The owner's rulings of 2026-09-30 written where they govern

[Japabu]

The owner's rulings of 2026-09-30, each written where it governs. Issue files only, plus one string in src/licence.rs.

This PR lands after #636, whose amendment of root CLAUDE.md's firmware rule admits the CPU microcode the microcode ruling orders the kernel to load; the defect points at that rule.

The rulings

• Q2 (stage 2): an end reads as an exit, a kill or a fault kind alike on every architecture, and a bare code reads the last two as failures.
• Q6d (stage 2): no end reads as a quit's reason.
• Q3a (stage 3): libc imitates SIGCHLD.
• Q3b (stage 3): a C handler runs at once, beside the program: in a blocking libc call, or on libc's own thread while no thread is in one.
• Q3c (stage 3): a C child starts with descriptors 0–2 and exactly what its file actions name, a stated departure from POSIX.
• Q5a (stage 5): a login's session is a program that logout ends and that only parents what its user starts.
• Q5b (stage 5): init hands the compositor or sshd, as it starts a session, a right to start programs in that session alone.
• Q5c (stage 5): sshd's right also quits and kills its own session.
• Q6a (stage 6): a quit carries its reason: interrupt, hang-up or terminate.
• Q6b (stage 6): a quit kills a process that never watched its notice. The stage already said so.
• Q6c (stage 6): a quit reaches the process's subtree by stage 4's walk.
• Q6e (stage 6): std hands a program its notice, the ctrlc fork waits there, and rustc stops skipping its handler. The stage already said so.
• Q6f (stage 6): libc takes the three reasons as SIGINT, SIGHUP and SIGTERM, and kill with one of them is a quit. The stage already said so.
• Q7 (stage 7): a second Ctrl+C kills only a program that has not yet taken the first.
• Microcode: the kernel is to load CPU microcode signed by the CPU's maker and pinned by version and hash. The question becomes the defect issues/kernel/the-kernel-loads-no-cpu-microcode.md, whose exit is current microcode loaded early on every CPU, at least as current as Linux's.
• std::os::toyos::io::{AsRawFd, FromRawFd} are renamed the next time the trait is touched in the fork. os-toyos-io-traits-keep-a-posix-name.md becomes an open defect with that exit.
• A boot start's device refusal is fatal only for a device the [programs] row marks as required. Otherwise init logs it loudly and starts the program without it. a-boot-start-refused-a-present-device-runs-without-it.md becomes an open defect whose exit is the row's mark and both outcomes.
• doomgeneric's fetch stays, so its question file is deleted.
• doom.jpg is the owner's own screenshot and he keeps it; no licence question applies. The question file is deleted, and the row's third column (free text: "names the file that carries the attribution, or says why it is ours") now says it is his screenshot and that no licence question applies. The terms column stays NOASSERTION.
• The blocked-task dump keeps painting its report on the panel and holding it there. One line at the small-kernel track's step 5 replaces both step 5's open question and stage 6's hold on it. Step 6's heading "The scheduler knows nothing about devices" goes, since step 5 now keeps the pass painting the panel.

Decisions

• Track length. a-childs-end-is-an-event-and-a-parent-takes-its-children-down.md is 260 lines, the same as before. It is 35 bytes longer (18608 → 18643). Q2, Q6d, Q3a, Q5a–Q5c and Q7 state things the track did not say before. These were cut as moot:

  • the pointer to the question file;
  • the Q labels and (ruled) marks on the stage headings, which distinguish nothing once every stage is ruled;
  • the Ruled block's session clause, which moved into stage 5's Q5a line;
  • stage 6's reason why the shell relays nothing, which the Q6c line now states;
  • stage 7's "Stop and continue need a suspend primitive and are not proposed". That sentence scoped the proposal the rulings closed.

  Stages 2, 3, 5, 6 and 7 are re-wrapped where the lines changed.

• Citations. The supervisor track's open item "the ask's ABI, which is Q6a of …" is removed. Q6a is ruled, and that track's stage 3 already asks by stage 6's quit. The security track's list follows the microcode rename. git grep of each deleted slug, and of its bare name, finds nothing.

• opened stays as it was on the three rewritten files.

Gates

• cargo run -- --ci host at 71b3a57: EXIT=0 (56 steps, all green)

🤖 Generated with Claude Code

https://claude.ai/code/session_016t9wjdQkB8SH7bmfUoiy6L

[Japabu]

Review, round 1, head 7eb4428

CI: host passed at 7eb4428 (run 36779929674, conclusion success). The branch adds no test and targets no hardware.
Net lines: +113 −459 (−346). Production is one src/licence.rs string (+2 −2), issues are +111 −457, and there are no tests.

The points the brief asked about:

• Rulings 1, 3, 4, 5 and 6 each landed where they govern, and nothing left in those files contradicts them. Ruling 1's 14 answers sit in the child track: Q2 and Q6d at :64–66, Q3a and Q3c at :72–73, Q3b at :81–84, Q5a–Q5c at :171–175, Q6a and Q6c at :182–185, Q6b at :187–189, Q6e at :191–194, Q6f at :198–201 and Q7 at :242–243.
• Rulings 2 and 7 each leave a line that contradicts them (below).
• Citations: git grep finds none of the four deleted slugs or their bare titles in the tree, and rg finds none in rust/ or ~/.cargo/git/checkouts/. No Q-label is left anywhere.
• 98 bytes longer at 260 lines: the new lines state rulings the track did not carry before, so that growth is accepted. 63 of the 98 bytes are the eight stage labels, which now all read "(ruled)" and so distinguish nothing (see REMOVE).
• The stop-and-continue sentence cut from stage 7: it contradicts no ruling, since none touched stop or continue. The stages' own lists still decide the scope, because kill answers EINVAL for every signal except those stages 3 and 6 name. The cut stands.

BLOCKER

• CLAUDE.md:63 — The rule still admits only vendor firmware "loaded only by that device's own driver through its IOMMU domain; it never executes on the CPU". That bars the CPU microcode ruling 2 orders the kernel to load. issues/kernel/the-kernel-loads-no-cpu-microcode.md:9 states the ruling but drops the deleted question's pointer at this conflict. An agent briefed on the defect would then meet an always-loaded rule that forbids the fix, so the ruling has not landed where it governs. Fix: in this PR, an agent the orchestrator briefs for it amends that sentence to admit CPU microcode that the kernel loads, signed by the CPU's maker and pinned by version and hash, without the file growing.

NOTE

• none

REMOVE

• issues/kernel/the-kernel-is-small-interrupts-post-and-threads-wait.md:147 — "The scheduler knows nothing about devices." — Ruling 7 makes it false: step 5 (:184–186) now keeps the pass painting the panel, "a device the pass reaches".
• issues/kernel/a-childs-end-is-an-event-and-a-parent-takes-its-children-down.md:40,54,64,70,120,171,182,240 — " (ruled)", and "ruled; " in stage 3 — Every stage now carries it, so it distinguishes nothing. This is 63 of the 98 bytes.
• src/licence.rs:399–400 — "he keeps it, and rules that" — The committed file already shows it is kept, and "no licence question applies" carries the ruling.
• PR body, "## Unsure" — It is addressed to this review and carries nothing main's merge record needs.

SEND BACK