medical-source-failover

Medical Source Failover

Retrieve the intended medical document reliably without laundering a secondary source into primary evidence.

Non-negotiable invariant

Degrade the access path, never the document identity or evidentiary standard.

• Keep authority, document quality, access health, and retrieval priority as separate judgments.
• A domain's reputation does not grade every item published on it.
• A repost that reproduces the same document is an access carrier, not independent corroboration.
• A summary, interpretation, search snippet, news item, or social post may help discover an original; it must not support a precise recommendation, dose, contraindication, or evidence grade.

Load the relevant references

• For any domestic medical retrieval, read references/domestic-source-registry.md before selecting sources. It defines authoritative scope, discovery-only sources, fallback relationships, and how to extend the registry.
• When a page is blocked, slow, moved, paywalled, incomplete, or unparseable, read references/failover-policy.md before retrying or switching hosts.
• Before using a repost, mirror, alternate PDF, cached copy, or database full text, read references/document-identity.md and verify that it is the same document and version.

Workflow

1. Classify the requested fact before searching:
   • regulation or legal status;
   • approved drug indication, contraindication, dosage, or label;
   • pharmacovigilance or adverse-reaction notice;
   • reimbursement or payment scope;
   • public-health policy, surveillance, or prevention;
   • national or industry standard;
   • clinical guideline or consensus;
   • effectiveness, diagnosis, prognosis, or harm evidence;
   • teaching or explanatory material.
2. Define the intended document identity from all available fields: normalized title, issuing/developing organization, year/version, document number or DOI, publication venue, and date.
3. Use the registry to choose a source authoritative for that fact type. Do not choose solely by generic tier or domain.
4. Attempt the canonical route within the retry budget. Classify access failures; do not repeatedly hammer a protected host.
5. Follow the failover ladder. Prefer alternate endpoints of the same publisher, official republications, and formal journal versions before verified third-party copies.
6. Verify document identity and version before extracting content from any fallback. Reject mismatches and incomplete copies.
7. Check for a newer version, correction, update, retraction, or withdrawal before relying on the text.
8. Extract only claims supported by the retrieved document. For recommendations, preserve population, intervention, conditions, recommendation strength, evidence certainty, and exceptions when reported.
9. Cite the original issuer or formal publication. If a fallback carried the text, disclose the fallback separately instead of presenting it as a second source.