Methodology
How data is sourced, scored, reviewed, and corrected: evidence tiers, the editorial pipeline, and error correction.
where the data comes from
Every value on the site traces to a source. Entries are drafted from published literature indexed in PubMed, the TripSit interaction dataset, drug-checking organizations, clinical trial reports, community forums and experience reports.
Initial substance data was seeded from the Bifrost API (api.psychonautwiki.org), the public data layer behind PsychonautWiki. Seeded values are checked against the sources above and rewritten as they pass through review.
Where sources disagree, that disagreement is reflected in the evidence metrics rather than hidden behind a single averaged value.
the editorial pipeline
Entries move through three stages. AI never auto-publishes. A human approves every entry before it reaches readers.
evidence tiers & confidence
Two separate gradings travel with the data, and they answer different questions. A confidence tier grades a value — this dose range, this duration, this contraindication. A quality tier grades a source — the paper the value came from. A well-supported value can cite a weak source, and a strong paper can support a value we still grade low, so neither tier is derivable from the other.
Value confidence
Shown as a dot beside each section of a substance card. A section covering several values takes the lowest grade among them: the dot never claims more than the weakest value beneath it.
Source quality
Shown beside each entry in a page’s reference list, and in full on the evidence page for that substance.
Alongside the tier, a value carries what kind of evidence stands behind it — controlled study, case series, mechanistic reasoning, expert consensus, community report, animal data. A value the pipeline reached by analogy rather than from a source is marked inferred and is graded low by construction.
Our editorial standard requires a minimum of two independent citations for dosage and interaction data. Safety-critical values — doses, interaction severities, toxicity assessments — come only from cited sources. Any value AI infers to fill a gap is labeled as inferred in the database. Where a value is uncertain, the uncertainty is stated rather than smoothed over.
how errors get corrected
Every substance page has a “suggest a correction” link. Corrections that cite a source are prioritized. You can also reach the editors at contact@psychedex.org.
A correction follows the same path as any other change: an editor reviews it, and an approved fix publishes as a new version. Prior versions are retained, so what changed, when, and why stays on the record. Both are readable per substance: version history under /history, and the sources and tiers behind each value under /evidence.
what we don't host
Psychedex publishes reference data and phenomenological reports. It does not host synthesis instructions, precursor or sourcing information, vendor details, or anything describing how to acquire a substance. Experience reports describe effects, timing, and setting — never acquisition. See the content policy for the full scope.