intended question.

## Unit And Structure

- What is one row: learner, attempt, event, message, session, class, artifact, or time window?
- Is there a stable participant or group ID?
- Are repeated observations nested inside learners, teams, classes, schools, or tasks?
- Are timestamps, turn numbers, or sequence positions available when the question is temporal?

## Measurement

- What construct is each variable supposed to represent?
- Is the construct directly observed, coded, inferred, or proxied?
- Are scales scored consistently?
- Are reliability or coding agreement details available?
- Are missing values meaningful, accidental, or structurally impossible?

## Design Support

- Was there an intervention, assignment, comparison group, baseline, or repeated measurement?
- Was assignment randomized, quasi-experimental, convenience-based, or observational?
- Does outcome timing occur after the predictor/intervention?
- Is there leakage between predictors and outcomes?

## Validity And Generalization

- What population/context does the sample actually represent?
- Was the model or coding scheme validated outside the development sample?
- Could demographics, platform access, consent, attrition, or missingness bias the answer?
- What would a skeptical reviewer say the data cannot show?