corpus(jcl): plant the PROC, teardown and executor idioms and re-bless against gitgalaxy#2755 (#2748–#2751)

[squid-protocol]

Companion to squid-protocol/gitgalaxy#2755 (jcl api, cleanup, globals rules; high_risk_execution narrowed — #2748–#2751), now merged. Part of #59's pairing table.

Cross-repo: engine PR #2755 merged first (rosetta:rebless-owed); this is step 2 of GATING.md's "Cross-repo flow" — the corpus re-bless against engine main, green by construction. Verified locally against engine main (3b3568e0): verify_language.py jcl PASS, 88 assertions across 4 files; na_check.py --ci 0 unreviewed; bias_data.json engine_mode full-precision. After merge bias-history.yml regenerates the report, chart and findings.

Plants

file	plant	cells
main.jcl	//ROSPROC PROC … // PEND around STEP2 — the procedure declaration the EXEC ROSPROC steps invoke	api 0 → 1 (file-level, api-non-function-surfaces); func_start stays 4, args stays 13 corpus-wide
main.jcl	PROBERISK / STEP2 re-planted as PGM=IKJEFT01 / PGM=BPXBATCH	high_risk_execution stays 2 under the narrowed rule (IEFBR14 no longer counts)
c.jcl	//DD3 DD DSN=&&TMP1,DISP=(OLD,DELETE) and //DD4 DD DSN=&&TMP2,DISP=(,DELETE),UNIT=SYSDA on probe_cleanup	cleanup 0 → 2 (planted); io 0 → 2 (ledgered probe-idiom-collides-with-measured-signal); DD3 is also the DISP=OLD plant gitgalaxy#2742 owed (sync_locks 1); && temp names add no DAG edge
b.jcl	none — its two existing SET lines	globals 0 → 2 (= median; batch4-dual-keyword-overlaps, the dockerfile ENV dual)

Ledger

• jcl-sync-locks-disp-enq-question — the DISP=OLD plant Plantable cells still unplanted (css/yaml/dockerfile safety), DISP=OLD and %union plants owed, and the pairings for gitgalaxy#2748–#2753 #59 declined is now made. Plantable cells still unplanted (css/yaml/dockerfile safety), DISP=OLD and %union plants owed, and the pairings for gitgalaxy#2748–#2753 #59 declined it because the DD line's io cost bought only one signal; the entry itself named the revisit condition ("revisit if gitgalaxy#2749 lands a cleanup rule on DISP=(...,DELETE)"), and #2749 shipped in #2755. The same line now pays cleanup as well as sync_locks, so the trade is the one Plantable cells still unplanted (css/yaml/dockerfile safety), DISP=OLD and %union plants owed, and the pairings for gitgalaxy#2748–#2753 #59 said would be worth making. still_reproduces: false, decision recorded, entry closed.
• jcl-2610-rebaseline-residual-morphology — cleanup and globals clauses retired (#2749/#2750 reversed #2610's overlap posture and found the scoped-vs-global distinction the entry said JCL lacked); signal narrowed to comment_lines|dependency_links|doc|test. doc/test re-checked against the language crucible (0 structured header keys in 1,610 //* lines; no TYPRUN=SCAN/IEBCOMPR) and stand.
• api-no-rule-defined — narrowed to markdown. api-non-function-surfaces, batch4-dual-keyword-overlaps, probe-idiom-collides-with-measured-signal — jcl added with the shape named.
• batch5-tier2-morphology-shapes — jcl clause rewritten: api now reads via the rule, and high_risk_execution reads the planted 2 (the median 3 is the danger-only string decoy JCL cannot carry — a //* line is comment stream and an operand literal would have to be a PARM value).

The two collective entries also carry #61's safety-plant text; this branch merged main and kept both sides.

Measured effect (local full-precision regen against engine main, not shipped)

bias_report.py --gate → exit 0, 0 unexplained out-of-band cells (10 undefined, 171 ledgered, 116 derived); consistency avg 87% across 49 metrics; length leaks 0/2 weak. jcl cells that move against corpus main:

metric	main	this PR	median	standing
cleanup	0	2	2	green — planted, at the median
high_risk_execution	0	2	3	amber, ledgered. Main reads 0: #2755's narrowed rule stopped counting IEFBR14, so the re-plant is what restores the paired 2
io	3	5	3	out of band, ledgered probe-idiom-collides-with-measured-signal — the accepted cost of the two teardown DDs
sync_locks	0	1	0	ungated unplanted input, so the +1 costs nothing — the point #59 turned on
raw_arch_api	0	0.25	3	file-level surface, ledgered api-non-function-surfaces
risk_safety_score	19.2	40.9	42.6	moves onto the median
risk_api_exposure / risk_documentation	5.3 / 29.3	6.2 / 33.4	7.1 / 38.3	both move toward the median

globals already read 2 on main (the rule landed in #2755 and b.jcl's SET lines carry it with no plant). func_start 13 and args 13 are untouched — the PROC wrapper and the PGM= re-plant disturb neither.

The regenerated docs/bias_* are deliberately not in this PR (AGENTS.md rule 6): bias-history.yml regenerates them against engine main after merge.

🤖 Generated with Claude Code