A transcribed benchmark landed in a .dag authority on the branch that ported the rule against it

[briansrls]

#9154 merged at 03:01:12Z. These two corrections were committed at 03:02:50Z and lost the race by 98 seconds, so they are landing here instead. Verified by ancestry rather than assumed: 5cdb3ec093 is not an ancestor of main, and the defect below is live on main right now.

1. A transcribed measurement is sitting in a .dag authority

dag/gunbc/repo_self_build.dag carries, on main:

a measured 59.8s/2.5GB against build at 155s/4.7GB

That is transcribed output preserved as authority inside a module, which the measurement-bankruptcy rule forbids in as many words: name the instrument, never transcribe its output. It arrived by an ordinary route — review 55633 correctly backed out the CargoBuildSubject/CargoVerb axes as speculative widening, and their deferral note stayed behind with the numbers still in it.

Worth recording plainly, because it is the strongest available argument for the rule: this was committed by the same branch that ported the corrected rule into DESIGN.md. The number was authored faster than the rule against it was remembered, by the author who had just written the rule down.

The enduring claim is qualitative and is what the note now carries — cargo check establishes the required compile diagnostics without paying for final binary production — with the re-deriving producer named and the magnitudes left as a property of the run that took them.

2. The argv fork is given a terminus, not only a blocker

The previous commit measured why extdeps.rust.cargo_build cannot be consumed: every cargo argv word there — cargo, build, -p, --bin — appears only inside a transport shell block, so no fn can read it. Stating a blocker is not bounding a debt.

This declares the string renderer a bounded workflow adapter, not a second cargo authority of equal standing, with the three-way split written down (cargo invocation shape → extdeps; package/bin policy → here; workflow shell rendering → the workflow realization) and a dissolution trigger naming its end: it dissolves when the operation shape is de-fused from transport shell and can bind a workflow-text realization, at which point this module produces a typed invocation and the argv words leave it entirely.

Because a terminus nobody enforces is decoration, the reviewer instruction is stated on the carrier: refuse any new argv word added here that the modeled surface already owns.

Provenance

Both corrections are from the v2 management thread's #9154 review. Its other two items are moot and were checked rather than assumed — they were written against e4b942a, before the axes were backed out, so the free manifest_path they object to no longer exists and the axis controls they ask for have no axes to cover.

Verification

• transcribed values: 0, dissolution trigger present: 1 on this branch
• parse-checked via gunbc run --entry dag/gunbc/repo_self_build.dag — no diagnostics beyond the known // annotation-routing noise that clean main also produces
• one file, +2/-2