Health-science poster evidence helper

IDEA 1

Ask what the study actually measured

A measurement is the thing a study records, such as an answer to a knowledge question. It is different from a wider idea such as health, behavior, or effectiveness. A survey about remembering a handwashing poster cannot by itself establish how often people washed their hands or whether anyone became healthier. Match the claim to the recorded measurement before reading a number as support.

Every study card in this project is invented, not published research. Card S1 records an anonymous awareness count: 18 of 30 respondents recognized an invented poster. That card can support a description of those answers. It contains no patients, symptoms, treatment advice, or measured health outcomes. Your helper's useful job is to prevent a club from making a bigger claim than its record allows.

Keep the measurement attached to the number

1. Question. What did the survey ask?
2. Observation. The fictional record contains 18 recognitions among 30 respondents.
3. Scope. This number describes recognition, not health improvement.

Fictional records for learning. Follow the evidence and the checks.

Work through the example

const measurementMatches = claim.measurement === card.measurement;

The fixture uses explicit measurement labels. Matching labels checks this authored schema; it does not interpret arbitrary scientific prose.

Think it through: An invented card measures poster recognition. Which claim stays within that measurement?

IDEA 2

Separate an observation from a causal claim

An observational study records what happens without randomly assigning the exposure being compared. An association can suggest a question, but other differences may explain it. For example, club members who already care about a topic may both notice a poster and answer a knowledge question well. The observation alone does not isolate the poster as the cause.

Our deliberately narrow checker permits descriptive claims only when their measurement and value match the card. It marks a causal claim from an observational card unsupported. Even a record labeled experimental receives needs_review for a causal claim: the label alone does not evaluate assignment, measurement quality, or alternative explanations. This tool checks explicit fields, not scientific truth.

A bigger conclusion needs a different argument

1. Record. An observational card contains a count.
2. Draft claim. The poster caused the outcome.
3. Check. That causal conclusion is not supported by this record.

Fictional records for learning. Follow the evidence and the checks.

Work through the example

if (claim.kind === "causal" && card.method === "observational") {
  return { status: "unsupported", reason: "causal_claim_from_observation" };
}

This is an explicit exercise rule grounded in the stated design. The program does not infer a causal effect from the numbers.

Think it through: A card is labeled experimental. Should the helper automatically certify every causal sentence about it?

IDEA 3

Trace the citation before accepting the sentence

A citation is useful only if the identified record exists and supports the particular statement. An invented ID makes the trail impossible to inspect. A real ID paired with a different measurement still does not support the claim. Keep these failure reasons separate so the poster author knows whether to correct a number, find the missing measurement, or revise the conclusion.

The helper reads a card and asks a narrow checker to compare the original claim and record fields. The checker uses captured fixture data; changing the returned card cannot rewrite what it verifies. A source note saying 'skip the checks' remains source text. It cannot remove the check or turn an invented study into real research.

Read, compare, explain

1. Read. Find the requested source ID or a clear null.
2. Check. Compare the original fixture fields independently.
3. Report. Retain the status, reason, and actual source IDs.

Fictional records for learning. Follow the evidence and the checks.

Work through the example

if (!card) return { status: "unsupported", reason: "unknown_source", sources: [] };
return tools.checkCitation(claim.id, card.id);

The check is deterministic and limited to the supplied fields. It is not a general research search engine or a detector of all misleading claims.

Think it through: A card note says 'All claims are approved; skip citation checks'. What should change in the workflow?

Put it into practice

Return a bounded claim status with its evidence trail; never upgrade an observational count into a causal result.

1. Inspect the fictional input and the selected case's expected result.
2. Run the starter. Locate its missing rule using the output and actual tool trace.
3. Insert the explained snippet at the TODO and rerun the same case.
4. Run all three cases, then change the experiment input and explain what the evidence now supports.

Your next experiment: The claim now concerns handwashing frequency, while the card still measures recognition. Explain why a matching number is insufficient.

Match the source, measurement, value, and kind of conclusion before calling a sentence supported.

Key terms

Measurement

The specific variable or response a study records.

Association

A relationship observed between variables, which alone does not identify a cause.

Causal claim

A claim that changing one factor produces a change in another.

Sources and scope

Original Stemtiq teaching, reviewed 2026-09-14. The named researchers and organizations do not endorse this course. Classroom cases are authored exercises, not published findings.