evidence · 4.0 KiB

mainline-ordinary-baseline-repeat-2026-10-05.md

Committed at docs/research/mainline-ordinary-baseline-repeat-2026-10-05.md. This is what grounds the requirements that cite it.

Open raw source

Repeat the unchanged ordinary baseline before another intervention

The quiet p2 ordinary run reached systemd, mounted sysroot and started closure lookup. Subsequent debug/console, info/console and info/kmsg comparisons showed only fresh kernel/pre-exec init output. None identifies a failing instruction. Defaults can later select journal-or-kmsg, so explicit info/kmsg is not equivalent to every baseline backend state.

Group5r repeats the exact quiet policy once: same p2/24h artifacts and controls, synchronous initramfs, markers disabled, no rd.systemd.log_* tokens. Existing typed parent selectors suffice; no new selector/source change is needed. Actual preparation must recover299 argument/317 literal command bytes, identical root/init/masks/sole console and unchanged source/config/Image/DT/initrd/init/manifest/loads/helpers. Rebind the report to NEW protected recovery after the kmsg capture before host qualification and one180-second passive board capture. Complete private logs, exact printed/live arguments and unknown-no-input rules remain; recovery is separate. Reproduction is a temporal control, not proof of logging causality, ordinary-root acceptance or a console fix.

Read-only inspection of the exact p2 archive finds a oneshot closure unit, RemainAfterExit and RequiresMountsFor=/sysroot/nix/store, without closure drop-ins. Its built script SHA256 is77fb33e11767ed5e1b01f3e2d4acf23640f3886fa737130ae5a256350d9a9218. It reads cmdline, resolves init inside sysroot, links the volatile closure, checks prepare-root and writes switch-root configuration; activation runs prepare-root later. Pinned NixOS unit/script source.

Pinned job.c emits Starting after unit_start returns; service.c spawns/registers the main process before SERVICE_START. That status supports progress beyond merely queueing, but does not prove Bash/loader/helper body or a nested operation completed. Its own output-call return is separate.

The selected resolver exists as a RISC-V ELF in the script PATH; absence of a /bin alias is not a missing-tool result. Read source path.rs digest6673235a2bd91864aa4b676fc157e16dac22f8763959a83bf5a4d606861793ba shows Root::open, root.resolve then procfs fd readlink/output; this does not identify a stalled syscall on the board. If the closure boundary reproduces, separately scoped finite helper-entry/before-resolve/after-resolve markers could bracket helper progress, preserving semantics. Marker writes perturb observation; missing markers remain unknown. That instrumentation is NOT part of5r and has not been implemented. If baseline stays pre-PID1, helper instrumentation is premature.

NEW protected recovery and actual unchanged-baseline qualification passed independent review. Physical repeat completed with independent review PASS: fresh kernel/pre-exec init only, no systemd/closure/login in180s; the earlier closure boundary did not reproduce. Closure-helper instrumentation is therefore premature. Subsequent NEW protected recovery passed independent review. Ordinary root/panel/glass remain UNVERIFIED; task5b.5 stays open. No kernel/initrd/DT build, masks/targets/tracing/PID1/IRQ/MMIO change or claim from the source scout.