From 4788c4dda7b0ef8273a5dfc4f551dcb2b4290069 Mon Sep 17 00:00:00 2001 From: squid-protocol Date: Sat, 29 Aug 2026 17:45:17 -0400 Subject: [PATCH] test(cpp): C++ special-member name reconciliation -> func precision 100% `tree_sitter_accuracy_audit` measured cpp function precision at 95.0% (68 "extra" functions). Every one is a real C++ special member that GitGalaxy correctly extracts and tree-sitter-cpp either names differently or drops -- zero genuine GitGalaxy false positives. Three shared-reader fixes (used by both the accuracy audit and tri_comparison_gatherer): 1. Conversion-operator naming. tree-sitter's `operator_cast` / `qualified_identifier` node text bakes in the parameter list and cv-qualifiers (`Variant::operator String() const`) and, for a template conversion, the argument list (`operator Vector() const`). GitGalaxy's `func_start` stores just `Variant::operator String` / `operator Vector`. Added `_cpp_canonical_operator_name` -- strips the trailing `(...)` + cv-quals and `<...>` args, keeps the `operator ` keyword and `Ns::` qualifier. Guarded to CONVERSION operators only (`operator ` + a type token); symbol operators (`operator()`, `operator==`, `operator[]`) are left untouched. 2. `_FORCE_INLINE_`-macro parse errors. A function-like macro with no visible definition immediately before a special member desyncs tree-sitter-cpp: it drops the `operator` keyword or the `~` into a sibling ERROR node and names the member by its bare type (`_FORCE_INLINE_ operator T() const` -> "T"; `_FORCE_INLINE_ ~Variant()` -> "Variant"). Recover the prefix from the ERROR node so the readers agree instead of double-counting. 3. Blind-spot promotion. Added cpp to `_find_blind_spot_ranges` (small tree-sitter-cpp ERROR nodes, padded +/-2 lines for the `template <...>` header line GitGalaxy anchors on), so GitGalaxy's correct extraction of a constructor tree-sitter's error recovery loses (`_FORCE_INLINE_ RequiredResult(...) : ... {}`) is promoted to ground truth via the existing #1849 Phase 2 mechanism. Result: cpp function precision 95.0% -> 100.0% (extra_functions 68 -> 0), recall 87.0% -> 89.3% (the operator matches now count as found), args exact-match 99.8% -> 100.0%. Baseline + summary table regenerated. `tree_sitter_accuracy_audit --all --ci` 31/31 OK; C unchanged. Tri-comparison: the shared `_get_node_name` fix also lands in `tri_comparison_gatherer`, moving ~55 conversion-operator occurrences from `cpp/function/existence/agree[tree_sitter]_vs[ctags,gitgalaxy]` (200 -> 136) and `agree[ctags,gitgalaxy]_vs[tree_sitter]` (66 -> 14) into GitGalaxy+tree-sitter consensus. Verdicts, chart, points-of-interest regenerated; every reproducing ledger shape stays validated. No engine change -- `gitgalaxy/core/*` untouched, no golden-master re-bless. Co-Authored-By: Claude Sonnet 5 Claude-Session: https://claude.ai/code/session_019Zm1uVBpVEZJ9SB7bWFR8S --- docs/self_scan/tri_comparison_chart.svg | 14 +- docs/self_scan/tri_comparison_ledger.json | 475 +++++++++--------- .../tri_comparison_points_of_interest.md | 279 +++++----- gitgalaxy/standards/language_standards.py | 2 +- tests/tools/tree_sitter_accuracy_audit.py | 77 ++- tests/tree_sitter_accuracy_baseline_cpp.json | 10 +- 6 files changed, 454 insertions(+), 403 deletions(-) diff --git a/docs/self_scan/tri_comparison_chart.svg b/docs/self_scan/tri_comparison_chart.svg index a39a4e2dc..9458541c4 100644 --- a/docs/self_scan/tri_comparison_chart.svg +++ b/docs/self_scan/tri_comparison_chart.svg @@ -196,10 +196,10 @@ 1378 - -1357/1369 - -1291/1491 + +1369/1369 + +1355/1491 1348/1378 @@ -220,11 +220,11 @@ 63/65 -1281 +1333 -1281 +1333 -1281 +1333 csharp diff --git a/docs/self_scan/tri_comparison_ledger.json b/docs/self_scan/tri_comparison_ledger.json index 35b30bb31..dac0cddf0 100644 --- a/docs/self_scan/tri_comparison_ledger.json +++ b/docs/self_scan/tri_comparison_ledger.json @@ -13,7 +13,7 @@ "investigated_at": "2026-08-20T00:00:00Z", "investigated_by": "Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep", "language": "agc_assembly", - "last_reconciled_at": "2026-08-29T21:16:28Z", + "last_reconciled_at": "2026-08-29T21:42:55Z", "last_seen_count": 215, "last_seen_examples": [ { @@ -118,7 +118,7 @@ "investigated_at": "2026-08-20T00:00:00Z", "investigated_by": "Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep", "language": "agc_assembly", - "last_reconciled_at": "2026-08-29T21:16:28Z", + "last_reconciled_at": "2026-08-29T21:42:55Z", "last_seen_count": 37, "last_seen_examples": [ { @@ -221,7 +221,7 @@ "investigated_at": "2026-08-20T22:17:39Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep (direct investigation, no Gemini dispatch needed -- root cause was immediately clear from source)", "language": "apex", - "last_reconciled_at": "2026-08-29T21:16:30Z", + "last_reconciled_at": "2026-08-29T21:42:56Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -260,7 +260,7 @@ "investigated_at": "2026-08-27T00:00:00Z", "investigated_by": "Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep", "language": "assembly", - "last_reconciled_at": "2026-08-29T21:16:30Z", + "last_reconciled_at": "2026-08-29T21:42:58Z", "last_seen_count": 26, "last_seen_examples": [ { @@ -365,7 +365,7 @@ "investigated_at": "2026-08-27T00:00:00Z", "investigated_by": "Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep", "language": "assembly", - "last_reconciled_at": "2026-08-29T21:16:30Z", + "last_reconciled_at": "2026-08-29T21:42:58Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -405,7 +405,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Claude Sonnet 5 (resolved + fixed directly, no dispatch needed)", "language": "c", - "last_reconciled_at": "2026-08-29T21:16:36Z", + "last_reconciled_at": "2026-08-29T21:43:01Z", "last_seen_count": 23, "last_seen_examples": [ { @@ -519,7 +519,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Gemini (dispatched via tri-comparison-ledger-sweep), confirmed by Claude Sonnet 5", "language": "c", - "last_reconciled_at": "2026-08-29T21:16:36Z", + "last_reconciled_at": "2026-08-29T21:43:01Z", "last_seen_count": 9, "last_seen_examples": [ { @@ -624,7 +624,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Gemini (dispatched via tri-comparison-ledger-sweep), confirmed by Claude Sonnet 5", "language": "c", - "last_reconciled_at": "2026-08-29T21:16:36Z", + "last_reconciled_at": "2026-08-29T21:43:01Z", "last_seen_count": 525, "last_seen_examples": [ { @@ -738,7 +738,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Claude Sonnet 5 (resolved directly, no dispatch needed) -- corrected after cross-referencing #1837", "language": "c", - "last_reconciled_at": "2026-08-29T21:16:36Z", + "last_reconciled_at": "2026-08-29T21:43:01Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -771,7 +771,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Claude Sonnet 5 (resolved + fixed directly, no dispatch needed)", "language": "c", - "last_reconciled_at": "2026-08-29T21:16:36Z", + "last_reconciled_at": "2026-08-29T21:43:01Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -804,7 +804,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Gemini (dispatched via tri-comparison-ledger-sweep), confirmed by Claude Sonnet 5", "language": "c", - "last_reconciled_at": "2026-08-29T21:16:36Z", + "last_reconciled_at": "2026-08-29T21:43:01Z", "last_seen_count": 13, "last_seen_examples": [ { @@ -921,7 +921,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Claude Sonnet 5 (resolved directly, no dispatch needed)", "language": "c", - "last_reconciled_at": "2026-08-29T21:16:36Z", + "last_reconciled_at": "2026-08-29T21:43:01Z", "last_seen_count": 3, "last_seen_examples": [ { @@ -972,7 +972,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Claude Sonnet 5 (resolved + fixed directly, no dispatch needed)", "language": "c", - "last_reconciled_at": "2026-08-29T21:16:36Z", + "last_reconciled_at": "2026-08-29T21:43:01Z", "last_seen_count": 7, "last_seen_examples": [ { @@ -1059,7 +1059,7 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "c", - "last_reconciled_at": "2026-08-29T21:16:36Z", + "last_reconciled_at": "2026-08-29T21:43:01Z", "last_seen_count": 3, "last_seen_examples": [ { @@ -1112,7 +1112,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Claude Sonnet 5 (resolved directly, no dispatch needed)", "language": "c", - "last_reconciled_at": "2026-08-29T21:16:36Z", + "last_reconciled_at": "2026-08-29T21:43:01Z", "last_seen_count": 4, "last_seen_examples": [ { @@ -1171,7 +1171,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "c", - "last_reconciled_at": "2026-08-29T21:16:36Z", + "last_reconciled_at": "2026-08-29T21:43:01Z", "last_seen_count": 17, "last_seen_examples": [ { @@ -1275,7 +1275,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Gemini (dispatched via tri-comparison-ledger-sweep), confirmed by Claude Sonnet 5", "language": "c", - "last_reconciled_at": "2026-08-29T21:16:36Z", + "last_reconciled_at": "2026-08-29T21:43:01Z", "last_seen_count": 74, "last_seen_examples": [ { @@ -1388,7 +1388,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "c", - "last_reconciled_at": "2026-08-29T21:16:36Z", + "last_reconciled_at": "2026-08-29T21:43:01Z", "last_seen_count": 88, "last_seen_examples": [ { @@ -1493,7 +1493,7 @@ "investigated_at": "2026-08-19", "investigated_by": "gemini-3.1-pro-high (agy), dispatched via tri-comparison-ledger-sweep, self-corrected and reviewed by claude-sonnet-5", "language": "cobol", - "last_reconciled_at": "2026-08-29T21:16:48Z", + "last_reconciled_at": "2026-08-29T21:43:07Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -1534,7 +1534,7 @@ "investigated_at": "2026-08-28", "investigated_by": "Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep", "language": "cobol", - "last_reconciled_at": "2026-08-29T21:16:48Z", + "last_reconciled_at": "2026-08-29T21:43:07Z", "last_seen_count": 6, "last_seen_examples": [ { @@ -1605,7 +1605,7 @@ "investigated_at": "2026-08-28", "investigated_by": "Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep", "language": "cobol", - "last_reconciled_at": "2026-08-29T21:16:48Z", + "last_reconciled_at": "2026-08-29T21:43:07Z", "last_seen_count": 1316, "last_seen_examples": [ { @@ -1708,7 +1708,7 @@ "investigated_at": "2026-08-19", "investigated_by": "gemini-3.1-pro-high (agy), dispatched via tri-comparison-ledger-sweep, reviewed by claude-sonnet-5", "language": "cobol", - "last_reconciled_at": "2026-08-29T21:16:48Z", + "last_reconciled_at": "2026-08-29T21:43:07Z", "last_seen_count": 45, "last_seen_examples": [ { @@ -1812,7 +1812,7 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "cpp", - "last_reconciled_at": "2026-08-29T21:16:53Z", + "last_reconciled_at": "2026-08-29T21:43:10Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -1854,7 +1854,7 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "cpp", - "last_reconciled_at": "2026-08-29T21:16:53Z", + "last_reconciled_at": "2026-08-29T21:43:10Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -1896,7 +1896,7 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "cpp", - "last_reconciled_at": "2026-08-29T21:16:53Z", + "last_reconciled_at": "2026-08-29T21:43:10Z", "last_seen_count": 84, "last_seen_examples": [ { @@ -2010,7 +2010,7 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "cpp", - "last_reconciled_at": "2026-08-29T21:16:53Z", + "last_reconciled_at": "2026-08-29T21:43:10Z", "last_seen_count": 22, "last_seen_examples": [ { @@ -2124,7 +2124,7 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "cpp", - "last_reconciled_at": "2026-08-29T21:16:53Z", + "last_reconciled_at": "2026-08-29T21:43:10Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -2166,7 +2166,7 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "cpp", - "last_reconciled_at": "2026-08-29T21:16:53Z", + "last_reconciled_at": "2026-08-29T21:43:10Z", "last_seen_count": 61, "last_seen_examples": [ { @@ -2280,7 +2280,7 