What we mean by evidence
The word "evidence" gets used in everyday speech and in research, but not always with the same meaning. Here, evidence means carefully designed and reported observations from research studies — not personal experience or single anecdotes.
For example, saying "creatine improves strength" counts as evidence-based when several independent research teams run randomized controlled trials (RCTs), and a meta-analysis pooling those results finds a consistent effect in the same direction. A single observation or personal report is not yet evidence.
When a term slows you down: Every bolded research term below (meta-analysis, RCT, confidence interval, and so on) is defined with an everyday analogy in the companion piece Evidence glossary. Skim it once, and come back whenever a word gets in your way.
Studies have a hierarchy of strength
Not all evidence carries the same weight. Different study designs give us different levels of confidence in the conclusion. Roughly, higher on the ladder means the design lets you draw more direct causal conclusions.
- Meta-analyses / systematic reviews: Pool results from many studies on the same question. Generally the strongest form of evidence
- Randomized controlled trials (RCTs): Randomly assign people to intervention or control, giving the most direct view of cause and effect
- Cohort studies: Follow groups over time without intervening. Large scale, but observational
- Case-control / cross-sectional: Data collected at one point in time. Good for observing associations, weak for causation
- Case reports: Single cases. Useful as hypothesis starting points but not conclusive
The three-step confidence bar shown on MiraProof (weak / moderate / strong) reflects this hierarchy along with sample size, number of studies, and how well the studies agree with each other.
The space between "works" and "doesn't work"
Evidence is rarely black and white. Most topics sit in the middle:
- Consistently supportive: Multiple RCTs and meta-analyses agree on the direction of effect
- Conditional: Results depend on population, dose, or duration
- Mixed: Supportive and negative studies roughly balance each other
- Insufficient: Not enough high-quality research yet, or quality is too low
The verdict labels MiraProof uses ("Effective," "Partial," "Limited") try to convey this middle band. Read them expecting nuance, not a yes/no answer.
Reading the numbers — "significant" does not mean "large"
This is the single easiest thing to misread.
"Statistically significant (p < 0.05)" means that the result is unlikely to have arisen by pure chance. It does not mean the effect is large. In a study with tens of thousands of participants, differences too small to notice in daily life can still be "significant."
- Effect size: how big the observed difference is (e.g. sleep duration up by 15 minutes on average)
- Confidence interval: the range in which the true effect most plausibly lies (narrower means more precise)
Always read significance, effect size, and the width of the confidence interval together, never one on its own.
The limits of evidence
Even strong studies have limits.
- Limited populations: Results from young healthy adults may not apply directly to older adults or people with specific conditions (see population)
- Short observation periods: An effect seen over 8–12 weeks says little about what happens after a year or five years
- Researcher conflicts of interest (COI): Funding sources can shape how results are interpreted
- Reproducibility: A single dramatic finding is weaker than a modest finding that other teams have reproduced
That's why the honest phrasing is often "based on current evidence, we can say this." Evidence is not a fixed fact — it updates as new research comes in.
How to read the articles here
Each MiraProof article shows the current state of evidence for a topic together with the type and number of studies, the direction of results, and the population conditions. The goal is to make the reasoning visible: not just a claim, but the research path that supports it.
To go one step deeper on how to weigh a study in front of you, the companion piece How to judge study quality walks through four practical lenses — design, sample, conflicts of interest, and reproducibility.
This article is not medical advice. For questions about your own symptoms or treatment, consult a physician or pharmacist.