Measuring a draft
Overview
Measure-Loom inspects recorded timings from the most recent executed draft in the current runspace. It warns if there has been no draft; it does not execute one itself.
In a draft
There is no measurement verb. Run a draft, then inspect its timings:
Invoke-Loom {
Thread Reed
Sley git { Command status }
}
Measure-Loom
From the command line
Measure-Loom | Format-Table Phase,Name,Harness,Line,Depth,Inclusive,Exclusive,Count
Measure-Loom -GroupBy Harness
Measure-Loom -GroupBy Verb
Get-Shed -State Failed
Rules
Rows cover prepass, imports per harness, validation, verbs and total execution, together with staged application under Deferred. Inclusive includes nested verbs; Exclusive subtracts their time. Line and Depth identify an invocation; grouped rows use Count. Grouping aggregates verb rows only, ordered by exclusive duration, so use ungrouped output for imports and staged work. Do not sum nested inclusive rows as a wall-clock total.
The exact Phase values are Prepass, Import, Install (missing harnesses on first runs), Validate, Verb, Capture (staging/rewrite), Deferred and Total.
The steady-state budgets, with harnesses already installed, are kernel import under 50 ms and a typical Reed draft under 150 ms over import. CI uses median measurements with a 1.2 noise tolerance. First-call import/type initialization and gallery downloads should not be confused with a cached style lookup or Tab benchmark. Deferred work remains real startup work even when it runs after the initial total row.
See development setup for the independent startup and BenchmarkDotNet budget checks.