An annotation at scope end names nothing, and it has main red

[gunbai-bot]

main is red for every PR right now. One-file repair that unblocks the fleet; not part of my own line of work.

Symptom

Every PR that test-merges current main fails required-witnesses-floor:

required-ci: FAILED PHASE parse (16 error(s))
required-ci: FAILED PHASE namespace-wave-admission (no head index)

The floor itself is clean in the same job — verdict=FloorClean unexpected_failures=0 claims_failed=0, 3621/3621 terminal. No witness is failing.

Root cause — one file, one block

All 16 errors are dag/test/claim/emit_copy_qualification_witness_test.dag:483-500:

source annotation names no subject: no module item follows it.

#10390 deleted the three workflow-subject rows at the end of that module and left their explanatory block as a trailing epilogue at EOF. DESIGN §4c admits only a leading block attached to a module-scope declaration — an annotation names the declaration that follows it, so at scope end it names nothing (AnnotationAttachmentRefusal::UnattachedAtScopeEnd, once per line).

The second blocker is not independent. claim_executor pushes namespace-wave-admission (no head index) only when the parse phase produced no index, so the wave never ran. One root, two reported blockers.

Why it surfaced after #10390 landed

The parse wall is not new, and #10325 only changed which receipt arm carries blockers — not whether parse fails.

The timing is the answer. Run 33869134455 (#10358's last green run on the old tree) was created 11:41:28Z; #10390 merged 11:48:31Z. The run's merge ref predates the breaking merge by about seven minutes, so it never parsed these lines. The first runs to actually test-merge the broken main are the ones now failing.

(An earlier revision of this body said 11:28Z and "twenty minutes". That was a transcription of a different run — 33868113495, created 11:28:30Z, on a superseded head. The causal conclusion is unchanged; the interval was wrong and is corrected here.)

The repair

Moves the block to the module head, where the imports that follow give it a subject. Every sentence preserved. Deictic words rewritten — "stood here" → "stood at the END OF THIS MODULE", "the rows above" / "the mutants below" → "in this module" — because a relocated pointer that still says "below" is a false citation of the kind this repo files as a_live_authority_name_carries_a_superseded_claim.

The blank line placement is load-bearing, and the first revision of this PR got it wrong. std.source_annotation module_header_gap_subject binds a post-module block to the module root only when the block opens immediately after the module line:

if preceded_by_blank_line || preceded_by_annotation_line { none }

A blank line between module … and the first // disables that arm and falls through to ordinary nearest-following attachment — which bound this module-wide block to import std.measure { byte_size }. That made the parse green while the annotation acquired the wrong subject: a plausible answer standing where a typed refusal used to be. The blank line is now removed; the one after the block is kept, because it ends the block.

Nothing else changes — no row, no assertion, no rung drop. The content already has a typed home at gunbc.rung_drop emit_copy_qualification_without_a_consumer, untouched here.

Evidence

	result
before	FAILED PHASE parse (16 error(s)) + namespace-wave-admission (no head index)
after	parse clean, zero parse FAIL lines
attachment battery	source_annotation_attachment_witness_test 13/13 PASS

Run locally with claim_executor --required-ci --source-root dag --source-root src/v2 --required-lane witnesses.

Known coverage gap — named, not fixed here

No enrolled witness covers module_header_gap_subject's blank-line arm — the exact rule that silently mis-bound this block. The attachment battery covers leading, trailing, body-grain and block-splitting cases, but nothing discriminates module-root attachment from nearest-following attachment across that one bit. That witness belongs in its own change; this PR is a fleet unblocker for a red main and stays one file wide.

The gate cannot pass until this merges

namespace-wave-admission now reports:

NotEvaluated — dag/test/claim/emit_copy_qualification_witness_test.dag does not parse
at the base revision (16 diagnostic(s)), so its base-side declarations cannot be read

The head parses; the base does not, because this PR is the fix. The wave needs base-side declarations, so a repair for a main parse breakage cannot satisfy the gate the breakage disables. That refusal is correct fail-closed behaviour, not a defect — but it does mean this PR needs an operator merge rather than a green required lane.

Note for whoever next touches the admission roster

With the wave running again it reports the 255 #10358 rows as consumed at base — deletion owed on the roster's next touch. This PR does not touch namespace_wave_admission.rs, so they are not due here.

🤖 Generated with Claude Code

https://claude.ai/code/session_01U397y4s3dSBof7vGPAX87G

[briansrls]

REQUEST_CHANGES on exact head 46fc682.

The root diagnosis and the one-file remedy are otherwise accepted. Current main carries a standalone leading annotation block after every module item, so it reaches EOF with no following subject and produces UnattachedAtScopeEnd once per line; namespace-wave's no head index is downstream of that parse failure, not an independent admission-roster defect. Moving the block rather than deleting it is correct, and the rewritten deictic wording no longer points "above" or "below" from a new location. The 255 consumed-at-base #10358 admissions are not due in this PR because this PR does not touch their roster.

One source-placement blocker remains. The relocated file currently starts:

module test.claim.emit_copy_qualification_witness_test

// block...

That blank line is load-bearing. std.source_annotation.module_header_gap_subject attaches post-module prose to the module root only when the block begins immediately after the module line; preceded_by_blank_line deliberately disables that arm and sends the block through ordinary nearest-following attachment. On this head the module-wide block therefore attaches to import std.measure { byte_size }, not to the module it repeatedly describes. The parse becomes green, but the annotation acquires the wrong subject—a plausible wrong answer instead of the previous refusal.

Repair narrowly: delete the blank line between the module declaration and the first // line. Keep the blank line after the block before the imports. That makes the existing post-module-header rule bind the whole consecutive block to the module root. No prose redesign, row change, or roster edit is requested.

One body-only evidence correction is also required. GitHub records run 33869134455 as created at 11:41:28Z, not 11:28Z; #10390 merged at 11:48:31Z. The causal conclusion remains correct—the run predates the breaking merge—but the interval is about seven minutes, not twenty. Correct the timestamp/duration without moving the source head.

Run 33884160310 is nonterminal. No merge authorization until the placement is repaired, the body timestamp is truthful, a replacement exact-head review is filed, and exact-head CI succeeds.