Under what architectural conditions does research data survive long horizons?

Some information systems have survived multi-decade operational horizons on the open Internet — DNS for 43 years, email for 44, BitTorrent for 25, Git for 21 — without any single organization underwriting their continuity. Other systems, including many built specifically to preserve research data, have failed within their first 12 to 20 years.

This question asks which architectural properties distinguish the systems that survive from the systems that do not, and whether those properties are transferable to scientific data preservation.

The framework developed by the paper organizes the answer around three properties: distribution of independent copies across independent failure domains, verifiable integrity without trusting the holder, and independence of persistence from any single organization's governance, funding, or operational continuity. The four-tier taxonomy (Tier 0/1/2/3) is the structural classification under which research data infrastructure is read against those properties.

This question is the analytical foundation for the rest of the paper. Every Claim about loss mechanisms, costs, prevention, and verification reduces to whether the underlying architecture supplies these three properties.