The Evidence Half-Life: Modeling Decay Rates Across Trust Signal Classes

The working assumption is that once a trust signal is established it remains useful until something visibly breaks. A clean registration record, a set of corroborating mentions, a published claim, or a correctly configured infrastructure element is treated as durable evidence. Organizations record the fact, file it, and continue to rely on it for later decisions about citability and source reliability.

That assumption fails because different classes of evidence lose their usefulness at different rates. A registration fact that was accurate two years ago may still describe the same entity. A content claim that was accurate two years ago may no longer be true, or may no longer be the version the domain stands behind. The age of the observation is not the same as the age of its reliability.

Four Classes of Evidence, Four Decay Profiles

Evidence that supports a domain's trust position falls into four classes. Each class has a characteristic rate at which its usefulness declines and a different reason for that decline. The classes are registration records, third-party corroboration, content claims, and infrastructure configurations. They are not interchangeable. A fresh observation in one class does not refresh the others.

Registration records change slowly. Ownership, registrar continuity, and core identity facts tend to persist until an explicit transfer or lapse occurs. Their half-life is long; the main risk is an unrecorded change rather than gradual obsolescence. A registration snapshot from eighteen months earlier is often still a usable description of the entity, provided no transfer has been observed.

Third-party corroboration decays at a moderate rate. Mentions, directory listings, and independent references remain valid only while the citing sources themselves continue to treat the domain as current. When those sources update, consolidate, or stop linking, the earlier corroboration loses force. A dense set of references from three years ago does not automatically constitute present corroboration.

Content claims decay fastest. Statements about products, capabilities, locations, or operational status are accurate only for the period in which they were published. Domains revise pages without notice. A claim that was correctly extracted last year may now be contradicted by the current page, or may have been removed. Content evidence must be revalidated against the live asset; historical extraction is not current evidence.

Infrastructure configurations occupy a middle ground with sharp failure modes. Email authentication records, certificate status, and nameserver stability can remain unchanged for long periods, yet a single misconfiguration or expiration can nullify them immediately. Their decay is often step-like rather than gradual: useful until the moment they are not.

Decay Categories and Revalidation Priority

The practical taxonomy treats each class as belonging to one of three illustrative decay categories: slow, moderate, or rapid. Registration records are slow. Third-party corroboration and infrastructure configurations are moderate, with infrastructure carrying an additional binary failure risk. Content claims are rapid. The categories are not numerical half-lives; they are relative priorities for how often an observation must be refreshed before it can still be treated as current evidence.

Revalidation follows the decay order. Content claims require the shortest interval. Infrastructure configurations require the next shortest, especially where expiration or revocation is possible. Third-party corroboration can be checked on a longer cycle but must be sampled for continued independence. Registration records need the least frequent refresh, yet still require confirmation that no transfer or lapse has occurred. A schedule that treats every signal as equally durable will systematically over-weight the fastest-decaying class.

What Stale Evidence Costs

An organization that continues to rely on a content claim extracted two years earlier will treat a domain as supporting a proposition the live page no longer asserts. Citation systems that evaluate current evidence will not. The mismatch appears as unexplained absence: the historical record looks strong, yet the domain is not selected. The same pattern occurs when infrastructure evidence is assumed stable after an unmonitored certificate or authentication change. The cost is not merely outdated internal notes. It is the continued allocation of effort against a trust position that no longer matches the evidence retrieval systems can actually use.

The age of the observation is not the same as the age of its reliability.

The Question That Precedes Reliance

The useful question is not whether a given signal was once reliable. It is how long ago the observation was made relative to the decay profile of its class. Evidence that has passed its class-specific usefulness window is no longer evidence of the current state. Maintaining a usable record of trust preconditions therefore requires a revalidation schedule ordered by decay rate, not by the convenience of the original collection. Domains whose evidence has been refreshed in the correct order remain evaluable; domains whose evidence has been allowed to age uniformly do not.