corpus(jcl,yaml): plant the #2732 comment-stream idioms, retire the ledger entry

[squid-protocol]

Paired corpus plant for squid-protocol/gitgalaxy#2736 (issue gitgalaxy#2732), which fills the three comment-stream rule absences the epic #2560 n/a sweep flagged: jcl/dead_code, jcl/spec_exposure, yaml/spec_exposure.

Why this is required rather than optional

docs/GATING.md's n/a semantics end a cell's incomparability the moment its rule stops being None, whether or not the corpus has anything for it to match. The engine PR moves nothing in yaml at all across the ~80-repo crucible — the anchored rule finds no [SPEC-n] tag anywhere in real-world YAML — so unplanted, these three cells would land as manufactured measured-0s scored against the median. This is the same paired-work shape as gitgalaxy#2644/#2645/#2646/#2647.

What is planted

Both plants sit on the c file, beside the two comment-stream signals already there (fragile_debt, planned_debt):

language	file	plant	signal
jcl	c.jcl	//*PROBEDEAD EXEC ROSPROC,PARM='D'	dead_code 1
jcl	c.jcl	//* [SPEC-2732] traceability tag for the rosetta corpus	spec_exposure 1
yaml	c.yml	# [SPEC-2732] traceability tag for the rosetta corpus	spec_exposure 1

dead_code was already a plant-if-idiomatic signal in SPEC.md, and JCL has the idiom (turning a statement's // into //* is the universal practice — 13 of the 443 licensed pool .jcl/.prc files do it). spec_exposure joins that same list.

yaml/dead_code is deliberately not touched. yaml has always defined that rule, so its 0 is a pre-existing comparable measurement, not an n/a-to-0 flip — gitgalaxy#2732 scopes it out explicitly. Recorded in the manifest notes so the next sweep sees the asymmetry is intentional.

Ledger

comment-stream-rules-missing-jcl-yaml retires: still_reproduces → false, resolved_by → gitgalaxy#2736.

It also carries a correction, recorded because the filed issue inherited the entry's original reasoning. The entry called yaml's absence arbitrary on the grounds that

flow-sequence brackets cannot FP because spec_exposure never sees the code stream

That is false. detector.coding_analysis applies every non-underscore rule to the code stream, and comment_analysis then adds a second pass over the comments — it supplements the code-stream pass rather than replacing it. Measured: the unanchored generic rule scores spec_exposure=1 on a comment-free workflow whose only bracket is needs: [audit, lint]. The rules that shipped are therefore anchored to each language's comment marker (^[ \t]*#, ^//\*), and gained a \b after the tag alternation because the bare spec branch otherwise matches "specified"/"species".

Verification

Against the engine PR's build (GALAXYSCOPE_BIN → gitgalaxy#2736's full-precision venv):

PASS jcl:  88 assertions across 4 files
PASS yaml: 84 assertions across 4 files
na_check: unreviewed n/a cells: 0 (0 baselined) — no new unreviewed absences

The --report scans confirm no collateral drift: every other signal in all eight files is unchanged, and yaml/c.yml gains spec_exposure only (no dead_code).

Bias report/chart artifacts are not shipped here — bias-history.yml regenerates them at full precision against engine main after merge (AGENTS.md rule 6).

Cross-repo

• Companion: feat(core-engine): jcl dead_code + jcl/yaml spec_exposure comment-stream rules (#2732) gitgalaxy#2736 (engine rules + strict tests + golden-master bless)
• Merge order: the engine PR merges FIRST. This repo's CI checks out the engine at main, so this PR is green by construction only once #2736 has landed — hence draft.
• What re-runs after: this repo's verify.yml on un-drafting; bias-history.yml regenerates the bias report/chart against engine main post-merge; gitgalaxy's rosetta-audit picks up the plants on its next run.

To check this against the unmerged engine PR before it lands, dispatch verify.yml with engine_ref=pull/2736/head.

🤖 Generated with Claude Code

https://claude.ai/code/session_01M8F1cmMsujEjWnnw4d1e4W