What Different Source Types Can Establish

A source can be genuine, accurately quoted and still be incapable of supporting the conclusion attached to it.

The key question is not only:

Is this source real?

It is also:

What kind of claim can this source actually establish?

Primary research creates observations

Primary studies collect or analyze original data.

But different designs support different inferences.

A randomized trial may estimate intervention effects under controlled assignment.

A cohort study can estimate associations and sometimes support causal inference with appropriate design and assumptions.

A case report documents an event but usually cannot estimate frequency or causation.

A laboratory experiment can establish behavior under experimental conditions without automatically establishing real-world effects.

Randomized trials are strong for some causal questions

Random assignment can reduce confounding in intervention studies.

That makes randomized trials powerful for estimating treatment effects.

But a randomized trial may still be poor evidence for:

• very rare harms;
• decades-long outcomes;
• populations not studied;
• exposures that cannot ethically be assigned;
• mechanisms outside the trial’s design.

“RCT” is not a universal evidence trump card.

Observational studies answer many indispensable questions

Observational data are essential for:

• long-term harms;
• environmental exposures;
• prognosis;
• prevalence;
• rare outcomes;
• natural experiments.

Their causal interpretation depends heavily on confounding, measurement and study design.

A strong observational study can answer a question that no randomized trial can ethically address.

Systematic reviews synthesize; they do not manufacture stronger inputs

A systematic review asks what the whole eligible literature shows.

A meta-analysis can combine effect estimates.

But synthesis quality depends on the included evidence.

A systematic review of weak, biased or highly indirect studies may still yield low-certainty conclusions.

Guidelines and consensus statements add judgment

Guidelines often combine:

• evidence synthesis;
• certainty assessment;
• benefits and harms;
• values;
• feasibility;
• expert judgment.

They can be excellent sources for current practice recommendations.

But a recommendation is not identical to the underlying empirical evidence.

The two layers should remain distinguishable.

Patents establish intellectual-property claims, not efficacy

A patent can show that an invention was described and that legal criteria were met.

It does not prove:

• that the device works;
• that a treatment is clinically effective;
• that a proposed mechanism is correct;
• that regulators approved it.

This distinction is why Patent as Proof Error is a recurring Reasoning Error.

Regulatory documents answer regulatory questions

An authorization, approval, warning or classification has a defined legal and technical scope.

It can establish what the regulator concluded under its framework.

It should not be paraphrased into a broader scientific claim without checking that scope.

News reports are navigation, not primary evidence

High-quality reporting can:

• identify documents;
• interview relevant experts;
• explain context;
• surface investigations;
• summarize a complex event.

For central empirical claims, the evidence trail should usually continue to the underlying paper, dataset, filing or primary record.

A news article saying “a study found” is not a second study.

Preprints require special care

A preprint can contain valuable and high-quality research.

It has not yet completed the journal peer-review process.

That does not make it false.

It means its review status should be visible, especially for consequential claims.

Source hierarchy depends on the question

There is no universal ladder where one source type always beats another.

For “What did the minister say?” the primary recording may be strongest.

For “Does this treatment reduce mortality?” a well-conducted systematic review of relevant trials may be stronger.

For “Was this patent filed?” the patent record itself is primary.

The best source is the one structurally capable of answering the exact question.

The Claim by Claim rule

For every citation, ask:

What proposition is this source doing evidential work for?

Then verify that the source type, design and scope can actually carry that proposition.