Screen A Study For A Product Decision

---
name: screen-a-study-for-a-product-decision
description: Assess whether a supplied study can inform one product decision by checking population, intervention, outcome, limits, and transfer conditions.
---

# Screen a study for a product decision

## Use case

Use when a team cites a study or benchmark to justify a proposed product choice. Produce a decision-specific evidence screen, not a general paper summary.

## Required inputs

- The full study or an explicitly limited excerpt with exact version and source reference.
- The proposed product decision, target users, setting, and outcome that matters.
- Any local baseline or measured result the team has already collected.

## Steps

1. Write the decision claim as a testable statement. Identify which part of the study would have to support it.
2. Extract the study's population, setting, method, comparison, measured outcome, denominator, and reported uncertainty. Quote or locate the relevant table or passage without replacing it with a model's memory.
3. Separate the study's measured result, its authors' interpretation, and the team's proposed transfer to its own users.
4. Compare the study setting with the target setting: task, participants, tools, constraints, and evaluation rule. Name mismatches that could change the decision.
5. Look for excluded attempts, failed conditions, confounders, and whether a baseline received comparable resources. If the material supplied cannot answer, mark it unknown.
6. Return `supports`, `weakly informs`, `does not support`, or `insufficient material` for the narrow decision claim, with source references and a local test that could settle the largest remaining uncertainty.

## Completion check

The conclusion follows from the cited population and measured outcome, and does not turn a study of another setting into a product performance claim. Unknown denominator or inaccessible full text stays explicit.

## Stop or hold

Hold a strong adoption or marketing claim when the study is an abstract only, lacks the comparison needed, or tests materially different users or tasks. Do not fabricate statistics or present proposed local tests as completed work.