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "cpp", - "last_reconciled_at": "2026-08-29T21:16:53Z", + "last_reconciled_at": "2026-08-29T21:43:10Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -2312,7 +2312,7 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "cpp", - "last_reconciled_at": "2026-08-29T21:16:53Z", + "last_reconciled_at": "2026-08-29T21:43:10Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -2343,7 +2343,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "cpp", - "last_reconciled_at": "2026-08-29T21:16:53Z", + "last_reconciled_at": "2026-08-29T21:43:10Z", "last_seen_count": 76, "last_seen_examples": [ { @@ -2447,8 +2447,8 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "cpp", - "last_reconciled_at": "2026-08-29T21:16:53Z", - "last_seen_count": 66, + "last_reconciled_at": "2026-08-29T21:43:10Z", + "last_seen_count": 14, "last_seen_examples": [ { "file_path": "godot/gdscript_vm.cpp", @@ -2486,15 +2486,6 @@ "tree_sitter": null } }, - { - "file_path": "godot/object.cpp", - "name": "Object::Connection::operator Variant", - "readings": { - "ctags": 108, - "gitgalaxy": 108, - "tree_sitter": null - } - }, { "file_path": "godot/object.h", "name": "operator=", @@ -2539,13 +2530,22 @@ "gitgalaxy": 1003, "tree_sitter": null } + }, + { + "file_path": "godot/object.h", + "name": "RequiredResult", + "readings": { + "ctags": 1013, + "gitgalaxy": 1012, + "tree_sitter": null + } } ], "metric": "existence", "status": "validated", "still_reproduces": true, "symbol_type": "function", - "verdict": "Confirmed real tree-sitter-cpp grammar limitation, not a GitGalaxy or ctags defect. Every sampled case (GDScriptFunction::call, Main::setup, Main::setup2, Main::start, Object::Connection::operator Variant) is a large, complex function -- GDScriptFunction::call in particular (godot/gdscript_vm.cpp:499) is a bytecode interpreter's main dispatch loop using GNU 'labels as values' computed-goto syntax (`&&OPCODE_LABEL`) via the same OPCODES_TABLE/OPCODE macro family documented in the sibling agree[gitgalaxy,tree_sitter]_vs[ctags] entry -- a non-standard GNU extension tree-sitter-cpp's grammar does not support, which plausibly causes a parse error cascade that loses the enclosing function_definition node entirely rather than just misreading the body. ctags and GitGalaxy both correctly find and name these functions regardless of body content, since neither one needs to fully parse the function body to recognize its signature. tree-sitter's non-detection is a confirmed limitation in tree-sitter itself -- but no credit_tools adjustment applies: ctags and GitGalaxy are already a 2-of-3 AGREEING PAIR on this shape (agreeing_tools has 2 members), which already satisfies reconcile_symbols' own `len(present) >= 2` precision-credit condition naturally with no ledger adjustment needed. credit_tools exists for a LONE, single-tool claim (agreeing_tools with exactly 1 member) the base algorithm can't otherwise corroborate -- applying it to an already-mutually-corroborating pair would double-count (confirmed: this exact mistake briefly pushed ctags' precision past 100% before being caught and reverted in this same session)." + "verdict": "Confirmed real tree-sitter-cpp grammar limitation, not a GitGalaxy or ctags defect. Every sampled case (GDScriptFunction::call, Main::setup, Main::setup2, Main::start, Object::Connection::operator Variant) is a large, complex function -- GDScriptFunction::call in particular (godot/gdscript_vm.cpp:499) is a bytecode interpreter's main dispatch loop using GNU 'labels as values' computed-goto syntax (`&&OPCODE_LABEL`) via the same OPCODES_TABLE/OPCODE macro family documented in the sibling agree[gitgalaxy,tree_sitter]_vs[ctags] entry -- a non-standard GNU extension tree-sitter-cpp's grammar does not support, which plausibly causes a parse error cascade that loses the enclosing function_definition node entirely rather than just misreading the body. ctags and GitGalaxy both correctly find and name these functions regardless of body content, since neither one needs to fully parse the function body to recognize its signature. tree-sitter's non-detection is a confirmed limitation in tree-sitter itself -- but no credit_tools adjustment applies: ctags and GitGalaxy are already a 2-of-3 AGREEING PAIR on this shape (agreeing_tools has 2 members), which already satisfies reconcile_symbols' own `len(present) >= 2` precision-credit condition naturally with no ledger adjustment needed. credit_tools exists for a LONE, single-tool claim (agreeing_tools with exactly 1 member) the base algorithm can't otherwise corroborate -- applying it to an already-mutually-corroborating pair would double-count (confirmed: this exact mistake briefly pushed ctags' precision past 100% before being caught and reverted in this same session). 2026-08-29 follow-up (#2455): the tree_sitter_accuracy_audit / tri_comparison_gatherer shared tree-sitter reader now canonicalizes C++ conversion-operator and destructor names to GitGalaxy's own convention -- `Variant::operator String() const` -> `Variant::operator String`, `operator BitField() const` -> `operator BitField`, and the `_FORCE_INLINE_`-macro-mangled `~Variant` / `operator T` forms recovered from the sibling ERROR node -- and treats a small tree-sitter-cpp ERROR node (a function-like macro with no visible definition immediately before a special member) as a blind-spot region, promoting GitGalaxy's correct extraction there to ground truth. cpp function precision 95.0% -> 100.0%, extra_functions 68 -> 0. `Object::Connection::operator Variant` (doubly-qualified conversion operator) is now in consensus. The residual (`GDScriptFunction::call`, `Main::setup*`) is the computed-goto / OPCODES_TABLE parse cascade that loses the whole `function_definition` node -- GitGalaxy and ctags both still correct, count ~66 -> ~14." }, "cpp/function/existence/agree[ctags,tree_sitter]_vs[gitgalaxy]": { "agreeing_tools": [ @@ -2561,7 +2561,7 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "cpp", - "last_reconciled_at": "2026-08-29T21:16:53Z", + "last_reconciled_at": "2026-08-29T21:43:10Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -2594,7 +2594,7 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "cpp", - "last_reconciled_at": "2026-08-29T21:16:53Z", + "last_reconciled_at": "2026-08-29T21:43:10Z", "last_seen_count": 30, "last_seen_examples": [ { @@ -2708,9 +2708,18 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "cpp", - "last_reconciled_at": "2026-08-29T21:16:53Z", - "last_seen_count": 9, + "last_reconciled_at": "2026-08-29T21:43:10Z", + "last_seen_count": 21, "last_seen_examples": [ + { + "file_path": "godot/object.h", + "name": "operator Ref", + "readings": { + "ctags": null, + "gitgalaxy": 1035, + "tree_sitter": 1036 + } + }, { "file_path": "godot/rendering_server_default.h", "name": "mesh_create_from_surfaces", @@ -2757,39 +2766,39 @@ } }, { - "file_path": "powertoys/FancyZones.cpp", - "name": "FancyZones::OnKeyDown", + "file_path": "godot/variant.cpp", + "name": "Variant::operator ::RID", "readings": { "ctags": null, - "gitgalaxy": 473, - "tree_sitter": 473 + "gitgalaxy": 2002, + "tree_sitter": 2002 } }, { - "file_path": "powertoys/FancyZones.cpp", - "name": "FancyZones::Run", + "file_path": "godot/variant.cpp", + "name": "Variant::operator Vector", "readings": { "ctags": null, - "gitgalaxy": 214, - "tree_sitter": 214 + "gitgalaxy": 2219, + "tree_sitter": 2219 } }, { - "file_path": "powertoys/FancyZones.cpp", - "name": "FancyZones::Destroy", + "file_path": "godot/variant.cpp", + "name": "Variant::operator Vector", "readings": { "ctags": null, - "gitgalaxy": 286, - "tree_sitter": 286 + "gitgalaxy": 2229, + "tree_sitter": 2229 } }, { - "file_path": "powertoys/FancyZones.cpp", - "name": "FancyZones::VirtualDesktopChanged", + "file_path": "godot/variant.cpp", + "name": "Variant::operator Vector", "readings": { "ctags": null, - "gitgalaxy": 303, - "tree_sitter": 303 + "gitgalaxy": 2247, + "tree_sitter": 2247 } } ], @@ -2797,7 +2806,7 @@ "status": "validated", "still_reproduces": true, "symbol_type": "function", - "verdict": "UPDATED after a real production fix. The OPCODE-family shared mistake documented in this shape's prior verdict (~96 of the original 105 occurrences) is now FIXED: GitGalaxy's func_start (and by the same mechanism, C's) now scans each file for `#define NAME(...)` function-like macro definitions and excludes any match whose captured name is a known macro -- the exact same fact universal-ctags itself already used (confirmed via a direct `ctags -f -` run: ctags tags `OPCODE` only once, at its own #define line, kind `d`, and produces zero tags at any invocation site -- it isn't smarter about the invocation's shape, it just already knows the name is a macro and never re-tags it). This also fixed the already-documented C `RICHCMP_WRAPPER`/`SLOT0`/`SLOT1`/`SLOT1BINFULL`/`DICT___REVERSED___METHODDEF` false positives as a bonus (same mechanism, `lang_id in (\"c\", \"cpp\")`). Verified via 3 repeated full-corpus scans producing byte-identical function lists, the full extraction gauntlet, and both golden masters re-blessed. The remaining residual (9 occurrences, previously miscounted as more of the same shared mistake in the earlier verdict's blanket debit -- corrected here) is a DIFFERENT, unrelated finding: GitGalaxy and tree-sitter are CORRECT here, ctags is wrong. Two sub-patterns confirmed: (1) the already-documented macro-as-return-type-prefix pattern (`IFACEMETHODIMP_(void)` immediately before the real `FancyZones::Run() noexcept` -- ctags tags the macro invocation itself and loses the real name, powertoys/FancyZones.cpp, 4 occurrences); (2) a newly-confirmed, NOT yet root-caused ctags miss on an ordinary virtual method with no macro involvement at all (`virtual RID mesh_create_from_surfaces(const Vector &p_surfaces, int p_blend_shape_count = 0) override { ... }`, godot/rendering_server_default.h -- ctags produces zero tags anywhere near this real method, 5 occurrences). No credit_tools adjustment applies: GitGalaxy and tree-sitter are already a 2-of-3 agreeing pair here, which already satisfies reconcile_symbols' own natural precision-credit condition." + "verdict": "UPDATED after a real production fix. The OPCODE-family shared mistake documented in this shape's prior verdict (~96 of the original 105 occurrences) is now FIXED: GitGalaxy's func_start (and by the same mechanism, C's) now scans each file for `#define NAME(...)` function-like macro definitions and excludes any match whose captured name is a known macro -- the exact same fact universal-ctags itself already used (confirmed via a direct `ctags -f -` run: ctags tags `OPCODE` only once, at its own #define line, kind `d`, and produces zero tags at any invocation site -- it isn't smarter about the invocation's shape, it just already knows the name is a macro and never re-tags it). This also fixed the already-documented C `RICHCMP_WRAPPER`/`SLOT0`/`SLOT1`/`SLOT1BINFULL`/`DICT___REVERSED___METHODDEF` false positives as a bonus (same mechanism, `lang_id in (\"c\", \"cpp\")`). Verified via 3 repeated full-corpus scans producing byte-identical function lists, the full extraction gauntlet, and both golden masters re-blessed. The remaining residual (9 occurrences, previously miscounted as more of the same shared mistake in the earlier verdict's blanket debit -- corrected here) is a DIFFERENT, unrelated finding: GitGalaxy and tree-sitter are CORRECT here, ctags is wrong. Two sub-patterns confirmed: (1) the already-documented macro-as-return-type-prefix pattern (`IFACEMETHODIMP_(void)` immediately before the real `FancyZones::Run() noexcept` -- ctags tags the macro invocation itself and loses the real name, powertoys/FancyZones.cpp, 4 occurrences); (2) a newly-confirmed, NOT yet root-caused ctags miss on an ordinary virtual method with no macro involvement at all (`virtual RID mesh_create_from_surfaces(const Vector &p_surfaces, int p_blend_shape_count = 0) override { ... }`, godot/rendering_server_default.h -- ctags produces zero tags anywhere near this real method, 5 occurrences). No credit_tools adjustment applies: GitGalaxy and tree-sitter are already a 2-of-3 agreeing pair here, which already satisfies reconcile_symbols' own natural precision-credit condition. 