[Japabu]

Problem

Both dev hosts, the Mac and the x86 guest runner's host, now run QEMU 11.1.3. The owner's rule is to upgrade rather than downgrade. build/ci/qemu-version declares 11.1.2 (#1177), so --ci guest jobs on either host red at the instrument step.

Change

On top of #1177:

• build/ci/qemu-version: 11.1.2 → 11.1.3.
• issues/the-nightly-is-frozen-so-nothing-runs-the-guest-matrix.md: ci: #1173's review fixes — the guest runner never holds main's scope, iowait is idle, non-green runs keep logs #1177's text, updated. The pin is now 11.1.3, the guest runner's and the dev host's. guest.yml's container still installs 11.1.1 and will red at its instrument step until it carries 11.1.3. The exit names 11.1.3.
• issues/the-build-runs-host-tools-outside-rust-and-qemu.md: the declaration is now 11.1.3, the guest runner's and the dev host's.

Fixtures and citations that name 11.1.1 are unchanged. They record what was measured.

Checks

All runs are on the Apple-silicon dev Mac with QEMU 11.1.3 for both architectures.

• cargo run -- --ci host at 24d164c: exit 0, 108 steps green.
• Before the rebase, cargo run -- --ci rows virt_user_mode at f83f1f2 (the same pin change on the older base): exit 1, ending HOST NOT QUIET. Both instrument steps passed on QEMU 11.1.3, and virt_user_mode passed. Other agents' builds used 2.74 CPUs during the run, so the harness gave no verdict. - cargo run -- --ci rows virt_user_mode at 24d164c, started when the 1-minute load average was 3.71 after about 27 minutes of waiting: exit 1, ending HOST NOT QUIET. Both instrument steps passed on QEMU 11.1.3, and virt_user_mode passed. Other work used 4.43 CPUs during the run, so the harness gave no verdict. The version check this PR changes is exercised and passes. No row on this Mac has a quiet verdict.

Risks

• guest.yml (frozen nightly) reds at its instrument step until its container carries 11.1.3, as the nightly issue records.

🤖 Generated with Claude Code

Orchestrator note: the x86 guest runner host reports QEMU emulator version 11.1.3 for both qemu-system-x86_64 and qemu-system-aarch64 (checked over ssh, 2026-11-04).

[Japabu]

Review of #1176 at 24d164c

Architecture. The change is the right size. build/ci/qemu-version is the instrument's only declaration, and build/ci.rs's instrument and qemu_version_note read it at run time. Changing the single line moves both the gate and the dev-host note. Nothing else in the tree hard-codes the declared version: guest-rows.yml refers to the file, and build/firmware.rs derives the machine type from the running QEMU's version. 11.1.3 still selects pc-q35-11.1 and virt-11.1, so firmware selection is unchanged. Following the hosts' upgrade, rather than holding them back, applies the owner's rule as stated. The PR claims no design decision beyond that.

Declarations moved, records untouched. git grep '11\.1\.2' at the head finds nothing. The remaining 11.1.0/11.1.1 mentions all record a past measurement or a cited source: the ACPI fixtures and their SOURCE files, the licence.rs entries, the vdagent and virtio-console measurements, the hw/.../target/i386 line citations and the issue evidence. The two version strings in tests are parser and machine-type inputs. They are not the declaration. All of these are rightly unchanged.

Issue texts. These are accurate. guest.yml still installs qemu-system-* from the snapshot archive named by the digest-pinned sid-20260918 image. The host-tools row records that archive as carrying 11.1.1, so the frozen nightly's instrument step will red against 11.1.3. The nightly issue says so, and its exit now names 11.1.3. The host-tools row's "the guest runner's and the dev host's" matches the owner's statement that both hosts were upgraded.

Evidence. This is sufficient for the change. Moving the declaration does not change how any guest behaves. The QEMU binaries changed when the owner upgraded the hosts, not in this PR. The PR's behavioural claim is narrow: the instrument step accepts 11.1.3. Both architectures' instrument steps passed on the Mac at the head, and --ci host passed at the head (the_declared_version_is_a_version covers the file's shape). The HOST NOT QUIET ending withholds a timing verdict, which this change doesn't need. It does not cast doubt on the observed instrument lines.

Findings

No blockers.

• SUGGESTION — issues/the-build-runs-host-tools-outside-rust-and-qemu.md:27, issues/the-nightly-is-frozen-so-nothing-runs-the-guest-matrix.md:18 — the guest runner's 11.1.3 rests on the owner's statement. No run on that host observed it; the Mac's x86_64 check was TCG on the Mac. The owner is the right source for their own host, and a mismatch reds loudly at the instrument step, so this does not block. The PR exists to unblock guest-rows on that runner. A single-row guest-rows.yml dispatch of this branch would observe its instrument line cheaply, if the orchestrator wants that before landing.
• For the orchestrator, about the PR description only: in Checks, the second virt_user_mode bullet is joined onto the first ("...gave no verdict. - cargo run..."). It needs a line break.

LAND