Expand description
Host repeat sets, rustev.host-repeat-report/1 (spec 061, 3.2 to 3.5).
N host runs of one host flow version, each bound to its views and
scores, aggregate with spec 029’s statistics, intervals, bands and
slices, computed by the same functions over the host case measures. The
source is host: statistically a captured set, never replayed, never
deterministic. Every figure is computed by Rustev from host-asserted
observations, and the report says so.
Structs§
- Host
Cohort Report - Host
Flow - The host identities every run of a set shares (3.2.3); the run handle is per run.
- Host
Metric Report - Host
Repeat Report - Host
RunRef - HostSet
- An assembled host set: its report and each run’s case measures.
- Host
SetInputs - The inputs of a host set: the configurations and each run, in run order.
- Scorer
Check - A declared scorer, and what Rustev did not reproduce of it (3.4.6, 3.5.2).
Enums§
- Host
Observation - Every latency and cost figure of a host set is the host’s own observation (3.5.3), never a candidate’s or a historical one.
- Host
Source - A host set’s only source (3.4.3).
- NotChecked
- What Rustev did not check for a host set (3.4.6, 3.5.2).
Constants§
- DUPLICATE_
RUN - The
mixed-runsdetail for a repeatedhost.run(3.2.3). - HOST_
LEDGER_ UNSUPPORTED - Why a host set on a held-out split has no count (3.8).
- HOST_
REPEAT_ REPORT - TOP_
K_ NOT_ OBSERVED - Why a repeat configuration with
top_kis refused (3.4.5).
Functions§
- import
- Assemble a host repeat set (3.2.3 to 3.5): bind and measure every run, check the runs against each other, and compute spec 029’s figures per cohort and metric.
- observed
- Whether a host metric is an observation of the host’s runs (3.5.3).
- reassemble
- Re-assembly (3.3.4): the set rebuilt from its inputs must reproduce the
supplied report and case documents byte for byte (
reconcile).