One aggregate in an Aggregate. field is None only for Count, and arg
is the fraction for Percentile (e.g. 0.95). alias names the output field
in each result row.
A grouped aggregation carrying one or more AggCalls, so a single query
can return several aggregates grouped by the same keys. An optional window
adds a time-bucket key.
The server-side gate that enforces a Consistency level the same way on
every backend. The client refuses an unadvertised level before sending,
but a backend that does advertise read_your_writes or strong_consistency
still has to honor the level, and the rule is fail-not-downgrade: serve only
when the projector’s applied offset for the queried source has reached the
required offset (the source log head at query time), else return
QueryError::Stale rather than a silently older read.
A lexical relevance search: the text to match and, optionally, the one
indexed field to match it in (None searches every text-hinted field).
Relevance is returned as an ordinary typed result field. Capability-gated by
KEYWORD_SEARCH: a backend without a lexical index answers unsupported,
never a contains approximation.
How fresh a query’s view of the materialized index must be. A materialized
view is a read model a projector builds by tailing the log, so it is
eventually consistent: a record is queryable once the projector has applied
it, not the instant it is appended. This level says what the query requires
of that lag, and the contract is fail-not-downgrade: a level that cannot be
met returns QueryError::Stale rather than silently serving older data.
A scalar value used by the graph and older non-query document surfaces. Query
predicates and results use the tagged TypedValue contract. This remains
untagged because it is part of the separate graph wire surface.