2026-08-29 follow-up (#2455): the tree_sitter_accuracy_audit / tri_comparison_gatherer shared tree-sitter reader now canonicalizes C++ conversion-operator and destructor names to GitGalaxy's own convention -- `Variant::operator String() const` -> `Variant::operator String`, `operator BitField() const` -> `operator BitField`, and the `_FORCE_INLINE_`-macro-mangled `~Variant` / `operator T` forms recovered from the sibling ERROR node -- and treats a small tree-sitter-cpp ERROR node (a function-like macro with no visible definition immediately before a special member) as a blind-spot region, promoting GitGalaxy's correct extraction there to ground truth. cpp function precision 95.0% -> 100.0%, extra_functions 68 -> 0. Several conversion operators GitGalaxy and tree-sitter now agree on (ctags names them without the `operator ` prefix or misses the `_FORCE_INLINE_`-macro forms) shifted into this shape, count ~9 -> ~21 -- still GitGalaxy+tree-sitter correct, ctags limited." }, "cpp/function/existence/agree[gitgalaxy]_vs[ctags,tree_sitter]": { "agreeing_tools": [ @@ -2813,7 +2822,7 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "cpp", - "last_reconciled_at": "2026-08-29T21:16:53Z", + "last_reconciled_at": "2026-08-29T21:43:10Z", "last_seen_count": 12, "last_seen_examples": [ { @@ -2909,7 +2918,7 @@ ], "metric": "existence", "status": "validated", - "still_reproduces": true, + "still_reproduces": false, "symbol_type": "function", "verdict": "Compound shape, three distinct causes confirmed via source, not one. (1) Most of the sample (OPCODE_WHILE x2, OPCODE_SWITCH, OPCODE) is the same GG+tree-sitter shared macro-misparse family documented in the sibling agree[gitgalaxy,tree_sitter]_vs[ctags] entry (godot/gdscript_vm.cpp's OPCODE/OPCODE_WHILE/OPCODE_SWITCH dispatch macros) -- here landing as 'GitGalaxy alone' because tree-sitter's own error recovery on this repeated macro pattern isn't fully deterministic across every occurrence, not because the underlying cause differs. GitGalaxy is WRONG for this portion (same debit as the sibling entry, not double-counted here since debit_tools is per-shape). (2) `attribute_buffer_applier_factories_`/`m_draggingState`/`std::thread` are a separate, real GitGalaxy FALSE POSITIVE: a lambda passed as a constructor argument or member-initializer-list entry (`m_draggingState([this]() {...}),`, `std::thread([...]() {...}).detach();`) is misread as a function definition. Filed as https://github.com/squid-protocol/gitgalaxy/issues/2013 . (3) `Variant::operator ::RID`/`Variant::operator ::AABB`/`Variant::operator Object *` are real functions GitGalaxy correctly finds (confirmed via godot/variant.cpp source) that didn't rank-match ctags/tree-sitter's own readings of the same functions by exact name string -- a residual, low-priority naming-comparison edge case (global-scope `::`-prefixed conversion-operator return types) not chased further in this pass. No credit/debit applied given the mixed, three-cause nature of this shape." }, @@ -2926,7 +2935,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "cpp", - "last_reconciled_at": "2026-08-29T21:16:53Z", + "last_reconciled_at": "2026-08-29T21:43:10Z", "last_seen_count": 64, "last_seen_examples": [ { @@ -3030,8 +3039,8 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "cpp", - "last_reconciled_at": "2026-08-29T21:16:53Z", - "last_seen_count": 200, + "last_reconciled_at": "2026-08-29T21:43:10Z", + "last_seen_count": 136, "last_seen_examples": [ { "file_path": "godot/gdscript_vm.cpp", @@ -3128,7 +3137,7 @@ "status": "validated", "still_reproduces": true, "symbol_type": "function", - "verdict": "UPDATED: count grew (98 -> 194) as a direct, correct consequence of the OPCODE-macro fix documented in the sibling agree[gitgalaxy,tree_sitter]_vs[ctags] entry -- tree-sitter's OWN OPCODE-family misparse (godot/gdscript_vm.cpp's bytecode-dispatch macro) was previously MASKED by GitGalaxy sharing the identical mistake (both wrong == they 'agreed', landing in the other shape instead of this one). Now that GitGalaxy correctly excludes known macro invocations, tree-sitter's own grammar limitation on this exact pattern is cleanly isolated here instead, confirmed via the shape's own examples (repeated `OPCODE` entries at godot/gdscript_vm.cpp, tree_sitter line set, ctags/gitgalaxy both None). The remaining, previously-documented causes (bare control-flow-keyword/`void` hallucination, conversion-operator naming-suffix convention) are unchanged and still contribute the bulk of the original 98. No credit/debit applies -- tree-sitter is the one that's wrong here, alone." + "verdict": "UPDATED: count grew (98 -> 194) as a direct, correct consequence of the OPCODE-macro fix documented in the sibling agree[gitgalaxy,tree_sitter]_vs[ctags] entry -- tree-sitter's OWN OPCODE-family misparse (godot/gdscript_vm.cpp's bytecode-dispatch macro) was previously MASKED by GitGalaxy sharing the identical mistake (both wrong == they 'agreed', landing in the other shape instead of this one). Now that GitGalaxy correctly excludes known macro invocations, tree-sitter's own grammar limitation on this exact pattern is cleanly isolated here instead, confirmed via the shape's own examples (repeated `OPCODE` entries at godot/gdscript_vm.cpp, tree_sitter line set, ctags/gitgalaxy both None). The remaining, previously-documented causes (bare control-flow-keyword/`void` hallucination, conversion-operator naming-suffix convention) are unchanged and still contribute the bulk of the original 98. No credit/debit applies -- tree-sitter is the one that's wrong here, alone. 2026-08-29 follow-up (#2455): the tree_sitter_accuracy_audit / tri_comparison_gatherer shared tree-sitter reader now canonicalizes C++ conversion-operator and destructor names to GitGalaxy's own convention -- `Variant::operator String() const` -> `Variant::operator String`, `operator BitField() const` -> `operator BitField`, and the `_FORCE_INLINE_`-macro-mangled `~Variant` / `operator T` forms recovered from the sibling ERROR node -- and treats a small tree-sitter-cpp ERROR node (a function-like macro with no visible definition immediately before a special member) as a blind-spot region, promoting GitGalaxy's correct extraction there to ground truth. cpp function precision 95.0% -> 100.0%, extra_functions 68 -> 0. That resolved the 'conversion-operator naming-suffix convention' cause named above (~55 occurrences moved into GitGalaxy+tree-sitter consensus, count ~194 -> ~136). The residual is the genuine OPCODE-macro and bare-control-flow-keyword / bare-`void` hallucinations, still tree-sitter's alone." }, "cpp/function/existence/agree[tree_sitter]_vs[gitgalaxy]": { "agreeing_tools": [ @@ -3143,7 +3152,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "cpp", - "last_reconciled_at": "2026-08-29T21:16:53Z", + "last_reconciled_at": "2026-08-29T21:43:10Z", "last_seen_count": 195, "last_seen_examples": [ { @@ -3247,7 +3256,7 @@ "investigated_at": "2026-08-21T18:13:27Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "csharp", - "last_reconciled_at": "2026-08-29T21:16:58Z", + "last_reconciled_at": "2026-08-29T21:43:13Z", "last_seen_count": 9, "last_seen_examples": [ { @@ -3352,7 +3361,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "csharp", - "last_reconciled_at": "2026-08-29T21:16:58Z", + "last_reconciled_at": "2026-08-29T21:43:13Z", "last_seen_count": 5, "last_seen_examples": [ { @@ -3421,7 +3430,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "csharp", - "last_reconciled_at": "2026-08-29T21:16:58Z", + "last_reconciled_at": "2026-08-29T21:43:13Z", "last_seen_count": 6, "last_seen_examples": [ { @@ -3498,7 +3507,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "csharp", - "last_reconciled_at": "2026-08-29T21:16:58Z", + "last_reconciled_at": "2026-08-29T21:43:13Z", "last_seen_count": 9, "last_seen_examples": [ { @@ -3597,7 +3606,7 @@ "investigated_at": "2026-08-21T20:06:45Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "csharp", - "last_reconciled_at": "2026-08-29T21:16:58Z", + "last_reconciled_at": "2026-08-29T21:43:13Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -3630,7 +3639,7 @@ "investigated_at": "2026-08-21T20:06:45Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "csharp", - "last_reconciled_at": "2026-08-29T21:16:58Z", + "last_reconciled_at": "2026-08-29T21:43:13Z", "last_seen_count": 7, "last_seen_examples": [ { @@ -3720,7 +3729,7 @@ "investigated_at": "2026-08-21T20:06:45Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "csharp", - "last_reconciled_at": "2026-08-29T21:16:58Z", + "last_reconciled_at": "2026-08-29T21:43:13Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -3761,7 +3770,7 @@ "investigated_at": "2026-08-21T20:06:45Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "csharp", - "last_reconciled_at": "2026-08-29T21:16:58Z", + "last_reconciled_at": "2026-08-29T21:43:13Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -3794,7 +3803,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "csharp", - "last_reconciled_at": "2026-08-29T21:16:58Z", + "last_reconciled_at": "2026-08-29T21:43:13Z", "last_seen_count": 26, "last_seen_examples": [ { @@ -3898,7 +3907,7 @@ "investigated_at": "2026-08-21T18:13:27Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "csharp", - "last_reconciled_at": "2026-08-29T21:16:58Z", + "last_reconciled_at": "2026-08-29T21:43:13Z", "last_seen_count": 271, "last_seen_examples": [ { @@ -4012,7 +4021,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "csharp", - "last_reconciled_at": "2026-08-29T21:16:58Z", + "last_reconciled_at": "2026-08-29T21:43:13Z", "last_seen_count": 4, "last_seen_examples": [ { @@ -4072,7 +4081,7 @@ "investigated_at": "2026-08-21T18:13:27Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "csharp", - "last_reconciled_at": "2026-08-29T21:16:58Z", + "last_reconciled_at": "2026-08-29T21:43:13Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -4105,7 +4114,7 @@ "investigated_at": "2026-08-21T18:13:44Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "csharp", - "last_reconciled_at": "2026-08-29T21:16:58Z", + "last_reconciled_at": "2026-08-29T21:43:13Z", "last_seen_count": 108, "last_seen_examples": [ { @@ -4221,7 +4230,7 @@ "investigated_at": "2026-08-21T21:45:53Z", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "csharp", - "last_reconciled_at": "2026-08-29T21:16:58Z", + "last_reconciled_at": "2026-08-29T21:43:13Z", "last_seen_count": 46, "last_seen_examples": [ { @@ -4334,7 +4343,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "csharp", - "last_reconciled_at": "2026-08-29T21:16:58Z", + "last_reconciled_at": "2026-08-29T21:43:13Z", "last_seen_count": 317, "last_seen_examples": [ { @@ -4438,7 +4447,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "csharp", - "last_reconciled_at": "2026-08-29T21:16:58Z", + "last_reconciled_at": "2026-08-29T21:43:13Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -4471,7 +4480,7 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep", "language": "css", - "last_reconciled_at": "2026-08-29T21:16:59Z", + "last_reconciled_at": "2026-08-29T21:43:15Z", "last_seen_count": 3, "last_seen_examples": [ { @@ -4520,7 +4529,7 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep", "language": "dart", - "last_reconciled_at": "2026-08-29T21:17:03Z", + "last_reconciled_at": "2026-08-29T21:43:17Z", "last_seen_count": 6, "last_seen_examples": [ { @@ -4591,7 +4600,7 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep", "language": "dart", - "last_reconciled_at": "2026-08-29T21:17:03Z", + "last_reconciled_at": "2026-08-29T21:43:17Z", "last_seen_count": 12, "last_seen_examples": [ { @@ -4694,7 +4703,7 