Subscribe →
How I evaluate evidence

The standard behind
every claim.

This is the canonical version of how I think about evidence — on the channel, in claim reviews, and on your project. If a verdict ever seems harsh or generous, this is the reasoning underneath it.

Citation policy

Distilled Science prioritizes peer-reviewed human evidence when available, clearly distinguishes mechanistic plausibility from clinical outcomes, and flags conflicts of interest, endpoint switching, preregistration issues, and overgeneralized claims. Every scientific claim is cited.

The hierarchy of evidence

Strongest → weakest
  1. 01
    Systematic reviews & meta-analyses of randomized human trials — the top of the ladder when the underlying trials are sound.
  2. 02
    Randomized controlled trials (RCTs) in humans, adequately powered, pre-registered, with a clinically meaningful endpoint.
  3. 03
    Prospective cohort / observational useful for signals and rare outcomes, but confounding means association ≠ causation.
  4. 04
    Case-control & cross-sectional hypothesis-generating; prone to recall and selection bias.
  5. 05
    Mechanistic & animal / in-vitro explains *why* something might work — does not prove it works in people.
  6. 06
    Anecdote, testimonial, expert opinion the weakest evidence on its own, however confident the source sounds.

Higher isn't automatically "true" and lower isn't automatically "false" — a well-run cohort beats a sloppy RCT. But the ladder is where I start.

What I check on every claim

Outcome, not mechanism

A plausible mechanism ("it reduces inflammation in a dish") is not a human outcome. I separate the two every time.

Endpoint switching

Did the study measure what it set out to, or did a surrogate marker quietly replace the outcome that matters?

Absolute vs relative risk

"50% reduction" can mean 2% → 1%. Relative numbers inflate; I look for the absolute change.

Sample size & power

Small studies produce big, unstable effects. Underpowered trials get over-read.

Preregistration & p-hacking

Was the analysis declared in advance, or fished out of the data after the fact?

Conflicts of interest

Who funded it, who stands to gain, and is that disclosed? Funding doesn’t void a study, but it changes the prior.

Surrogate endpoints

A biomarker moving is not the same as a person living longer or feeling better.

Generalizability

Does a result in 22 young male athletes apply to the person reading the headline?

Supplement-study red flags

When I get more skeptical, not less.

  • A single small study presented as settled science
  • Mouse or petri-dish results described as if they happened in people
  • Relative-risk numbers with no absolute baseline
  • "Clinically proven" with no citation you can actually read
  • Proprietary blends and undisclosed doses
  • A supplement brand citing a study it paid for, on an outcome it didn’t measure
How I communicate uncertainty

"We don't know yet" is a real answer.

Most health questions don't have a clean yes/no. I'd rather tell you the evidence is mixed, early, or absent than manufacture a confident take for the algorithm. When I say something is supported, situational, or a myth, that label carries the weight of the evidence behind it — and I update publicly when better data arrives.