Cannavec Skill
The evidence library arranges cannabis research by condition: an evidence ladder from systematic reviews down to preclinical work, one plain-language brief per paper, every record checked against Retraction Watch, drawn from PubMed and Europe PMC.
When a reading list is not enough, run an Evidence Review from the same place: frame the question in PICO, search, screen, extract and grade, with every statement traceable to a verbatim quote in a source you approved. Built for researchers, medical affairs and regulatory teams who have to defend their conclusions.
An Evidence Review stops at four points and waits for a decision: the protocol before anything is searched, the records screening could not settle, the evidence table, and the final report. Everything between those points runs on its own — and everything it did is recorded.
Each extracted value arrives with a verbatim quote, and that quote is checked against the source document. A value whose quote cannot be located is withheld from the evidence table and queued for you — it never enters the record unnoticed.
Every record is scored for relevance and ranked so the strongest matches surface first — nothing is excluded by the score. Every exclusion beyond what the search metadata itself rules out is a decision you make, not the model's.
Risk of bias uses the right instrument for the design — RoB 2, ROBINS-I, or AMSTAR-2. Certainty is calculated in code from design and downgrade domains, never produced as a number by a model.
Search logs, exclusion rationale, and every approval are recorded in an append-only, hash-chained dossier. Editing any entry invalidates every entry after it — enforced by the database, not by application code.
Searches run only against named databases, and every query is recorded exactly as it was sent — so the search can be re-run and compared a year later. Retractions are screened on every record, under every configuration, without exception.
The Evidence Library and Evidence Review are available to paid Cannavec subscribers. Both can also be driven via MCP from Claude Desktop, Cursor and other AI tools on paid plans.
Every review exports a complete audit dossier