rosetta[css]: cross-language consistency — 14 red / 5 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: 3 vs median 13 (-77%)

2. Signal / keyword deviations

• 🔴 cleanup: 0 vs median 2 (-100%)
• 🔴 doc: 0 vs median 1 (-100%)
• 🔴 globals: 0 vs median 2 (-100%)
• 🔴 io: 0 vs median 3 (-100%)
• 🔴 ownership: 0 vs median 1 (-100%)
• 🔴 safety_bypasses: 0 vs median 2 (-100%)
• 🔴 state_mutation: 0 vs median 6 (-100%)
• 🔴 telemetry: 0 vs median 2 (-100%)
• 🟡 branch: 3 vs median 5 (-40%)
• 🟡 keyword_hits: 147 vs median 247.5 (-41%)
• 🟡 safety: 1 vs median 2 (-50%)

3. Risk-exposure consequences (downstream)

• 🔴 risk_api_exposure: 0 vs median 5.31104 (-100%)
• 🔴 risk_cognitive_load: 4 vs median 15.1444 (-74%)
• 🔴 risk_documentation: 26.4404 vs median 57.1208 (-54%)
• 🔴 risk_spec_match: 45.3333 vs median 96 (-53%)
• 🔴 risk_state_flux: 0 vs median 19.9999 (-100%)
• 🟡 risk_safety_score: 19.8485 vs median 37.297 (-47%)
• 🟡 risk_verification: 1.27414 vs median 2.3628 (-46%)

Ledgered shapes touching css

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…

Decoy outcomes (file-level evidence)

• main.css · @media / @supports / @container (code): the three planted branch at-rules are TRIPLE-counted: each is func_start (the CSS at-rules (@layer / @media / @supports) counted as functions; css tree-sitter-accuracy baseline stale (real_functions 24→25) #2507/sqlite tri-comparison validation: class extraction to 100%, both ledger shapes validated #2513 'closest function-shaped construct' precedent), @media/@supports/@container are branch, and @supports is also safety — one @supports line = 3 signals

Reproduce / evidence

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