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep", "language": "dart", - "last_reconciled_at": "2026-08-29T21:17:03Z", + "last_reconciled_at": "2026-08-29T21:43:17Z", "last_seen_count": 9, "last_seen_examples": [ { @@ -4790,7 +4799,7 @@ "investigated_at": "2026-08-21T23:35:49Z", "investigated_by": "Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "fortran", - "last_reconciled_at": "2026-08-29T21:17:08Z", + "last_reconciled_at": "2026-08-29T21:43:22Z", "last_seen_count": 8, "last_seen_examples": [ { @@ -4891,7 +4900,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "fortran", - "last_reconciled_at": "2026-08-29T21:17:08Z", + "last_reconciled_at": "2026-08-29T21:43:22Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -4923,7 +4932,7 @@ "investigated_at": "2026-08-21T00:00:00Z", "investigated_by": "Claude Sonnet 5, direct investigation (surfaced while re-blessing the tri-comparison chart after #1973/#1985)", "language": "fortran", - "last_reconciled_at": "2026-08-29T21:17:08Z", + "last_reconciled_at": "2026-08-29T21:43:22Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -4955,7 +4964,7 @@ "investigated_at": "2026-08-21T23:35:49Z", "investigated_by": "Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "fortran", - "last_reconciled_at": "2026-08-29T21:17:08Z", + "last_reconciled_at": "2026-08-29T21:43:22Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -4986,7 +4995,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "fortran", - "last_reconciled_at": "2026-08-29T21:17:08Z", + "last_reconciled_at": "2026-08-29T21:43:22Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -5018,7 +5027,7 @@ "investigated_at": "2026-08-21T23:35:49Z", "investigated_by": "Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "fortran", - "last_reconciled_at": "2026-08-29T21:17:08Z", + "last_reconciled_at": "2026-08-29T21:43:22Z", "last_seen_count": 15, "last_seen_examples": [ { @@ -5132,7 +5141,7 @@ "investigated_at": "2026-08-21T23:35:49Z", "investigated_by": "Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "fortran", - "last_reconciled_at": "2026-08-29T21:17:08Z", + "last_reconciled_at": "2026-08-29T21:43:22Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -5176,7 +5185,7 @@ "investigated_at": "2026-08-21T23:35:49Z", "investigated_by": "Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "fortran", - "last_reconciled_at": "2026-08-29T21:17:08Z", + "last_reconciled_at": "2026-08-29T21:43:22Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -5208,7 +5217,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "fortran", - "last_reconciled_at": "2026-08-29T21:17:08Z", + "last_reconciled_at": "2026-08-29T21:43:22Z", "last_seen_count": 16, "last_seen_examples": [ { @@ -5317,7 +5326,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "go", - "last_reconciled_at": "2026-08-29T21:17:10Z", + "last_reconciled_at": "2026-08-29T21:43:24Z", "last_seen_count": 75, "last_seen_examples": [ { @@ -5429,7 +5438,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "go", - "last_reconciled_at": "2026-08-29T21:17:10Z", + "last_reconciled_at": "2026-08-29T21:43:24Z", "last_seen_count": 75, "last_seen_examples": [ { @@ -5533,7 +5542,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Claude Sonnet 5 (session investigation)", "language": "haskell", - "last_reconciled_at": "2026-08-29T21:17:12Z", + "last_reconciled_at": "2026-08-29T21:43:27Z", "last_seen_count": 16, "last_seen_examples": [ { @@ -5645,7 +5654,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Claude Sonnet 5 (session investigation)", "language": "haskell", - "last_reconciled_at": "2026-08-29T21:17:12Z", + "last_reconciled_at": "2026-08-29T21:43:27Z", "last_seen_count": 9, "last_seen_examples": [ { @@ -5750,7 +5759,7 @@ "investigated_at": "2026-08-29", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "haskell", - "last_reconciled_at": "2026-08-29T21:17:12Z", + "last_reconciled_at": "2026-08-29T21:43:27Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -5783,7 +5792,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "haskell", - "last_reconciled_at": "2026-08-29T21:17:12Z", + "last_reconciled_at": "2026-08-29T21:43:27Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -5816,7 +5825,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Claude Sonnet 5 (dispatched agent investigation)", "language": "haskell", - "last_reconciled_at": "2026-08-29T21:17:12Z", + "last_reconciled_at": "2026-08-29T21:43:27Z", "last_seen_count": 102, "last_seen_examples": [ { @@ -5930,7 +5939,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Claude Sonnet 5 (dispatched agent investigation)", "language": "haskell", - "last_reconciled_at": "2026-08-29T21:17:12Z", + "last_reconciled_at": "2026-08-29T21:43:27Z", "last_seen_count": 69, "last_seen_examples": [ { @@ -6046,7 +6055,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Claude Sonnet 5 (session investigation)", "language": "haskell", - "last_reconciled_at": "2026-08-29T21:17:12Z", + "last_reconciled_at": "2026-08-29T21:43:27Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -6078,7 +6087,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "haskell", - "last_reconciled_at": "2026-08-29T21:17:12Z", + "last_reconciled_at": "2026-08-29T21:43:27Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -6110,7 +6119,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "haskell", - "last_reconciled_at": "2026-08-29T21:17:12Z", + "last_reconciled_at": "2026-08-29T21:43:27Z", "last_seen_count": 91, "last_seen_examples": [ { @@ -6224,7 +6233,7 @@ "investigated_at": "2026-08-29", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "html", - "last_reconciled_at": "2026-08-29T21:17:13Z", + "last_reconciled_at": "2026-08-29T21:43:28Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -6257,7 +6266,7 @@ "investigated_at": "2026-08-29", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "html", - "last_reconciled_at": "2026-08-29T21:17:13Z", + "last_reconciled_at": "2026-08-29T21:43:28Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -6290,7 +6299,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "java", - "last_reconciled_at": "2026-08-29T21:17:13Z", + "last_reconciled_at": "2026-08-29T21:43:30Z", "last_seen_count": 28, "last_seen_examples": [ { @@ -6404,7 +6413,7 @@ "investigated_at": "2026-08-22T01:15:53Z", "investigated_by": "Sonnet 5, direct investigation via tri-comparison-ledger-sweep", "language": "java", - "last_reconciled_at": "2026-08-29T21:17:13Z", + "last_reconciled_at": "2026-08-29T21:43:30Z", "last_seen_count": 12, "last_seen_examples": [ { @@ -6518,7 +6527,7 @@ "investigated_at": "2026-08-22T01:15:53Z", "investigated_by": "Sonnet 5, direct investigation via tri-comparison-ledger-sweep", "language": "java", - "last_reconciled_at": "2026-08-29T21:17:13Z", + "last_reconciled_at": "2026-08-29T21:43:30Z", "last_seen_count": 28, "last_seen_examples": [ { @@ -6631,7 +6640,7 @@ "investigated_at": "2026-08-22T01:15:53Z", "investigated_by": "Sonnet 5, direct investigation via tri-comparison-ledger-sweep", "language": "java", - "last_reconciled_at": "2026-08-29T21:17:13Z", + "last_reconciled_at": "2026-08-29T21:43:30Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -6664,7 +6673,7 @@ "investigated_at": "2026-08-22T01:15:53Z", "investigated_by": "Sonnet 5, direct investigation via tri-comparison-ledger-sweep", "language": "java", - "last_reconciled_at": "2026-08-29T21:17:13Z", + "last_reconciled_at": "2026-08-29T21:43:30Z", "last_seen_count": 3, "last_seen_examples": [ { @@ -6715,7 +6724,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "java", - "last_reconciled_at": "2026-08-29T21:17:13Z", + "last_reconciled_at": "2026-08-29T21:43:30Z", "last_seen_count": 315, "last_seen_examples": [ { @@ -6829,7 +6838,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "java", - "last_reconciled_at": "2026-08-29T21:17:13Z", + "last_reconciled_at": "2026-08-29T21:43:30Z", "last_seen_count": 3, "last_seen_examples": [ { @@ -6880,7 +6889,7 @@ "investigated_at": "2026-08-21", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "javascript", - "last_reconciled_at": "2026-08-29T21:17:16Z", + "last_reconciled_at": "2026-08-29T21:44:29Z", "last_seen_count": 93, "last_seen_examples": [ { @@ -6994,7 +7003,7 @@ "investigated_at": "2026-08-21", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "javascript", - "last_reconciled_at": "2026-08-29T21:17:16Z", + "last_reconciled_at": "2026-08-29T21:44:29Z", "last_seen_count": 7, "last_seen_examples": [ { @@ -7081,7 +7090,7 @@ "investigated_at": "2026-08-21", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "javascript", - "last_reconciled_at": "2026-08-29T21:17:16Z", + "last_reconciled_at": "2026-08-29T21:44:29Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -7121,7 +7130,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "javascript", - "last_reconciled_at": "2026-08-29T21:17:16Z", + "last_reconciled_at": "2026-08-29T21:44:29Z", "last_seen_count": 45, "last_seen_examples": [ { @@ -7225,7 +7234,7 @@ "investigated_at": "2026-08-21", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "javascript", - "last_reconciled_at": "2026-08-29T21:17:16Z", + "last_reconciled_at": "2026-08-29T21:44:29Z", "last_seen_count": 13, "last_seen_examples": [ { @@ -7339,7 +7348,7 @@ "investigated_at": "2026-08-21", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "javascript", - "last_reconciled_at": "2026-08-29T21:17:16Z", + "last_reconciled_at": "2026-08-29T21:44:29Z", "last_seen_count": 6, "last_seen_examples": [ { @@ -7417,7 +7426,7 @@ "investigated_at": "2026-08-21", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "javascript", - "last_reconciled_at": "2026-08-29T21:17:16Z", + "last_reconciled_at": "2026-08-29T21:44:29Z", "last_seen_count": 265, "last_seen_examples": [ { @@ -7533,7 +7542,7 @@ "investigated_at": "2026-08-21", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "javascript", - "last_reconciled_at": "2026-08-29T21:17:16Z", + "last_reconciled_at": "2026-08-29T21:44:29Z", "last_seen_count": 189, "last_seen_examples": [ { @@ -7646,7 +7655,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "javascript", - "last_reconciled_at": "2026-08-29T21:17:16Z", + "last_reconciled_at": "2026-08-29T21:44:29Z", "last_seen_count": 194, "last_seen_examples": [ { @@ -7750,7 +7759,7 @@ "investigated_at": "2026-08-21", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "javascript", - "last_reconciled_at": "2026-08-29T21:17:16Z", + "last_reconciled_at": "2026-08-29T21:44:29Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -7783,7 +7792,7 @@ "investigated_at": "2026-08-22T11:40:48Z", "investigated_by": "claude-sonnet-5, dispatched via tri-comparison-ledger-sweep (direct investigation, no Gemini dispatch -- small occurrence counts, answer was directly checkable)", "language": "kotlin", - "last_reconciled_at": "2026-08-29T21:17:18Z", + "last_reconciled_at": "2026-08-29T21:43:35Z", "last_seen_count": 3, "last_seen_examples": [ { @@ -7836,7 +7845,7 @@ "investigated_at": "2026-08-22T11:40:48Z", "investigated_by": "claude-sonnet-5, dispatched via tri-comparison-ledger-sweep (direct investigation, no Gemini dispatch -- small occurrence counts, answer was directly checkable)", "language": "kotlin", - "last_reconciled_at": "2026-08-29T21:17:18Z", + "last_reconciled_at": "2026-08-29T21:43:35Z", "last_seen_count": 15, "last_seen_examples": [ { @@ -7950,7 +7959,7 @@ "investigated_at": "2026-08-22T11:45:48Z", "investigated_by": "claude-sonnet-5, dispatched via tri-comparison-ledger-sweep (direct investigation, no Gemini dispatch)", "language": "kotlin", - "last_reconciled_at": "2026-08-29T21:17:18Z", + "last_reconciled_at": "2026-08-29T21:43:35Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -7983,7 +7992,7 @@ "investigated_at": "2026-08-22T11:40:48Z", "investigated_by": "claude-sonnet-5, dispatched via tri-comparison-ledger-sweep (direct investigation, no Gemini dispatch -- small occurrence counts, answer was directly