feat(livecode): tri-comparison manual verification -- class precision badge + 3 engine fixes

[squid-protocol]

What

tri-comparison-ledger-sweep skill, manual-verification fallback for livecode
(a gg_only language -- no tree-sitter grammar, no ctags parser, so the
tri-comparison tool has nothing to reconcile against). Verified by direct source
cross-check against the whole ../language-crucible/data/livecode corpus (99
code files, 97 ingested).

Result: livecode earns a Class Precision G badge

Signature	Before	After	Verdict
class_start named	0	57	57/57 precision, 0 false positives -- badge
args file signal	0	442	signal alive (per-function count blocked on #2410)
func_start raw signal	781	781	correct; named extraction blocked on #2410

All 57 detected classes (24 .livecodescript script "Name" + 33 .lcb
module com.x.y) checked against their declaration lines. map.lcb correctly
not counted (its module line is commented out).

Engine fixes (3)

1. args regex missing re.M (language_standards.py) -- ^-anchored per
   line exactly like its func_start sibling, but compiled re.I only, so ^
   matched only the file start -> 0 args matches corpus-wide (every file
   has a license header before its first handler). The pattern was already
   correct on a single line, so every prior single-line test missed it.
2. class_start bareword-only capture (language_standards.py) -- added a
   real quoted-name branch ("([^"\r\n]{1,200})") for .livecodescript
script "Name" object exports (~63 files, previously 0 detected). Two capture
   groups now; _resolve_class_start_match already handles the
   group-1-or-group-2 shape (fortran/lua/abap precedent).
3. livecode missing from _CLASS_START_NAMED_EXTRACTION_LANGS
   (detector.py) -- the generic fallback regex is lowercase
   class|struct|interface|trait|enum, so the named class list stayed empty even
   with the regex fixed. Added.

New regression tests: test_livecode_args_multiline_anchor_regression (catches
#1), quoted-name assertions in the class_start dotted-name test.

Filed, deliberately not fixed here

• livecode func_start/args miss LiveCode Builder (.lcb) 'handler' syntax entirely (0 of 471 handlers detected) #2409 -- func_start/args miss LiveCode Builder handler syntax
  entirely (0 of 471 .lcb handlers detected). Needs a dedicated
  parenthesized-typed-param branch, not just a keyword add.
• livecode: no ScopeParsingRegistry entry -> Mode B brace-slice fallthrough drops 780/781 named functions #2410 -- no ScopeParsingRegistry entry -> Mode B brace-slice fallthrough
  produces 1 FunctionNode for the whole corpus. Needs its own
  inline-if...then / else / next repeat depth guards (design work), so
  filed rather than fixed inline.
• Leading UTF-8 BOM never stripped -> defeats ^-anchored line-1 extraction (58 corpus files; 38 livecode classes missed) #2411 -- a leading UTF-8 BOM is never stripped, defeating ^-anchored
  line-1 extraction; 38 more real livecode classes missed. Engine-wide gap (58
  BOM files across 6 languages), not a livecode class_start defect.

Why the golden masters churn so much

crucible_check.py against the full ~80-repo corpus: all ~290 diffs are
livecode-only -- Function Parameters 0 -> real, Class/Entity Declarations
0 -> 1, and the topological-coordinate / structural-magnitude ripple that
recomputes from the corrected counts. Zero diffs in any other language.
livecode's global mass share rises ~6x purely because a whole language's args
signal went from artificially-zero to real (revliburl.livecodescript alone:
0 -> 129 function-parameter signals). Both golden master fixtures re-blessed;
update_golden_master.py --yes re-run confirms "No drift" for both modes.

The tri-comparison ledger / points-of-interest regen produced only
last_reconciled_at timestamp churn (no last_seen_count / still_reproduces
changes, and livecode has no ledger entries) -- left for the
tri-comparison-history.yml companion workflow to pick up post-merge, per
tri_comparison_README.md. tri_comparison_chart.svg is committed (shows the
earned badge), matching the jcl manual-verification precedent (#2203).

Verification

livecode extraction gauntlet + strict (113 passed) · core-engine suite (674
passed) · audit_check.py (ruff / mypy / dead-key / ast-accuracy clean) ·
crucible_check.py full corpus (livecode-only diffs) · both golden masters
re-blessed · tri_comparison_chart.py --all --write (livecode Class Precision
-> 57/57** + G badge).

🤖 Generated with Claude Code

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab