Methodology

Studies, identifiers, evidence grades, interaction classifications, regulatory statuses, reported quantities, durations, and every derived statistic originate from structured data or from deterministic services operating on that data.

No language model produces any of these values. When an assistant layer is added, it will call internal tools to retrieve verified results and will convert those results into readable explanation. It will not be permitted to originate a fact, and it will not be the database.

A compound can have strong human evidence for one endpoint and none at all for another. Grading the compound as a whole hides exactly the distinction that matters. So a peptide record can, and often does, carry claims graded B, D, and X simultaneously.

Evidence level and confidence are separate axes, because a serious hypothetical concern and a well-established mild one are different things and collapsing them into one score destroys that difference. See the full evidence grading scale.

Every claim also carries an explicit statement of what remains unestablished, rendered with the same prominence as the finding. For most compounds in this category, that gap is the more important half of the record.

A study record is marked verified only when its PubMed identifier was checked against the source registry. For every other record we publish no identifier at all and link to a PubMed search instead.

This matters more than it might appear. A citation that does not resolve is worse than no citation, because it manufactures an appearance of rigour that cannot be checked. Records that are not verified are excluded from every derived statistic on the platform.

Values such as the most commonly studied duration, the longest qualifying human study, and total participants are computed by a deterministic service that returns the exact study identifiers used. Every derived figure on the platform can be traced to the specific records that produced it.

Where no qualifying evidence exists, the service returns nothing and the interface states that an evidence gap exists. It never falls back to a plausible-looking number.

When you select a research duration, it is compared against the longest qualifying human study. If your horizon is longer, the platform states that an evidence gap exists rather than extrapolating. Findings from shorter studies cannot be assumed to persist, and several compounds catalogued here showed effects that reversed after discontinuation.

The calculation engine is a separate, tested package. Values are stored canonically in micrograms and microlitres using decimal arithmetic, so floating-point display formatting can never become the source of truth for a number someone acts on.

Beyond worked examples, the engine is tested against mathematical invariants across thousands of randomly generated valid inputs:

• Concentration multiplied by solvent volume must return the original material quantity.
• Target volume multiplied by concentration must return the target quantity.
• Required containers multiplied by quantity per container must meet or exceed total demand.
• One container fewer must always be insufficient. Testing only the first direction would allow systematic over-ordering to pass.

Inputs are rejected explicitly rather than silently coerced: zero or negative quantities, targets exceeding available material, NaN, Infinity, unit mismatches, results below the smallest readable graduation, and results exceeding device capacity. Where a displayed value differs from the exact value, both are shown.

Targeting WCAG 2.2 AA. Specifically for this platform:

• Evidence level and interaction severity are never communicated by colour alone. Every badge carries a letter, a word, or a distinguishing glyph.
• The measurement-device visualiser carries a full text alternative describing the calculated position and whether it falls between graduations.
• Sortable table headers are buttons with aria-sort state, and every table has a caption.
• Motion respects prefers-reduced-motion. Focus indicators are visible throughout.