Evidence grading scale
Two independent axes, applied per claim. A single “good or bad” score would be easier to read and considerably less honest.
Axis one
Evidence level
Multiple high-quality human trials, or strong systematic-review or meta-analytic evidence.
Human controlled evidence exists, but replication or sample size is limited.
Small human studies, observational evidence, or early-phase clinical work.
Predominantly animal or in-vitro evidence. No adequate human outcome data.
Biological plausibility or mechanistic reasoning without adequate outcome evidence.
Evidence is absent, irreconcilable, or too weak to support any directional statement.
Axis two
Confidence
How settled the finding is, reported separately from how strong the underlying design was. A single small but well-conducted trial and a large trial with contested methodology can occupy the same evidence level while differing sharply in confidence.
High confidence
Replicated, consistent, and unlikely to change materially with additional evidence of similar quality.
Moderate confidence
Reasonably consistent, but replication is limited or the evidence comes from a narrow set of sources.
Low confidence
Sparse, inconsistent, or concentrated within a small number of research groups.
Very low confidence
Effectively unreplicated, or originating from a single programme with limited independent scrutiny.
Interactions
Severity scoring
Severity describes how consequential an interaction would be if real. It is reported separately from certainty, because a hypothetical serious concern and a well-established mild one are different things.
No interaction meeting our evidence criteria is documented. This is not a statement of safety.
A shared pathway or overlapping domain worth knowing about, without a specific concern attached.
A concern warranting attention and discussion with a clinician.
A significant concern, typically supported by approved labelling or by clear mechanistic redundancy.
Flagged in the platform as a combination we will not present as researchable together.
Interactions
Classification types
Documented interaction
Reported in human evidence or in authoritative product labelling.
Potential interaction
Plausible and partially supported, but not established in dedicated study.
Labeled contraindication
An approved product's labelling states the combination should not be used.
Mechanistic overlap
The compounds act on the same pathway. Reasoning-based, not an observed interaction.
No qualifying signal identified
Nothing meeting our criteria was identified. Usually this means nobody has looked.
Inadequately studied
The question has been asked but not answered by adequate evidence.
Conflicting evidence
Evidence points in more than one direction and cannot currently be reconciled.
An X is not a failing grade