rosetta[jcl]: cross-language consistency — 13 red / 7 amber metrics, 2 ledgered shapes

[squid-protocol]

Part of #2560 — per-language tracking for the keyword-rosetta control corpus (built for #1096). Planted intent is identical in every language, so every deviation below is measured bias vs. the 46-language median, validated in the corpus's deviation ledger. 🔴 = beyond ±50% of the cross-language median, 🟡 = ±25–50%.

Triage order (per the epic): structure detection first, then signal/keyword rules, then risk-score consequences — §3's numbers are downstream of §1–§2 and re-baseline as those fix. Cross-cutting engine issues (e.g. #2535/#2546) fix many languages at once; this issue tracks this language's residual after each cross-cutting fix lands, and closes when the language sits inside the ±25% band on every comparable metric (or every remaining deviation is ledgered as intended morphology).

1. Structure detection (fix first)

• 🔴 args: 1 vs median 13 (-92%)
• 🟡 dependency_links: 4 vs median 3 (+33%)

2. Signal / keyword deviations

• 🔴 cleanup: 0 vs median 2 (-100%)
• 🔴 comment_lines: 5 vs median 30 (-83%)
• 🔴 doc: 0 vs median 1 (-100%)
• 🔴 fragile_debt: 0 vs median 1 (-100%)
• 🔴 globals: 0 vs median 2 (-100%)
• 🔴 planned_debt: 0 vs median 1 (-100%)
• 🔴 safety: 0 vs median 2 (-100%)
• 🔴 safety_bypasses: 0 vs median 2 (-100%)
• 🔴 state_mutation: 2 vs median 6 (-67%)
• 🔴 telemetry: 0 vs median 2 (-100%)
• 🔴 test: 0 vs median 2 (-100%)
• 🟡 branch: 3 vs median 5 (-40%)
• 🟡 keyword_hits: 158 vs median 247.5 (-36%)

3. Risk-exposure consequences (downstream)

• 🔴 risk_cognitive_load: 4 vs median 15.1444 (-74%)
• 🟡 risk_api_exposure: 6.93756 vs median 5.31104 (+31%)
• 🟡 risk_spec_match: 64 vs median 96 (-33%)
• 🟡 risk_tech_debt: 20 vs median 39.9576 (-50%)
• 🟡 risk_verification: 1.70246 vs median 2.3628 (-28%)

Ledgered shapes touching jcl

defect	type	issue	evidence	summary
census-requires-git-tracked	engine-semantic	—	—	GalaxyScope's census enumerates git-tracked files only; an untracked folder scans as '0 files mapped' with no per-file warning
batch5-tier2-morphology-shapes	keyword-overlap	—	—	Collective entry for Tier-2 finditer/report-confirmed shapes, detailed per-manifest: css at-rules triple-count (func+branch, @supports also safety; args tracks @import u…

Reproduce / evidence

• Corpus folder: data/jcl/ — expected_signals.json carries the shape-by-shape accounting
• Gate: GALAXYSCOPE_BIN=<venv>/bin/galaxyscope python tools/verify_language.py jcl in the corpus repo
• Full context: findings_by_language.md#jcl · bias_report.md · bias chart

[squid-protocol]

Status 2026-08-31: 13🔴/7🟡 → 6🔴/8🟡

Root-caused per the epic's triage order. Three distinct causes, three deliverables:

1. Real engine gap (filed #2610, fixed in PR #2611): prism never extracted JCL //* comments — jcl's line_exclusive delimiter list is deliberately empty (// prefixes every statement, so the stateless stripper can't express the whole-line positional form), and no positional path existed either. Every //* line stayed in the code stream: doc_loc 0 engine-wide, comment_analysis (doc/ownership/debt) structurally dead for JCL, ownership counting only by accident on the wrong surface. PR #2611 adds a line-count-preserving whole-line stripper (with a JES3 //*MAIN-verb guard) plus the signal rules JCL has genuine morphology for: safety = COND= RC-tests, safety_bypasses = COND=EVEN/ONLY, telemetry = MSGLEVEL=/MSGCLASS=, and the shared comment-anchored debt patterns.

