[avrabe]

The refuting command ran first, and did not refute

The v0.84 design answer still stands in the artifact tree, the bounding assertion that memory 0 carries no suffixed alias still holds, and the call-site count still reads nine.

The correction this lane owes on itself

The first draft of this PR took the reporter's fit measurement — two domains reserving 64 + 128 KiB against boards of 8, 64 and 96 KiB — as refuting the premise that the layering is the whole remaining blocker, and recommended doing the sizing half first.

That was wrong, and the dates settle it. Their comment is dated 2026-10-06 and names the version they measured. The release that honours a declared byte page on the relocatable path was tagged 2026-10-09 — three days later — and its changelog entry credits this same reporter.

Verified against this tree rather than inferred:

• the capability is reachable on exactly the flag set their transcript shows them compiling with (--relocatable);
• each memory is sized by its own declared page, through the twinned byte getters;
• __synth_mem_size_k and __synth_mem_region_k are emitted for every memory, not memory 0 alone.

So the fit number is stale by one release, the premise holds, and the layering is the remaining blocker — which is what the artifact said before this lane argued with it. The order follows from the shipped tree, not from a judgement: sizing is done, layering is next.

Worth naming, because it nearly shipped: the sizing issue is still open even though its capability shipped, so reading the issue state alone gives the wrong answer here. The tree had to be read. The apparent tension between the reporter's two asks is likewise resolved, not open — the issue itself argued the declared size was the one worth trusting, because an embedder need not believe a derivation, and the declared one is what shipped.

The four code measurements, which stand

1 — the nine call sites are the wrong target. The memory index is carried by a wrapper variant, not a field; the decoder wraps a load/store only when its memarg names a non-default memory, so memory 0 keeps the bare variants and single-memory streams stay bit-identical by construction. The selector's envelope arm spans 152 lines with zero references to the static-data classifier — it relocates through its own per-memory region symbol, and neither helper it calls reaches the classifier. All nine sites lie before that arm begins.

2 — the classification window is already memory-0-scoped. Upper bound: the first declared memory's initial size (all_memories.first() in the compile driver, the value handed to the selector's native-pointer setter). Lower bound: the stack-pointer global's initializer.

3 — a narrowed symbol claim. A search for the full region-symbol name cannot contain a prefix test spelled with a shorter one. Re-run over the short spellings, every match compares for equality, and a prefix form appears for no suffixed name at all. Positive control: the same spellings return 11 real matches elsewhere in the same crates. Scoped to shipped crate code — this lane did not locate the post-link in-range oracle and asserts nothing about it.

4 — now the sharpest remaining bound, promoted from a footnote. The envelope arm covers the i32 access family only (8 variants), so an i64, f32, f64 or v128 access on memory k still declines loudly. Lifting the two refusals would move the blocker for any module whose memory k is touched at another width — and nothing here establishes which widths the reporter's two-domain module uses.

What is actually costing the reporter

Found by verifying a held draft rather than by reading code: the reply that says the sizing shipped has been drafted and held for three releases, while their thread still carries the stale fit framing. Its claims were re-verified against this release's tree and they hold; only its version stamp is behind. A correct answer nobody sent is worth nothing to them.

What this lane deliberately did not do

Neither refusal is lifted. Both were located — a CLI pre-flight and the selector's own arm, the pre-flight being the one their transcript shows them hitting. Lifting them is gated on an execution differential under their exact flags, which per this repo's own rule lands first.

Conduct rule 4, stated against this lane

Measuring and correcting a scope is not progress on an externally reported defect that is still open. Nothing the reporter runs behaves differently after this release than before it.

Refs #1145
Refs #1441

🤖 Generated with Claude Code

https://claude.ai/code/session_01YJK5LZZEkV5smCY1jKn18L

[avrabe]

Superseded by #1537, which carries this branch's content in full.

Why: a dependabot auto-merge moved main and put every lane PR behind=1. The merge gate's currency check (rev-list --count head..base must be zero) means merging any one lane PR re-stales the rest, so the six lanes were forced sequential — six rebases and six full CI cycles, with three more auto-merge-enabled dependabot PRs still queued to move main again.

The consolidation was verified by simulation before being chosen: the whole wave applied to a detached worktree off origin/main, the single ci.yml conflict resolved with both steps kept, and gated there at no CI cost — status_evidence 0 failures, claim_check 75/75, verdict_prose clean, and the window anti-vacuity guard untripped.

Nothing was dropped. The artifact records are byte-identical apart from the PR number they name, plus one correction the simulation surfaced: SPELLSCAN now carries issue-scope: outlives, because #1515's own subject is the SELFTEST2 vacuity and that lane is partial.