fix(#2409,#2410): livecode func_start -- Mode D routing + LiveCode Builder handlers

[squid-protocol]

Fixes #2409, fixes #2410.

Problem

livecode extracted exactly one named function from ~900 real handlers across the 98-file
language-crucible corpus.

	reality	before
named functions	~934 handlers	1 (escape)
func_start raw signals	847	847 counted, 0 sliced into FunctionNodes
.lcb handlers	471 declarations (127 real defs)	0 detected

Two root causes, one PR (they're inseparable)

#2410 — no ScopeParsingRegistry entry → Mode B brace-slice fallthrough

LiveCode has no braces (end <name> terminators), so _slice_by_braces produced one
FunctionNode for the whole corpus. Added a mode_d entry:

• Statement-anchored openers (on/command/function/getprop/setprop/handler +
  repeat/if/switch/try/unsafe), bare end closer, function_opener for nested
  handlers. Anchoring is required here (unlike other mode_d configs) because the bare end
  closer also spells end if/end repeat/end handler, and function is a LiveCode operator.
• Anchoring makes next repeat / exit repeat / else if / else fall out for free.
• A dedicated _slice_by_keywords guard handles if COND then <statement> one-liners (no
  end if), analogous to the Ruby/Elixir inline-modifier guard.
• New livecode branch in _extract_semantic_name.

Same routing-only failure class as MATLAB #1266 / yacc #2351.

#2409 — .lcb handler syntax invisible to func_start/args

Added a second alternation arm for LiveCode Builder [public|private] handler <name>(...) —
parenthesized, typed. foreign handler (344 FFI binding declarations — C-prototype shaped,
binds to "<builtin>", no body) and handler type <Name>() (function-pointer typedefs) are
excluded by construction (the design decision on #2409: parallel to a C header prototype not
being a definition).

Verification

Independent-scanner name-diff over the whole corpus (ground truth = every real handler
definition header, from a line scanner sharing no implementation with func_start):

934 real handler defs · 934 GG named functions · 0 false positives · 0 missed.
(.livecodescript ~815, .lcb 121 real handler defs.)

• manual_verification.json: livecode function 1 → 934/934 (PRECISION record for handler
  existence, the chart's staleness anchor). Func Found / Func Precision panels go 0/N → 934/934** + GitGalaxy badge. gg_args_found 2 → 950.
• Differential Scan (crucible_check.py, full ~80-repo corpus): every diff is livecode or
  its downstream ripple. livecode's one monster function splits into ~934 correctly-sized ones,
  so its global mass share drops ~18× and dominant_language corrects livecode → c.
  Zero unrelated-language diffs. Both golden masters re-blessed, re-verified PASS both modes.
• ast_accuracy / ruff / mypy / dead-key: no change.

One limitation, filed separately (#2419)

Exactly 1 of the 934 (addExternalFromFile in the 2183-line
revsaveasandroidstandalone.livecodescript) over-extends its body because _apply_literal_shield
mis-reads /* / */ sitting inside that file's string literals ("Android/*", "*/R.java"),
blanking the handler's own end lines. A narrow shield edge, not a Mode D gap — every other
handler slices cleanly — and it does not affect the 934/934 handler-existence count. Its
own PR.

Tests

tests/core_engine/ + tests/extraction/ full: 6803 passed. New:
test_detector_mode_d_livecode_script_handlers, test_detector_mode_d_livecode_builder_handlers,
.lcb handler cases in test_livecode.py (func_start + args, incl. foreign handler /
handler type exclusions).

tri_comparison_chart.svg / tri_comparison_ledger.json are regenerated by the post-merge
tri-comparison-history.yml job, not here (language_standards.py/detector.py are
tri-comparison-audit.yml trigger paths, but that job only measures js/ts/zig and never
regenerates the chart).

🤖 Generated with Claude Code

https://claude.ai/code/session_019Zm1uVBpVEZJ9SB7bWFR8S

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab