The wasm umbrella reinstates what the wasm re-pricing demoted

Found 2026-09-01 by frankB while taking the wasm32 group off ready --track A, and independently by frank-coordinator the same hour. Neither of us was looking for it; it is what ready hands you.

The two instructions, both real, one dated later

2026-08-30, in the body of both wasm bug-a- tickets:

"it works, it tests our IR, we should be able to compile applications.. for now, that's good enough." The anchor is met — pascal26 runs under wasmtime and emits an ELF byte-identical to the native compiler's. These tickets stay OPEN and correct; they simply must not outrank ordinary Track A work.

Someone acted on it correctly: both are prio: 25.

2026-08-31, 8d9a5794b, landing the umbrella scheme, created umbrella-wasm-is-a-real-platform at prio 70 with four wasm tickets wired to it. effective_prio takes the max over dependents, so:

Nothing is broken. The ranker does exactly what CLAUDE.md specifies. The re-pricing was applied to the leaves and the goal above them was never given a number consistent with it, so the ruling was reversed by a mechanism that landed the next day and could not know about it.

Why this is probably an accident rather than a reversal

The five umbrellas in 8d9a5794b are priced 90, 85, 80, 75, 70 — exact five-point steps. That is a ladder assigned by ordering, not five independent ratings; wasm's 70 is "last of five", not "worth 70". The commit says the umbrellas were "derived strictly from the stated goal", and the goal document does name wasm in the platform list — so 70 is a defensible reading of the-goal-cross-cross.md by someone who had not seen the 08-30 ruling, which lives only in two ticket bodies.

That is the whole mechanism: a ruling recorded where the ranker cannot see it, overturned by a number chosen from a document that does not contain it.

The fork

• (a) Lower the umbrella to ~25. Honours the explicit, dated ruling; wasm tickets rank on their own prio and stay open and correct, which is exactly what the ruling asked for. Recommended — it is the only option that makes the board agree with a sentence the owner actually wrote.
• (b) Keep 70. Treats the 08-31 umbrella as a deliberate re-rating. Costs nothing to adopt, but no evidence supports it: the ladder tell above suggests the number was never a judgement about wasm.
• (c) Split the umbrella. "Emit correct wasm32" demoted, "host the compiler under a wasm runtime" kept high because the goal's second proof is pxx hosting itself off Linux/x86-64. Weaker than it looks — the 08-30 note says that anchor is already met under wasmtime, so the high half is largely done.

Scope, so nobody over-corrects

Only the wasm umbrella is in question. bug-a-managed-locals-leak-on-an-unwind-on-wasm32-and-xtensa is NOT affected: its 75 comes from umbrella-managed-memory-is-correct, an unwind leak is a managed-memory correctness hole on its own terms, and CLAUDE.md cites that very ticket as the worked example of legitimate multi-umbrella membership. It ranks at 75 under either answer here.

The general shape, which outlives this ticket

A re-pricing applied to leaves is undone by any umbrella later placed above them, silently, and the ranker will look correct while it happens. Worth a progress.sh check aperture: a ticket whose own prio: is far below its effective_prio and whose body contains an explicit ruling is the detectable form. Nobody misread anything here — frankA read 25, frankB read 70, both numbers are real, and only one reflects intent.

RULED 2026-09-01 — option (a), and wider

The owner, asked about umbrella pricing: "we focus on linux only for now. that means demoting bsd and wasm."

So it is not only wasm. umbrella-pxx-hosted-beyond-linux (85) had OpenBSD as its only children and is demoted with it. Both umbrellas to 25; the three BSD leaves whose OWN prio exceeded 25 were demoted too, since effective_prio takes the max of own and inherited and demoting only the umbrella would have left them ranking on their own numbers.

Not demoted: umbrella-cross-target-codegen-is-correct (80) and umbrella-managed-memory-is-correct (75). Those are architectures under Linux (xtensa, i386, arm32, riscv32), not other kernels, and the ruling was about kernels.

One residual, left deliberately and raised to the owner: bug-a-managed-locals-leak-on-an-unwind-on-wasm32-and-xtensa still comes out at 75, inherited from umbrella-managed-memory-is-correct. Its own prio is 25. It is genuinely both things — a managed-memory correctness bug whose remaining work is wasm32-only — so the edge is not false and cutting it would falsify the umbrella's dependency. Under a Linux-only focus it will sit near the head of Track A and not be takeable. Flagged rather than silently re-edged.

Log

• 2026-09-01 — decided, commit e9d1b7850.