fix(detector): exclude synthetic slicer bucket names from orphan/duplicate census (#2547)

[squid-protocol]

Summary

Closes #2547. Languages sliced by Mode D (_slice_by_keywords: shell/ruby/lua/elixir/livecode/matlab) or Mode E (_slice_by_terminator: sql/sqlite/plpgsql/t-sql/mysql/psql) have no real same-file call graph:

• Mode D bundles any top-level, non-comment code that sits before the file's first real scope into a synthetic satellite literally named "__global_context__" (galaxyscope.py's Contextual Baseline Fix downstream, detector.py ~5063).
• Mode E names every top-level statement generically from its leading keyword ("SELECT_Statement", "CREATE_Statement", ...) rather than any real captured identifier (detector.py ~5235) — sql's func_start/class_start regexes exist for signal-counting purposes only; Mode E's own satellite naming never uses their captured groups.

Neither is a genuine callable function, so the intra-file orphan census's "does this name appear anywhere else in the file" check (detector.py's token_counts[func_name] <= 1) is structurally guaranteed to flag them every time — not a signal about real dead code, just an artifact of how the slicer buckets non-function shapes.

Verified directly against the #1096 control corpus (keyword-rosetta's data/{shell,sqlite}/) via galaxyscope --debug + the recorder DB:

• a.sh/b.sh: 4 "orphans" reported for 3 real functions (the extra was __global_context__).
• main.sql: orphaned_logic = 7 for a file where every satellite is a synthetic Mode-E bucket (zero real callable names at all).

Root cause

The orphan/duplicate loop already tries to exclude known-synthetic bucket names (Unknown_Sat/Anonymous_Block/Main/Declarative_Block), but the set was incomplete:

1. Missing __global_context__ entirely.
2. Exact-match only — never stripped the _[Truncated]/_[Unterminated] suffixes the slicer appends to these same names elsewhere (detector.py ~5039, ~5043, ~5277).
3. Never covered Mode E's <KEYWORD>_Statement shape at all.
4. Only guarded the orphan branch — the duplicate-detection branch had no exclusion, so the same synthetic names could also be mis-flagged as "duplicate logic".

Fix

Replace the hardcoded exact-match set with _is_synthetic_satellite_name() (detector.py, module-level, right before StructuralExtractor): strips the known suffixes, checks the full synthetic-name family, and additionally recognizes Mode E's <KEYWORD>_Statement pattern (always synthesized, never a real identifier — see the "sql" in lang_key branch at detector.py ~5235). Gates both the orphan and duplicate branches of the classification loop.

Verification

• New regression test test_detector_orphan_census_excludes_synthetic_slicer_names (tests/core_engine/test_detector.py), covering both Mode D (__global_context__) and Mode E (<KEYWORD>_Statement).
• Full suite: 7168 passed, 2 skipped, 9 xfailed, 3 xpassed (all pre-existing/unrelated).
• tests/tools/audit_check.py: clean, no new ruff/mypy/dead-key/ast-accuracy findings.
• Re-ran against the real corpus post-fix: a.sh/b.sh api 4→3, main.sh orphans 2→1, main.sql orphans 7→0, a/b/c.sql's bogus api 2/4/3→0/0/0 (SQL statements were never real API surface — Mode E never captures a real identifier for any SQL statement type, so the correct count is zero, not one per CREATE INDEX as the issue's own reporter — me — initially assumed before this investigation).
• tests/tools/crucible_check.py: both full_precision and zero_dependency pass clean after fixture regen.
• Golden master regenerated (tests/tools/update_golden_master.py, both modes) — ~2200 diffs, all in tech_debt/api_exposure/documentation exposure ratios for shell/sql/lua/livecode/ruby/matlab corpus repos, exactly the intentional effect of no longer inflating those files' orphaned_logic/api with non-function shapes.

Notes

• #2536 (raw pre-adjustment hit_vector recording) is a separate, broader auditability enhancement — not superseded by this fix, left open.
• Dockerfile's Mode A RUN/CMD/HEALTHCHECK satellite-name reuse is a related-but-distinct mechanism (real captured keywords, not synthetic labels); the live corpus shows zero actual orphaned_logic inflation for the dockerfile control files today, so it's out of scope here (no observed symptom).
• keyword-rosetta's deviation_ledger.json entry (orphan-conversion-opaque-counts) will need its own re-validation sweep in that repo after this merges — not part of this PR.

🤖 Generated with Claude Code

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab