rosetta[sqlite]: cross-language consistency — 8 red / 2 amber metrics, 4 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%)
• 🔴 functions_found: 28 vs median 13 (+115%)

2. Signal / keyword deviations

• 🔴 io: 11 vs median 3 (+267%)
• 🔴 keyword_hits: 116 vs median 247.5 (-53%)
• 🔴 safety_bypasses: 0 vs median 2 (-100%)
• 🟡 state_mutation: 3 vs median 6 (-50%)

3. Risk-exposure consequences (downstream)

• 🔴 risk_cognitive_load: 5.53524 vs median 15.1444 (-63%)
• 🔴 risk_state_flux: 0 vs median 19.9999 (-100%)
• 🔴 risk_tech_debt: 60 vs median 39.9576 (+50%)
• 🟡 risk_spec_match: 64 vs median 96 (-33%)

Ledgered shapes touching sqlite

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
sqlite-dot-read-dual-import-io	keyword-overlap	—	(notes)	^.read counts BOTH import and io (finditer-confirmed) — every corpus chain link adds +1 io in sqlite
orphan-conversion-opaque-counts	upstream-question	#2547	(notes)	On languages without a same-file call graph the orphan census includes non-function shapes, so the orphan->api conversion count diverges from visible function counts: sh…
string-literal-selective-shielding	upstream-question	#2535	b.sql	Keywords inside one string literal ('if eval fails, try open') count toward branch (+2), safety (+1), io (+1) but eval does NOT count toward high_risk_execution (routes …

Decoy outcomes (file-level evidence)

• main.sql · never detaches a database and no shell word (comment): stripped: contributed 0
• b.sql · SELECT 'IF UPDATE FAILS TRY SELECT AGAIN' (string): UPDATE and SELECT inside the literal COUNT (io +2, state +1); IF/TRY inert (sqlite has no bare IF/TRY keywords). io 5 = planted UPDATE + carrier SELECT + .read + 2 from the literal. Ledger: string-literal-selective-shielding
• c.sql · hack_level; (code): inert (underscore + lowercase)

Reproduce / evidence

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