2. Corpus authoring gap (keyword-rosetta PR #4, draft until #2611 merges): args: 1 was not morphology — JCL expresses per-step arguments via PARM= and the engine rule already handled it (#2482); the shell just under-planted. Every EXEC step now carries PARM= → args 13, exactly at median, clearing the §1 structure red. Plants added for the new rules: safety 2, bypasses 2, telemetry 2, debt 1/1 — all land exactly (report-confirmed, strict gate PASS 76 assertions).

3. Ledgered intended morphology (jcl-2610-rebaseline-residual-morphology, validated): cleanup deliberately has no rule (DISP=(...,DELETE) would double-count io's DISP= — recorded decision in #2610); doc/test/globals have no JCL concept to plant; dependency_links 4 = the DD DSN= capture (deliberate blast-radius design); comment_lines residual = the no-blank-lines proxy artifact; branch 3 / state_mutation 2 match planted intent exactly — their deviation is median inflation by the #2545 return-in-branch family and #2546 flux weighting, and re-baselines when those cross-cutting fixes land.

Remaining after both PRs merge

• 🔴 morphology zeros (cleanup/doc/test/globals) — ledgered; final disposition is the epic's scoring-side "absent morphology = incomparable" work
• 🔴 state_mutation, 🟡 branch — blocked on state-flux ×3 branch-context weighting: undocumented, and phantom-triggered by branch keywords inside string literals #2546 / kotlin: branch rule counts 'return' — inconsistent with java/swift/c siblings #2545
• 🔴/🟡 §3 risk metrics (cognitive_load −74% etc.) — downstream shadows of the above, re-baseline per the epic; not tuned here

Everything in §1–§2 that is fixable at the jcl level is now fixed or ledgered.

[squid-protocol]

Closing: every remaining deviation is ledgered, fixed, or owned by a named cross-cutting issue

Final state after the full sweep (engine PRs #2611/#2618, corpus PRs keyword-rosetta#4/#6/#8):

13🔴 / 7🟡 at open → 2🔴 / 8🟡, with 4 ledgered n/a

• Fixed (engine, prism: JCL //* comments are never stripped — comment surface structurally dead for jcl (comment_lines ~0, debt/doc rules can never fire) #2610 → PR fix(jcl): //* comments reach the comment surface; COND=/MSGLEVEL/debt rules wired (#2610) #2611): the //* comment surface (prism gap, engine-wide for all JCL), plus new rules with genuine JCL morphology: safety=COND=, safety_bypasses=COND=EVEN/ONLY, telemetry=MSGLEVEL=/MSGCLASS=, shared debt patterns.
• Fixed (corpus authoring): args 1→13 via PARM= per EXEC step — was mislabeled morphology; the engine rule always worked.
• n/a (ledgered morphology, jcl-2610-rebaseline-residual-morphology): cleanup (deliberate — DISP= already feeds io), doc/test/globals (no JCL concept). Rendered incomparable, not −100%, since the n/a semantics landed (keyword-rosetta#6).
• Remaining 🔴, not jcl's fault: state_mutation (median inflated by state-flux ×3 branch-context weighting: undocumented, and phantom-triggered by branch keywords inside string literals #2546 ×3 flux weighting — jcl matches planted intent exactly) and risk_cognitive_load (downstream shadow; re-baselines per the epic's rule against tuning risk formulas on biased inputs). branch 🟡 likewise owned by kotlin: branch rule counts 'return' — inconsistent with java/swift/c siblings #2545.
• Capstone: docs/language_status/jcl.md §10 carries the full five-bucket reasoning; the reusable workflow is the keyword-rosetta rosetta-language-sweep skill; ongoing enforcement is the control corpus Phase 5: pin keyword-rosetta as a cross-language consistency gate in gitgalaxy CI #2557 rosetta-audit gate (live as of today, pin e92022d).

Residuals tracked by #2545 / #2546 / epic #2560's scoring-side item — nothing left that is jcl-specific.