What this is, and what it is not

Before generating anything, your wording is screened for patterns associated with medical emergencies — including chest pain with radiation, stroke symptoms, severe breathing difficulty, anaphylaxis, uncontrolled bleeding, suicidal ideation and infant fever. When one matches, an urgent-care message is placed above the answer, not inside it. This screening is deterministic, not left to the model's judgment.

• · AI systems can be confidently wrong, including about details inside a correct-sounding answer.
• · Medical evidence changes, and a source that was current may no longer be.
• · General information cannot account for your history, medications, allergies or examination findings.
• · Reference ranges, dosing and guidelines vary by country, laboratory and individual.

When evidence is weak or conflicting, we say so rather than smoothing it over. Uncertainty is stated plainly.

Reported answers enter a physician review queue. Flags raised automatically by the safety layer — emergency language, medication-related content, unverifiable citations — are logged for the same review process, and patterns feed back into the prompt and source policy.