Function append_stamps_a_recorded_time
pub async fn append_stamps_a_recorded_time<F: Fixture>(
open: impl AsyncFn() -> F,
) -> RuleOutcomeExpand description
VT-9 — the store stamps a time at append, and reports that same value on every read of that event.
Read the prohibitions before strengthening this rule. VT-9’s third
MUST forbids the contract from stating any relationship between
RecordedAt order and SequencePosition order, and the conformance suite
is the contract’s executable form: an adapter author reads a failing rule
as a requirement, so a rule that asserts an order states one. Therefore no
comparison of two events’ recorded times in either direction, no
comparison of a recorded time against a position, and no plausibility
check against the harness’s own clock — which CF-33 forbids independently.
RUNBOOK.md’s “a rule that it is non-decreasing with position” was struck
by ADR-0014 for precisely this reason, and a store on a machine whose
clock steps backwards under an NTP correction is conformant.
What is left is that the value is a fact about the append rather than
about the read, and that is not nothing. ReadTimeClockStore in
tests/mutation_coverage/mutants.rs fills the field from the
connection’s clock in its row mapper, which is what an adapter does when
recorded_at arrives on the port after its schema was written and the
field has to be given something. That store is correct in every other
respect and invisible to any rule that reads once.