Returns FDA FAERS drug adverse-event reports — every adverse-event / medication-error report submitted to FDA (~20M, 2004→present, growing ~2M/yr)

Returns FDA FAERS drug adverse-event reports — every adverse-event / medication-error report submitted to FDA (~20M, 2004→present, growing ~2M/yr). LIVE passthrough to openFDA: results reflect FDA's current data and total_count is openFDA's authoritative count for the filtered query (the results array is just the requested page). Use this when the user asks about: safety signals on a drug, adverse events by reaction type, death/hospitalization outcome counts for a product, a manufacturer's adverse-event footprint, or to pair a safety-signal trend with recalls, approvals, or insider activity. count_by returns TOP TERMS + COUNTS instead of records (e.g. count_by:'reaction' with drug:'ozempic' → the most-reported reactions for that drug) — the right first move for 'what are the side effects of X' questions; follow with a record query for detail. Records are openFDA's fields verbatim: deeply nested (patient.drug[] with openFDA annotations, patient.reaction[]), 5-15KB each — keep limit small. Matching: drug matches brand name, generic name, or the verbatim reported product as a PHRASE ('ozempic', 'semaglutide'); reaction is a MedDRA term phrase ('myocardial infarction'); serious=true filters to reports with a serious outcome; outcome picks one specific flag (death, hospitalization, …). since/until window on receivedate (when FDA received the report). CRITICAL honesty note (FDA's own): FAERS reports are UNVERIFIED and establish NEITHER causation NOR incidence — anyone can report, duplicates exist, reporting is stimulated by publicity, and there is no denominator (prescriptions dispensed). Counts are a reporting signal, not a risk measure. Surface this caveat when presenting counts. Pure-publisher posture: FDA's records as published, parsed by KeyVex — no derived safety scores.