A date option in an audit tool is not a clock stub. It decides what a claim is allowed to rest on, and getting that order wrong is how a report turns into a lie with a timestamp on it.
In a11ytrail, our open-source accessibility auditing project, the assessment reads an asOf date, defaults it to now, and rejects an unparseable value outright instead of quietly scoring with a broken clock.
Then, before anything gets scored, the evidence is filtered: only checks whose recorded time falls at or before asOf survive. Everything later is set aside.
The sequencing is the whole design.
The set of scopes under evaluation defaults to the scopes present in the filtered evidence, not to every scope in the full dataset. The per-criterion logic is handed the filtered evidence, so "latest check per scope" means latest as known by that date, not latest ever.
So a scope whose newest check is excluded falls back to its last earlier check. A scope with nothing earlier is simply uncovered: it lands in the unevaluated set, the criterion comes back not-evaluated, and the claim becomes incomplete instead of conformant.
That last part is the trade-off we would defend to any auditor. Incomplete is still a usable answer. A conformant claim built on evidence that did not exist yet is not usable at all.
Two details that cost little and buy a lot: the number of discarded checks is counted and, when positive, adds a sentence to the stated reasons naming the count and the date, so the exclusion is visible rather than silent. Later-dated checks also stay put, since nothing is deleted, they remain in the evidence and just do not decide this claim. Staleness rules then apply only to what survived the cut.
We build this way for the same reason we build payment and compliance systems this way: any claim about the past has to be reproducible from what was known at that moment, including the parts that were missing.
If you produce point-in-time reports, audit, financial or regulatory, the part worth checking is where the filtering happens in your pipeline: before scoring, or as a presentation-layer caveat bolted on afterwards.