Skip to content

Every claim about speed is a recorded measurement

Every externally callable operation is timed directly, over MCP and over HTTP, the denominator is generated from the registry, and a regression against the accepted baseline fails the check.

status
feature status: stable
domain
Evidence

What it does

majordomus bench times every operation through the real transports and reports the slowest first; bench coverage --check refuses a required target nobody covers; bench --check compares a run with this platform's baseline under the policy's thresholds; bench baseline update records a reviewable, tracked baseline and refuses a dirty tree. The executor's counters say whether an execution was answered from the cache, so a claim about work done is a count rather than a clock.

What it does not do

A baseline belongs to the machine that measured it; a value from one platform says nothing about another, and no number in prose is trusted over the file that recorded it.

Why it exists

Operational moments from the Why catalogue that name one of this feature's mechanisms. Nothing here was assigned: a moment appears because it names a rule, a command, a capability or a claim this feature also names.

How it works

The capability modules of the executable this feature names, each with the capabilities it composes. The interfaces each one reaches are listed under Interfaces below: they are projections of one declaration, not separate registrations.

Performance perf behaviorally verified

This process's work counters and phase timings: what happened once at startup and what happens per call, for the structural tests and the benchmark evidence.

Where it sits

domain
Evidence — A claim names the test that settles it, a recorded run is judged against the commit it ran at, and a claim with no recorded run says so.
areas of work
verificationcost
names
Declare onceCI
named by
EvidenceCI

Interfaces

Each capability behind it, on every interface it is projected onto. A link goes to this site's reference for that interface; an HTTP route and an MCP tool are answered by a running majordomus serve, not by this site.

The capabilities behind this feature, by interface
capabilitycommand lineHTTPMCP tool
perf.counters —none GET /api/v1/perf majordomus_perf

Evidence

What is guaranteed

Each claim this feature names, with the status its row declares, the verdict the recorded runs support, and the test that settles it.

Executable use cases

The rules that hold it

  • advisory A performance claim carries its measurementproject.performance-evidence
  • blocking No endpoint of the Rust executable without a benchmark, and no number without evidenceproject.rust-benchmark-coverage
  • advisory Every public command is benchmarkable from the registryproject.benchmarkable-commands
  • blocking A request never rebuilds canonical state, and a cache is never trusted without an equivalence testproject.rust-hot-path
  • advisory A command reads each canonical file at most onceproject.hot-path-reads-once
  • advisory Derived state is computed once per state versionproject.derived-once

The decisions behind it

Using it

The commands that run it

  • majordomus bench inspect

    Time every public command of the registry, cold and warm, and compare with the baseline.

Where this page comes from

the file
.ai/repo/features/benchmarks.md
the same answer
majordomus product show benchmarks
read further

Performance truth

Searches the titles and descriptions of every page this site publishes.