Module bench 

The benchmark projection: every externally callable operation of the registry, as a target, derived and never listed by hand. A capability with a required benchmark policy is a target directly and through every transport its exposure declares; the inputs come from its input type’s BenchmarkCases; the transports’ own operations (initialize, tools/list, /openapi.json, …) are system targets declared once in system. Coverage is covered / required with a generated denominator; the runners time targets through the executor, a real loopback socket and a real child process; the results are a versioned document; the accepted baseline is compared against a policy that is data.

Re-exports

pub use results::BenchmarkResult;

pub use results::ResultDocument;

pub use results::RESULT_SCHEMA;

Modules

baseline

Accepted baselines and the regression policy. A baseline is a tracked result document of one platform (os-arch-buildprofile), promoted explicitly from a local run by bench baseline update; the policy is data in .ai/repo/benchmarks/rust/policy.yaml (the shell tool keeps its own evidence beside it, under .ai/repo/benchmarks/). A check compares each measurement of a run with the baseline’s under the policy and reports every line; a target the baseline knows and the run does not is reported as stale, never silently attached to something else, and a baseline of another platform is not compared at all.

results

The result document: a versioned record of one benchmark run, with the provenance that makes it comparable (commit, dirty state, build profile, platform, registry fingerprint, profile) and, per target, the statistics of each cache mode measured. Local runs are written under the checkout’s local half; the accepted baseline is a tracked file of the same shape, minus the machine-local fields, promoted explicitly.

Structs

BenchmarkProjection

Every target of a registry against one repository.

BenchmarkTarget

One thing to time.

Coverage

The coverage document.

CoverageLine

One requirement: a capability on a transport, or a system target.

Profile

How much to measure.

Runner

Times targets against one context.

Statistics

The statistics of one set of samples, in microseconds.

Enums

CoverageState

Where one requirement stands.

SystemTarget

One system target.

TargetKind

What a target is. A data enum: the capability variant carries the case input and is the common one, so it is not boxed for the sake of the small system variant.

Transport

The way a target is reached.

View

Which reading of the repository a page request is answered from.