checkable)", "language": "kotlin", - "last_reconciled_at": "2026-08-29T21:17:18Z", + "last_reconciled_at": "2026-08-29T21:43:35Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -8016,7 +8025,7 @@ "investigated_at": "2026-08-29", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "lua", - "last_reconciled_at": "2026-08-29T21:17:22Z", + "last_reconciled_at": "2026-08-29T21:43:39Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -8047,7 +8056,7 @@ "investigated_at": "2026-08-29", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "lua", - "last_reconciled_at": "2026-08-29T21:17:22Z", + "last_reconciled_at": "2026-08-29T21:43:39Z", "last_seen_count": 42, "last_seen_examples": [ { @@ -8161,7 +8170,7 @@ "investigated_at": "2026-08-29", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "lua", - "last_reconciled_at": "2026-08-29T21:17:22Z", + "last_reconciled_at": "2026-08-29T21:43:39Z", "last_seen_count": 6, "last_seen_examples": [ { @@ -8239,7 +8248,7 @@ "investigated_at": "2026-08-29", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "lua", - "last_reconciled_at": "2026-08-29T21:17:22Z", + "last_reconciled_at": "2026-08-29T21:43:39Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -8281,7 +8290,7 @@ "investigated_at": "2026-08-29", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "lua", - "last_reconciled_at": "2026-08-29T21:17:22Z", + "last_reconciled_at": "2026-08-29T21:43:39Z", "last_seen_count": 486, "last_seen_examples": [ { @@ -8395,7 +8404,7 @@ "investigated_at": "2026-08-29", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "lua", - "last_reconciled_at": "2026-08-29T21:17:22Z", + "last_reconciled_at": "2026-08-29T21:43:39Z", "last_seen_count": 21, "last_seen_examples": [ { @@ -8509,7 +8518,7 @@ "investigated_at": "2026-08-29", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "lua", - "last_reconciled_at": "2026-08-29T21:17:22Z", + "last_reconciled_at": "2026-08-29T21:43:39Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -8542,7 +8551,7 @@ "investigated_at": "2026-08-29", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "lua", - "last_reconciled_at": "2026-08-29T21:17:22Z", + "last_reconciled_at": "2026-08-29T21:43:39Z", "last_seen_count": 12, "last_seen_examples": [ { @@ -8657,7 +8666,7 @@ "investigated_at": "2026-08-24T03:52:19Z", "investigated_by": "Antigravity (direct source read)", "language": "m4", - "last_reconciled_at": "2026-08-29T21:17:23Z", + "last_reconciled_at": "2026-08-29T21:43:41Z", "last_seen_count": 4, "last_seen_examples": [ { @@ -8714,7 +8723,7 @@ "investigated_at": "2026-08-24T03:52:19Z", "investigated_by": "Antigravity (direct source read)", "language": "m4", - "last_reconciled_at": "2026-08-29T21:17:23Z", + "last_reconciled_at": "2026-08-29T21:43:41Z", "last_seen_count": 76, "last_seen_examples": [ { @@ -8819,7 +8828,7 @@ "investigated_at": "2026-08-24T03:52:19Z", "investigated_by": "Antigravity (direct source read)", "language": "m4", - "last_reconciled_at": "2026-08-29T21:17:23Z", + "last_reconciled_at": "2026-08-29T21:43:41Z", "last_seen_count": 36, "last_seen_examples": [ { @@ -8923,7 +8932,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "makefile", - "last_reconciled_at": "2026-08-29T21:17:24Z", + "last_reconciled_at": "2026-08-29T21:43:42Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -8958,7 +8967,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "matlab", - "last_reconciled_at": "2026-08-29T21:17:25Z", + "last_reconciled_at": "2026-08-29T21:43:44Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -8991,7 +9000,7 @@ "investigated_at": "2026-08-25T00:00:00Z", "investigated_by": "Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep", "language": "objective-c", - "last_reconciled_at": "2026-08-29T21:17:26Z", + "last_reconciled_at": "2026-08-29T21:43:45Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -9035,7 +9044,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "objective-c", - "last_reconciled_at": "2026-08-29T21:17:26Z", + "last_reconciled_at": "2026-08-29T21:43:45Z", "last_seen_count": 89, "last_seen_examples": [ { @@ -9149,7 +9158,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "objective-c", - "last_reconciled_at": "2026-08-29T21:17:26Z", + "last_reconciled_at": "2026-08-29T21:43:45Z", "last_seen_count": 104, "last_seen_examples": [ { @@ -9265,7 +9274,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "objective-c", - "last_reconciled_at": "2026-08-29T21:17:26Z", + "last_reconciled_at": "2026-08-29T21:43:45Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -9297,7 +9306,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "objective-c", - "last_reconciled_at": "2026-08-29T21:17:26Z", + "last_reconciled_at": "2026-08-29T21:43:45Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -9331,7 +9340,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "objective-c", - "last_reconciled_at": "2026-08-29T21:17:26Z", + "last_reconciled_at": "2026-08-29T21:43:45Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -9372,7 +9381,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "objective-c", - "last_reconciled_at": "2026-08-29T21:17:26Z", + "last_reconciled_at": "2026-08-29T21:43:45Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -9412,7 +9421,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "perl", - "last_reconciled_at": "2026-08-29T21:17:31Z", + "last_reconciled_at": "2026-08-29T21:43:49Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -9445,7 +9454,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "perl", - "last_reconciled_at": "2026-08-29T21:17:31Z", + "last_reconciled_at": "2026-08-29T21:43:49Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -9477,7 +9486,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "perl", - "last_reconciled_at": "2026-08-29T21:17:31Z", + "last_reconciled_at": "2026-08-29T21:43:49Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -9507,7 +9516,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "perl", - "last_reconciled_at": "2026-08-29T21:17:31Z", + "last_reconciled_at": "2026-08-29T21:43:49Z", "last_seen_count": 12, "last_seen_examples": [ { @@ -9621,7 +9630,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "perl", - "last_reconciled_at": "2026-08-29T21:17:31Z", + "last_reconciled_at": "2026-08-29T21:43:49Z", "last_seen_count": 7, "last_seen_examples": [ { @@ -9708,7 +9717,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "perl", - "last_reconciled_at": "2026-08-29T21:17:31Z", + "last_reconciled_at": "2026-08-29T21:43:49Z", "last_seen_count": 7, "last_seen_examples": [ { @@ -9797,7 +9806,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "perl", - "last_reconciled_at": "2026-08-29T21:17:31Z", + "last_reconciled_at": "2026-08-29T21:43:49Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -9831,7 +9840,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "perl", - "last_reconciled_at": "2026-08-29T21:17:31Z", + "last_reconciled_at": "2026-08-29T21:43:49Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -9863,7 +9872,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), direct investigation correcting a prior misdiagnosis", "language": "perl", - "last_reconciled_at": "2026-08-29T21:17:31Z", + "last_reconciled_at": "2026-08-29T21:43:49Z", "last_seen_count": 122, "last_seen_examples": [ { @@ -9976,7 +9985,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "perl", - "last_reconciled_at": "2026-08-29T21:17:31Z", + "last_reconciled_at": "2026-08-29T21:43:49Z", "last_seen_count": 122, "last_seen_examples": [ { @@ -10080,12 +10089,12 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "php", - "last_reconciled_at": "2026-08-29T21:17:35Z", + "last_reconciled_at": "2026-08-29T21:43:52Z", "last_seen_count": 1, "last_seen_examples": [ { "file_path": "laravel_core/BladeCompiler.php", - "name": "AnonymousClassa08aab890100", + "name": "AnonymousClassda5cee600100", "readings": { "ctags": 342, "gitgalaxy": null, @@ -10113,7 +10122,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "php", - "last_reconciled_at": "2026-08-29T21:17:35Z", + "last_reconciled_at": "2026-08-29T21:43:52Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -10146,7 +10155,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "php", - "last_reconciled_at": "2026-08-29T21:17:35Z", + "last_reconciled_at": "2026-08-29T21:43:52Z", "last_seen_count": 68, "last_seen_examples": [ { @@ -10259,7 +10268,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "php", - "last_reconciled_at": "2026-08-29T21:17:35Z", + "last_reconciled_at": "2026-08-29T21:43:52Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -10291,7 +10300,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "powershell", - "last_reconciled_at": "2026-08-29T21:17:39Z", + "last_reconciled_at": "2026-08-29T21:43:55Z", "last_seen_count": 9, "last_seen_examples": [ { @@ -10396,7 +10405,7 @@ "investigated_at": "2026-08-29", "investigated_by": "Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep", "language": "powershell", - "last_reconciled_at": "2026-08-29T21:17:39Z", + "last_reconciled_at": "2026-08-29T21:43:55Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -10427,7 +10436,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "powershell", - "last_reconciled_at": "2026-08-29T21:17:39Z", + "last_reconciled_at": "2026-08-29T21:43:55Z", "last_seen_count": 5, "last_seen_examples": [ { @@ -10496,7 +10505,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "powershell", - "last_reconciled_at": "2026-08-29T21:17:39Z", + "last_reconciled_at": "2026-08-29T21:43:55Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -10529,7 +10538,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "powershell", - "last_reconciled_at": "2026-08-29T21:17:39Z", + "last_reconciled_at": "2026-08-29T21:43:55Z", "last_seen_count": 16, "last_seen_examples": [ { @@ -10642,7 +10651,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "powershell", - "last_reconciled_at": "2026-08-29T21:17:39Z", + "last_reconciled_at": "2026-08-29T21:43:55Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -10674,7 +10683,7 @@ "investigated_at": "2026-08-21T03:23:52Z", "investigated_by": "claude sonnet 5, tri-comparison-ledger-sweep (direct investigation, no dispatch needed)", "language": "python", - "last_reconciled_at": "2026-08-29T21:17:48Z", + "last_reconciled_at": "2026-08-29T21:44:00Z", "last_seen_count": 4, "last_seen_examples": [ { @@ -10733,7 +10742,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "python", - "last_reconciled_at": "2026-08-29T21:17:48Z", + "last_reconciled_at": "2026-08-29T21:44:00Z", "last_seen_count": 4, "last_seen_examples": [ { @@ -10789,7 +10798,7 @@ "investigated_at": "2026-08-21T03:23:52Z", "investigated_by": "claude sonnet 5, tri-comparison-ledger-sweep (direct investigation, no dispatch needed)", "language": "python", - "last_reconciled_at": "2026-08-29T21:17:48Z", + "last_reconciled_at": "2026-08-29T21:44:00Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -10822,7 +10831,7 @@ "investigated_at": "2026-08-21T12:03:03Z", "investigated_by": "claude sonnet 5, tri-comparison-ledger-sweep (direct investigation, no dispatch needed)", "language": "python", - "last_reconciled_at": "2026-08-29T21:17:48Z", + "last_reconciled_at": "2026-08-29T21:44:00Z", "last_seen_count": 100, "last_seen_examples": [ { @@ -10938,7 +10947,7 @@ "investigated_at": "2026-08-21T12:03:03Z", "investigated_by": "claude sonnet 5, tri-comparison-ledger-sweep (direct investigation, no dispatch needed)", "language": "python", - "last_reconciled_at": "2026-08-29T21:17:48Z", + "last_reconciled_at": "2026-08-29T21:44:00Z", "last_seen_count": 5, "last_seen_examples": [ { @@ -11007,7 +11016,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "python", - "last_reconciled_at": "2026-08-29T21:17:48Z", + "last_reconciled_at": "2026-08-29T21:44:00Z", "last_seen_count": 5, "last_seen_examples": [ { @@ -11075,7 +11084,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "python", - "last_reconciled_at": "2026-08-29T21:17:48Z", + "last_reconciled_at": "2026-08-29T21:44:00Z", "last_seen_count": 100, "last_seen_examples": [ { @@ -11181,7 +11190,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "ruby", - "last_reconciled_at": "2026-08-29T21:17:49Z", + "last_reconciled_at": "2026-08-29T21:44:02Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -11223,7 +11232,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "ruby", - "last_reconciled_at": "2026-08-29T21:17:49Z", + "last_reconciled_at": "2026-08-29T21:44:02Z", "last_seen_count": 6, "last_seen_examples": [ { @@ -11301,7 +11310,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "ruby", - "last_reconciled_at": "2026-08-29T21:17:49Z", + "last_reconciled_at": "2026-08-29T21:44:02Z", "last_seen_count": 3, "last_seen_examples": [ { @@ -11352,7 +11361,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "ruby", - "last_reconciled_at": "2026-08-29T21:17:49Z", + "last_reconciled_at": "2026-08-29T21:44:02Z", "last_seen_count": 6, "last_seen_examples": [ { @@ -11430,7 +11439,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "ruby", - "last_reconciled_at": "2026-08-29T21:17:49Z", + "last_reconciled_at": "2026-08-29T21:44:02Z", "last_seen_count": 9, "last_seen_examples": [ { @@ -11526,7 +11535,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "rust", - "last_reconciled_at": "2026-08-29T21:17:52Z", + "last_reconciled_at": "2026-08-29T21:44:04Z", "last_seen_count": 14, "last_seen_examples": [ { @@ -11640,7 +11649,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Claude Sonnet 5 (resolved directly via live ctags run, no dispatch)", "language": "rust", - "last_reconciled_at": "2026-08-29T21:17:52Z", + "last_reconciled_at": "2026-08-29T21:44:04Z", "last_seen_count": 3, "last_seen_examples": [ { @@ -11693,7 +11702,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Gemini (dispatched via tri-comparison-ledger-sweep), confirmed by Claude Sonnet 5", "language": "rust", - "last_reconciled_at": "2026-08-29T21:17:52Z", + "last_reconciled_at": "2026-08-29T21:44:04Z", "last_seen_count": 25, "last_seen_examples": [ { @@ -11806,7 +11815,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "rust", - "last_reconciled_at": "2026-08-29T21:17:52Z", + "last_reconciled_at": "2026-08-29T21:44:04Z", "last_seen_count": 25, "last_seen_examples": [ { @@ -11912,7 +11921,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "rust", - "last_reconciled_at": "2026-08-29T21:17:52Z", + "last_reconciled_at": "2026-08-29T21:44:04Z", "last_seen_count": 3, "last_seen_examples": [ { @@ -11963,7 +11972,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Gemini (dispatched via tri-comparison-ledger-sweep, 2 rounds with self-correction), confirmed by Claude Sonnet 5", "language": "rust", - "last_reconciled_at": "2026-08-29T21:17:52Z", + "last_reconciled_at": "2026-08-29T21:44:04Z", "last_seen_count": 21, "last_seen_examples": [ { @@ -12075,7 +12084,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Gemini (dispatched via tri-comparison-ledger-sweep), confirmed by Claude Sonnet 5", "language": "rust", - "last_reconciled_at": "2026-08-29T21:17:52Z", + "last_reconciled_at": "2026-08-29T21:44:04Z", "last_seen_count": 21, "last_seen_examples": [ { @@ -12189,7 +12198,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "rust", - "last_reconciled_at": "2026-08-29T21:17:52Z", + "last_reconciled_at": "2026-08-29T21:44:04Z", "last_seen_count": 83, "last_seen_examples": [ { @@ -12303,7 +12312,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "rust", - "last_reconciled_at": "2026-08-29T21:17:52Z", + "last_reconciled_at": "2026-08-29T21:44:04Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -12336,7 +12345,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Claude Sonnet 5 (resolved directly via live ctags run, no dispatch)", "language": "rust", - "last_reconciled_at": "2026-08-29T21:17:52Z", + "last_reconciled_at": "2026-08-29T21:44:04Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -12371,7 +12380,7 @@ "investigated_at": "2026-08-19T00:00:00Z", "investigated_by": "Claude Sonnet 5 (resolved from existing Claim 6 documentation, no dispatch)", "language": "rust", - "last_reconciled_at": "2026-08-29T21:17:52Z", + "last_reconciled_at": "2026-08-29T21:44:04Z", "last_seen_count": 152, "last_seen_examples": [ { @@ -12484,7 +12493,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "rust", - "last_reconciled_at": "2026-08-29T21:17:52Z", + "last_reconciled_at": "2026-08-29T21:44:04Z", "last_seen_count": 152, "last_seen_examples": [ { @@ -12590,7 +12599,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "scala", - "last_reconciled_at": "2026-08-29T21:17:54Z", + "last_reconciled_at": "2026-08-29T21:44:06Z", "last_seen_count": 16, "last_seen_examples": [ { @@ -12693,7 +12702,7 @@ "investigated_at": "2026-08-20", "investigated_by": "gemini-3.1-pro-high (agy), dispatched via tri-comparison-ledger-sweep, reviewed and fixed by claude-sonnet-5", "language": "scheme", - "last_reconciled_at": "2026-08-29T21:17:58Z", + "last_reconciled_at": "2026-08-29T21:44:09Z", "last_seen_count": 84, "last_seen_examples": [ { @@ -12800,7 +12809,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "scheme", - "last_reconciled_at": "2026-08-29T21:17:58Z", + "last_reconciled_at": "2026-08-29T21:44:09Z", "last_seen_count": 50, "last_seen_examples": [ { @@ -12902,7 +12911,7 @@ "investigated_at": "2026-08-28T23:59:00Z", "investigated_by": "Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep", "language": "shell", - "last_reconciled_at": "2026-08-29T21:18:01Z", + "last_reconciled_at": "2026-08-29T21:44:12Z", "last_seen_count": 10, "last_seen_examples": [ { @@ -13016,7 +13025,7 @@ "investigated_at": "2026-08-28T23:59:00Z", "investigated_by": "Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep", "language": "shell", - "last_reconciled_at": "2026-08-29T21:18:01Z", + "last_reconciled_at": "2026-08-29T21:44:12Z", "last_seen_count": 14, "last_seen_examples": [ { @@ -13130,7 +13139,7 @@ "investigated_at": "2026-08-28T23:59:00Z", "investigated_by": "Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep", "language": "shell", - "last_reconciled_at": "2026-08-29T21:18:01Z", + "last_reconciled_at": "2026-08-29T21:44:12Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -13172,7 +13181,7 @@ "investigated_at": "2026-08-28T23:59:00Z", "investigated_by": "Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep", "language": "shell", - "last_reconciled_at": "2026-08-29T21:18:01Z", + "last_reconciled_at": "2026-08-29T21:44:12Z", "last_seen_count": 41, "last_seen_examples": [ { @@ -13286,7 +13295,7 @@ "investigated_at": "2026-08-28T23:59:00Z", "investigated_by": "Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep", "language": "shell", - "last_reconciled_at": "2026-08-29T21:18:01Z", + "last_reconciled_at": "2026-08-29T21:44:12Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -13322,7 +13331,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "solidity", - "last_reconciled_at": "2026-08-29T21:18:02Z", + "last_reconciled_at": "2026-08-29T21:44:13Z", "last_seen_count": 6, "last_seen_examples": [ { @@ -13392,7 +13401,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "swift", - "last_reconciled_at": "2026-08-29T21:18:03Z", + "last_reconciled_at": "2026-08-29T21:44:15Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -13433,7 +13442,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "swift", - "last_reconciled_at": "2026-08-29T21:18:03Z", + "last_reconciled_at": "2026-08-29T21:44:15Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -13466,7 +13475,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "swift", - "last_reconciled_at": "2026-08-29T21:18:03Z", + "last_reconciled_at": "2026-08-29T21:44:15Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -13498,7 +13507,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "tcl", - "last_reconciled_at": "2026-08-29T21:18:04Z", + "last_reconciled_at": "2026-08-29T21:44:16Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -13536,7 +13545,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "tcl", - "last_reconciled_at": "2026-08-29T21:18:04Z", + "last_reconciled_at": "2026-08-29T21:44:16Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -13578,7 +13587,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "tcl", - "last_reconciled_at": "2026-08-29T21:18:04Z", + "last_reconciled_at": "2026-08-29T21:44:16Z", "last_seen_count": 4, "last_seen_examples": [ { @@ -13640,7 +13649,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "tcl", - "last_reconciled_at": "2026-08-29T21:18:04Z", + "last_reconciled_at": "2026-08-29T21:44:16Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -13672,7 +13681,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "tcl", - "last_reconciled_at": "2026-08-29T21:18:04Z", + "last_reconciled_at": "2026-08-29T21:44:16Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -13716,7 +13725,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "tcl", - "last_reconciled_at": "2026-08-29T21:18:04Z", + "last_reconciled_at": "2026-08-29T21:44:16Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -13761,7 +13770,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "tcl", - "last_reconciled_at": "2026-08-29T21:18:04Z", + "last_reconciled_at": "2026-08-29T21:44:16Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -13801,7 +13810,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "typescript", - "last_reconciled_at": "2026-08-29T21:18:11Z", + "last_reconciled_at": "2026-08-29T21:44:32Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -13843,7 +13852,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "typescript", - "last_reconciled_at": "2026-08-29T21:18:11Z", + "last_reconciled_at": "2026-08-29T21:44:32Z", "last_seen_count": 501, "last_seen_examples": [ { @@ -13955,7 +13964,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "typescript", - "last_reconciled_at": "2026-08-29T21:18:11Z", + "last_reconciled_at": "2026-08-29T21:44:32Z", "last_seen_count": 4, "last_seen_examples": [ { @@ -14015,7 +14024,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "typescript", - "last_reconciled_at": "2026-08-29T21:18:11Z", + "last_reconciled_at": "2026-08-29T21:44:32Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -14048,7 +14057,7 @@ "investigated_at": "2026-08-26T16:02:21Z", "investigated_by": "Claude (Sonnet 5), tri-comparison-ledger-sweep skill", "language": "typescript", - "last_reconciled_at": "2026-08-29T21:18:11Z", + "last_reconciled_at": "2026-08-29T21:44:32Z", "last_seen_count": 7, "last_seen_examples": [ { @@ -14135,7 +14144,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "typescript", - "last_reconciled_at": "2026-08-29T21:18:11Z", + "last_reconciled_at": "2026-08-29T21:44:32Z", "last_seen_count": 204, "last_seen_examples": [ { @@ -14249,7 +14258,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "typescript", - "last_reconciled_at": "2026-08-29T21:18:11Z", + "last_reconciled_at": "2026-08-29T21:44:32Z", "last_seen_count": 2223, "last_seen_examples": [ { @@ -14365,7 +14374,7 @@ "investigated_at": "2026-08-24T04:14:26Z", "investigated_by": "Antigravity (direct sweep)", "language": "typescript", - "last_reconciled_at": "2026-08-29T21:18:11Z", + "last_reconciled_at": "2026-08-29T21:44:32Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -14406,7 +14415,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "typescript", - "last_reconciled_at": "2026-08-29T21:18:11Z", + "last_reconciled_at": "2026-08-29T21:44:32Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -14448,7 +14457,7 @@ "investigated_at": "2026-08-26T16:02:21Z", "investigated_by": "Claude (Sonnet 5), tri-comparison-ledger-sweep skill", "language": "typescript", - "last_reconciled_at": "2026-08-29T21:18:11Z", + "last_reconciled_at": "2026-08-29T21:44:32Z", "last_seen_count": 3, "last_seen_examples": [ { @@ -14500,7 +14509,7 @@ "investigated_at": "2026-08-26T00:00:00Z", "investigated_by": "Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep", "language": "typescript", - "last_reconciled_at": "2026-08-29T21:18:11Z", + "last_reconciled_at": "2026-08-29T21:44:32Z", "last_seen_count": 183, "last_seen_examples": [ { @@ -14603,7 +14612,7 @@ "investigated_at": "2026-08-27", "investigated_by": "claude-sonnet-5, dispatched via tri-comparison-ledger-sweep", "language": "yacc", - "last_reconciled_at": "2026-08-29T21:18:11Z", + "last_reconciled_at": "2026-08-29T21:44:21Z", "last_seen_count": 18, "last_seen_examples": [ { @@ -14710,7 +14719,7 @@ "investigated_at": "2026-08-23T18:02:16Z", "investigated_by": "Antigravity", "language": "zig", - "last_reconciled_at": "2026-08-29T21:18:19Z", + "last_reconciled_at": "2026-08-29T21:44:36Z", "last_seen_count": 1, "last_seen_examples": [ { @@ -14745,7 +14754,7 @@ "investigated_at": "2026-08-23T18:02:16Z", "investigated_by": "Antigravity", "language": "zig", - "last_reconciled_at": "2026-08-29T21:18:19Z", + "last_reconciled_at": "2026-08-29T21:44:36Z", "last_seen_count": 2, "last_seen_examples": [ { @@ -14788,7 +14797,7 @@ "investigated_at": "2026-08-23T18:02:16Z", "investigated_by": "Antigravity", "language": "zig", - "last_reconciled_at": "2026-08-29T21:18:19Z", + "last_reconciled_at": "2026-08-29T21:44:36Z", "last_seen_count": 1, "last_seen_examples": [ { diff --git a/docs/self_scan/tri_comparison_points_of_interest.md b/docs/self_scan/tri_comparison_points_of_interest.md index 5b7deefd8..62c27e1ad 100644 --- a/docs/self_scan/tri_comparison_points_of_interest.md +++ b/docs/self_scan/tri_comparison_points_of_interest.md @@ -8,7 +8,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `agc_assembly` function existence: ctags agree, GitGalaxy differ -*2-vs-1 -- 215 occurrences as of 2026-08-29T21:16:28Z* +*2-vs-1 -- 215 occurrences as of 2026-08-29T21:42:55Z* **Verdict** (by Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep, 2026-08-20T00:00:00Z): > Mixed-cause shape, fully accounted for via a corpus-wide cross-reference of ctags' actual output against GitGalaxy's raw func_start regex matches and its real pipeline/DB output (215 + 50 = 265, no unexplained residual). (1) 215/265 (81%): ctags' Asm "l" kind tags EVERY line-start label unconditionally, including pure data/constant-definition labels (e.g. ERASCON1 OCTAL 00061, S10BITS, LSTBNKCH -- AGC_BLOCK_TWO_SELF-CHECK.agc:133 and nearby) that are never followed by an executable instruction. GitGalaxy's func_start regex deliberately requires the label be followed by a real instruction mnemonic from a fixed whitelist, so it does not count these as functions -- a genuine, intentional precision distinction (code label vs. data label), not a GitGalaxy defect; ctags' generic Asm parser has no way to make this distinction at all. (2) 50/265 (19%): a real, confirmed GitGalaxy engine defect in detector.py's _slice_by_labels (Mode A), independently root-caused to two separate bugs: (a) `RELINT` is incorrectly included in `self.assembly_returns`'s early-termination keyword list (detector.py:572-575) -- in real AGC assembly RELINT means "release interrupt inhibit" and commonly opens a long interrupt-handler routine rather than closing one, so it truncates the real body to one line (confirmed: ELOOPFIN, AGC_BLOCK_TWO_SELF-CHECK.agc:303, a 20+-line real routine collapsed to just its own label line); (b) the `len(block.splitlines()) < 2` guard (detector.py:1973) unconditionally discards legitimate single-instruction assembly subroutines when the next func_start match sits on the very next line (confirmed: SOPTION1-SOPTION5+, AGC_BLOCK_TWO_SELF-CHECK.agc:210-214, each a real one-instruction label). Filed as #1949 -- a follow-up read-only Gemini/agy dispatch confirmed both root causes generalize beyond agc_assembly to the shared Mode A mechanism (assembly, cobol, fortran, abap all independently exhibit bug 1; assembly and cobol also exhibit bug 2), plus two further bug-1 variants not visible from agc_assembly alone: the terminator regex also false-matches inside comments (assembly) and inside hyphenated identifier names (cobol), not just legitimate-but-misclassified instructions. See #1949 for the full cross-language evidence and fix scope. No credit/debit -- the 215 portion is an honest scope difference (not two tools independently wrong about the same fact), and the 50 portion is GitGalaxy's own unresolved bug, not something ctags corroborates or contradicts. @@ -28,7 +28,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `agc_assembly` function existence: GitGalaxy agree, ctags differ -*2-vs-1 -- 37 occurrences as of 2026-08-29T21:16:28Z* +*2-vs-1 -- 37 occurrences as of 2026-08-29T21:42:55Z* **Verdict** (by Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep, 2026-08-20T00:00:00Z): > Confirmed: all 35 occurrences are real AGC labels that GitGalaxy correctly extracts and Universal Ctags' generic Asm parser structurally cannot tag, due to AGC assembly's non-standard label-naming conventions. Two sub-patterns, both confirmed corpus-wide (14/35 + 21/35 = 35/35, not just the sample): (1) labels with an embedded hyphen -- AGC's own convention of naming a point relative to an event, e.g. TIG-35/TIG-30/CALLT-35 in BURN_BABY_BURN--MASTER_IGNITION_ROUTINE.agc:250/292/222 ("35/30 seconds before Time of Ignition"); (2) labels starting with a digit or a leading minus sign, e.g. 1CHK/2EBANK/-1CHK in AGC_BLOCK_TWO_SELF-CHECK.agc:184 (real label text is "-1CHK"). Directly verified via `ctags --language-force=Asm --kinds-Asm=l` against the real corpus files: ctags emits zero tags for any of these names (confirmed by grepping its actual output for the exact names and their surrounding CHK-suffixed siblings, which ARE tagged when they don't start with a hyphen/digit) -- ctags' Asm parser requires a tag name to start with a letter and contain no hyphen, neither of which is a real constraint in AGC assembly's own label syntax. GitGalaxy's func_start regex ([A-Z0-9_-]+ at line start) has no such restriction. This is a confirmed, structural ctags/Asm-parser limitation, not corroboration of anything wrong on GitGalaxy's side -- logged to docs/why_gitgalaxy_beats_ast_here.md. @@ -50,7 +50,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `assembly` function existence: ctags agree, GitGalaxy differ -*2-vs-1 -- 26 occurrences as of 2026-08-29T21:16:30Z* +*2-vs-1 -- 26 occurrences as of 2026-08-29T21:42:58Z* **Verdict** (by Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep, 2026-08-27T00:00:00Z): > Re-validated 2026-08-27 (same PR as the mirror shape). All 26 occurrences are ctags' generic Asm parser tagging line-start labels that are not subroutine entries; GitGalaxy correctly excludes every one, confirmed by reading corpus source for each. (1) Data-emission labels: `intro`/`error_message`/`commands`/`int_0x20` in bootos/os.asm (`db`/`dw` string and jump-table data), `A`/`B`/`C` in hellosilicon/matrixmultneon.s (`.short`/`.fill` matrices), `prtstr`/`getcreditcards`/`instr` (`.asciz`/`.ascii` strings). (2) Section / object markers in cosmopolitan/ape.S: `__ro`/`cstr`/`_gdt_end`/`sconf` (`.endobj`), `ape_loader` (`.incbin`), `ape_phdrs`/`ape_macho`/`ape_grub`/`ape_mz`/`apesh` (ELF/Mach-O/Multiboot/shell header data emitted via `.long`/`.ascii`), `_gdtr`/`_gdtrlo` (GDT register values), `ape_idata_idtend`/`ape_idata_iatend` (`.byte` terminators). (3) `.Lenv0`/`.Largv0` -- `.L`-prefixed GCC compiler-local labels `func_start`'s negative lookahead deliberately excludes. The four NASM local-label name-normalization cases from the earlier sample (`load_vec`/`loop`/`empty`/`find`) are gone: the reconciler now pairs them with GitGalaxy's dot-prefixed form. No GitGalaxy recall miss remains in this shape -- the #1949 `_slice_by_labels` cases were fixed and that issue is closed. ctags already pays for these in its own precision denominator (it alone claims them); no credit/debit adjustment applies -- there is no shared-consensus mistake to debit and nothing of GitGalaxy's to credit. @@ -70,7 +70,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `assembly` function existence: GitGalaxy agree, ctags differ -*2-vs-1 -- 2 occurrences as of 2026-08-29T21:16:30Z* +*2-vs-1 -- 2 occurrences as of 2026-08-29T21:42:58Z* **Verdict** (by Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep, 2026-08-27T00:00:00Z): > Re-validated 2026-08-27; down from 7 occurrences to 2 after two coordinated fixes in the same PR. Four of the original seven (`.load_vec`/`.loop`/`.empty`/`.find`, bootos/os.asm) were never a real disagreement: both tools found the identical label at the identical line, ctags' Asm parser just strips the leading `.` from NASM/GAS local labels while GitGalaxy keeps it verbatim -- `tri_comparison_reconcile.py` now normalizes a single leading dot before pairing, so these register as agreements. One (`ape.mbrpad`, cosmopolitan/ape.S:525) was a genuine GitGalaxy false positive -- a `.org`/`.endobj` MBR-padding object, not a subroutine -- now fixed by a `func_start` negative lookahead that rejects a label followed only by a pure data-emission / location-counter directive (the generic-assembly counterpart to agc_assembly's positive 'opcode must follow' lookahead). The remaining 2 (`.1`, `.2` in bootos/counter.asm:52,67) are real NASM numeric local CODE labels -- `.1:` opens a routine (`int 0x22` ...), `.2:` likewise (`mov [di],ax` ...) -- that Universal Ctags' Asm parser structurally cannot tag (a tag name must start with a letter). GitGalaxy is correct here and ctags has a confirmed structural limitation, not an open question -> credit_tools: [gitgalaxy]. @@ -84,7 +84,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `c` function existence: tree-sitter agree, GitGalaxy, ctags differ -*2-vs-1 -- 74 occurrences as of 2026-08-29T21:16:36Z* +*2-vs-1 -- 74 occurrences as of 2026-08-29T21:43:01Z* **Verdict** (by Gemini (dispatched via tri-comparison-ledger-sweep), confirmed by Claude Sonnet 5, 2026-08-19T00:00:00Z): > Confirmed, independently verified all 6 sampled names against real source -- GitGalaxy and ctags both correct, tree-sitter over-recalling from two related but distinct preprocessor-driven mechanisms, both already covered by existing infrastructure: (1) keyword/macro misparse -- 'if' (dictobject.c:522-527, an `#if SIZEOF_VOID_P > 4` / `else if` sequence desyncs the parse) and 'DICT___REVERSED___METHODDEF' (dictobject.c:5102, a PyMethodDef array-initializer macro, not a definition) are both ALREADY in tree_sitter_accuracy_audit.py's `_C_KNOWN_MACRO_HALLUCINATIONS` exclusion set (confirmed by reading it directly) -- this tri-comparison tool's raw walk deliberately doesn't apply that list (it's a curated, ground-truth-shaped judgment call, appropriately left to reconciliation per this module's own stated design, not baked into the walk). (2) dead #if 0 code -- '_PyObject_ManagedDictValidityCheck' (dictobject.c:7396) and 'tos_char'/'print_stack'/'print_stacks' (frameobject.c:1264-1313) are genuinely well-formed function definitions sitting entirely inside `#if 0 ... #endif` guards; tree-sitter has no preprocessor model and parses the dead branch as live code. Both mechanisms are already the exact shape docs/why_gitgalaxy_beats_ast_here.md's Claim 8 names generically ('a dead #if 0 block... macro definitions... that merely look structural') -- added these 4 new concrete citations to Claim 8's evidence section rather than treating this as a new finding. No GitHub issue -- both tools already behave as intended; this is expected, already-documented tree-sitter preprocessor-blindness surfacing under the new tri-comparison reconciliation, not a fresh defect. @@ -104,7 +104,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `c` function existence: GitGalaxy, ctags agree, tree-sitter differ -*2-vs-1 -- 13 occurrences as of 2026-08-29T21:16:36Z* +*2-vs-1 -- 13 occurrences as of 2026-08-29T21:43:01Z* **Verdict** (by Gemini (dispatched via tri-comparison-ledger-sweep), confirmed by Claude Sonnet 5, 2026-08-19T00:00:00Z): > 3 distinct causes, all genuine tree-sitter-c grammar limitations (GitGalaxy and ctags both correct in all 10 sampled cases) -- confirmed via dispatched investigation (which ran tree-sitter's C grammar directly and found ERROR nodes in every case) plus independent source-level spot-checks of all 3 trigger shapes. This is a C-scale instance of Claim 7 (CPP-directive-driven recall loss) -- added to that claim's evidence section. (1) 4 samples: an #if/#else pair splitting a single `if` condition inside a function body (ceval.c:33, _Py_ReachedRecursionLimitWithMargin). (2) 5 samples: bare, un-semicoloned macro invocations the grammar can't cleanly recover from, losing the next real function (object.c:1269-1271's _Py_COMP_DIAG_PUSH/IGNORE_DEPR_DECLS/POP before _PyObject_SetAttributeErrorContext; similar shape for the typeobject.c slot-getattr cluster). (3) 1 sample: #if/#endif wrapping only the `static` storage-class specifier, separated from the rest of the signature (micropython/compile.c:3473-3476, mp_compile_to_raw_code). Unlike Fortran's existing Claim 7 evidence (entire trailing sections lost), C's version is local -- one function lost per trigger, not a cascading region. No GitHub issue -- documented as new Claim 7 evidence, not a fixable tooling bug (this repo doesn't control tree-sitter-c's grammar). @@ -124,7 +124,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `c` function existence: ctags agree, GitGalaxy, tree-sitter differ -*2-vs-1 -- 7 occurrences as of 2026-08-29T21:16:36Z* +*2-vs-1 -- 7 occurrences as of 2026-08-29T21:43:01Z* **Verdict** (by Claude Sonnet 5 (resolved + fixed directly, no dispatch needed), 2026-08-19T00:00:00Z): > Confirmed real ctags limitation, not a GitGalaxy or tree-sitter defect -- resolved directly (no dispatch needed), source read confirms all 7 sampled names. RICHCMP_WRAPPER/SLOT0/SLOT1/SLOT1BINFULL are all-caps macro names; every occurrence is a MACRO INVOCATION (a call to a previously-#define'd boilerplate-generating macro), not a function definition -- confirmed at cpython/typeobject.c:10099 (`RICHCMP_WRAPPER(lt, Py_LT)`) and :10544 (`SLOT1(slot_mp_subscript, __getitem__, PyObject *)`), same shape as the multiple SLOT0/SLOT1 hits at different lines (each is a separate invocation of the same macro generating a different wrapper function). ctags' regex-based C parser tags the macro-invocation site itself as a function; GitGalaxy and tree-sitter both correctly don't. Deliberately NOT fixed with a curated name-exclusion list in the gatherer (unlike the sibling __anon* class shape, this would require hand-curating specific macro names -- the same ground-truth-judgment category tri_comparison_gatherer.py's own docstring reasons belongs in reconciliation, not the raw reader) -- documented in ctags_reader.py instead, alongside its existing per-language notes. @@ -141,7 +141,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `c` function existence: GitGalaxy agree, tree-sitter, ctags differ -*2-vs-1 -- 4 occurrences as of 2026-08-29T21:16:36Z* +*2-vs-1 -- 4 occurrences as of 2026-08-29T21:43:01Z* **Verdict** (by Claude Sonnet 5 (resolved directly, no dispatch needed), 2026-08-19T00:00:00Z): > Confirmed GitGalaxy correct, real finding -- a new, narrower instance of Claim 3 (parse-error cascade), added to docs/why_gitgalaxy_beats_ast_here.md. All 4 sampled names (slot_mp_ass_subscript:10544, slot_nb_inplace_power:10697, slot_tp_repr:10714, slot_tp_hash:10730, all cpython/typeobject.c) are ordinary, unremarkable function definitions -- nothing unusual individually -- but each sits directly after a bare SLOT0/SLOT1 macro-invocation LINE (`SLOT1(slot_mp_subscript, __getitem__, PyObject *)`, `SLOT0(slot_tp_str, __str__)`, etc.) that isn't valid freestanding C without macro expansion. GitGalaxy's regex has no adjacency sensitivity and finds all 4 correctly; both ctags and tree-sitter locally lose the SINGLE function immediately following each such line (recovers after just one function, not a full cascade to EOF -- confirmed by resolving all 4 as isolated single-function misses, not a growing region). Resolved directly, no dispatch needed -- same pattern verified at all 4 sample points before writing up. @@ -155,7 +155,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `c` function existence: tree-sitter, ctags agree, GitGalaxy differ -*2-vs-1 -- 3 occurrences as of 2026-08-29T21:16:36Z* +*2-vs-1 -- 3 occurrences as of 2026-08-29T21:43:01Z* **Verdict** (by Claude Sonnet 5 (resolved directly, no dispatch needed), 2026-08-19T00:00:00Z): > Not a new finding -- both sampled names are ALREADY in tree_sitter_accuracy_audit.py's _C_KNOWN_MACRO_HALLUCINATIONS exclusion set (confirmed by grep: 'EXPORT_FUN' and 'MICROPY_WRAP_MP_EXECUTE_BYTECODE' both present). This shape is the same already-documented macro-hallucination mechanism (Claim 8) as the earlier tree-sitter-alone shape, just with ctags ALSO independently hallucinating the same 2 names the same way (both tools' regex/grammar parsers get fooled by the same macro-definition text). GitGalaxy correctly excludes both. Resolved directly, no dispatch needed. @@ -168,7 +168,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `c` function existence: GitGalaxy, tree-sitter agree, ctags differ -*2-vs-1 -- 3 occurrences as of 2026-08-29T21:16:36Z* +*2-vs-1 -- 3 occurrences as of 2026-08-29T21:43:01Z* **Verdict** (by Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-21T00:00:00Z): > UPDATED after a real production fix (the same fix documented in cpp's agree[gitgalaxy,tree_sitter]_vs[ctags] entry -- c and cpp share the mechanism, `lang_id in ("c", "cpp")` in detector.py's `_slice_by_braces`). GitGalaxy now scans each file for `#define NAME(...)` function-like macro definitions and excludes any func_start match whose captured name is a known macro -- confirmed to eliminate the already-documented `RICHCMP_WRAPPER`/`SLOT0`/`SLOT1`/`SLOT1BINFULL`/`DICT___REVERSED___METHODDEF` cpython/typeobject.c false positives entirely (a direct `SELECT func_name ... WHERE func_name IN (...)` query against a fresh scan returned zero rows). The small residual (3 occurrences: `slot_nb_power`, `slot_nb_bool`, `wrap_next`, all cpython/typeobject.c) is a DIFFERENT, unrelated finding, not yet individually root-caused -- GitGalaxy and tree-sitter both correctly find these real functions and ctags doesn't; plausibly the same category of ctags miss documented in cpp's sibling entry (an ordinary function ctags' C++ parser fails to tag for reasons unrelated to macros) but not directly confirmed for these three specific names. No credit_tools adjustment applies -- already a naturally-corroborated 2-of-3 agreeing pair. @@ -181,7 +181,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `c` function args: GitGalaxy, tree-sitter agree, ctags differ -*2-vs-1 -- 1 occurrence as of 2026-08-29T21:16:36Z* +*2-vs-1 -- 1 occurrence as of 2026-08-29T21:43:01Z* **Verdict** (by Claude Sonnet 5 (resolved + fixed directly, no dispatch needed), 2026-08-19T00:00:00Z): > Not a GitGalaxy or ctags defect -- a bug in this tool's OWN _count_ctags_signature_params (tri_comparison_gatherer.py), now fixed. Confirmed directly: ran ctags against cpython/ceval.c, PyEval_GetLocals(void)'s raw signature field is literally the text '(void)'. GitGalaxy and tree-sitter both already special-case C's explicit empty-parameter-list idiom (0 real args, matching detector.py's own _count_top_level_args docstring) -- _count_ctags_signature_params did not, splitting '(void)' into one non-empty segment and counting it as 1 real parameter (the same class of bug its own docstring already describes fixing twice for Python's trailing-comma and bare * / marker cases). Added 'void' to the segment-exclusion set alongside the existing '*'/'/'/'**' -- verified fix: _count_ctags_signature_params('(void)') now returns 0. This corpus (cpython) uses the (void) idiom extremely heavily, plausibly explaining most/all of the 104 occurrences; not independently re-verified beyond the sample, but the mechanism is unconditional (any '(void)' signature was miscounted the same way, corpus-wide) so high confidence it generalizes. No GitHub issue needed -- fixed directly in this same commit, not a repo-code defect. @@ -194,7 +194,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `cobol` function existence: ctags agree, GitGalaxy differ -*2-vs-1 -- 1316 occurrences as of 2026-08-29T21:16:48Z* +*2-vs-1 -- 1316 occurrences as of 2026-08-29T21:43:07Z* **Verdict** (by Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep, 2026-08-28): > Re-verified 2026-08-28 after the corpus grew ~6x for this language (53 -> 308 files, issue-#4 provenance audit / language-crucible v1.1.0): the count grew from 133 to 773 (gather_language name-diff; ledger's own capped-sample count differs slightly, same shape) without a status change, per merge_and_save()'s normal behavior -- re-checked rather than assumed, per how_to_investigate_a_discrepancy.md's 'When a validated shape's count changes a lot'. Confirmed: still the single already-documented mechanism (ctags' Cobol parser tags ANY period-terminated word as a paragraph, kind 'p', regardless of COBOL division or what role the period plays), now generalizing to syntax shapes the smaller corpus never exercised. Full accounting, zero unexplained residual: scope terminators like END-IF./END-PERFORM. (original 2026-08-19 finding); IDENTIFICATION/ENVIRONMENT/DATA DIVISION section/paragraph headers like WORKING-STORAGE SECTION./CONFIGURATION SECTION./LINKAGE SECTION./AUTHOR. (confirmed 2026-08-28, now the single largest contributor -- 564/773 in a corpus-wide name-tally); embedded-SQL qualified-column periods like COMMERCIAL.POLICYNUMBER inside an EXEC SQL...END-EXEC block, where the period is a table/column separator, not a COBOL statement terminator (confirmed 2026-08-28 via direct ctags run on cics-genapp/lgipdb01.cbl -- POLICY/COMMERCIAL/MOTOR tagged as paragraphs from FROM POLICY,COMMERCIAL and MOTOR.POLICYNUMBER clauses). GitGalaxy correctly excludes all three via its reserved-word shield and division/section awareness. Issue #1892's hyphenated-verb fix remains fully effective -- a corpus-wide name tally of every 'missing from GitGalaxy' occurrence found zero real paragraph names, only the three ctags-side false-positive shapes above. Full cross-reference: tests/tools/ctags_reader.py's cobol notes. @@ -214,7 +214,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `cobol` class existence: GitGalaxy agree, ctags differ -*2-vs-1 -- 6 occurrences as of 2026-08-29T21:16:48Z* +*2-vs-1 -- 6 occurrences as of 2026-08-29T21:43:07Z* **Verdict** (by Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep, 2026-08-28): > Confirmed: GitGalaxy correct on all 6, two independent confirmed ctags 'program' (P) kind limitations, zero unexplained residual. (1) 5/6 (aws-mainframe-modernization-carddemo's COACTUPC/COACTVWC/COCRDLIC/COCRDSLC/COCRDUPC): PROGRAM-ID. and the program name are written on two separate lines (a legitimate, common COBOL style) -- `ctags -x --language-force=Cobol --kinds-Cobol=P` emits zero tags at all for this shape, confirmed directly on all 5 real occurrences in the corpus, not a sample. GitGalaxy's class_start matches the name regardless of which line it's on. (2) 1/6 (gnucobol/CBL_OC_DUMP.cob): ctags truncates the program name at its first underscore, tagging bare 'CBL' instead of 'CBL_OC_DUMP' -- confirmed via direct ctags run, compared against a hyphen-named sibling (cics-genapp's LGACDB01) tagged correctly in full in the same run. Same underscore-intolerant identifier convention as Claim 12's second instance (GnuCOBOL's own YACC grammar), manifesting as truncation instead of total rejection here. GitGalaxy's class_start correctly captures underscores as valid identifier characters. Full writeup: docs/why_gitgalaxy_beats_ast_here.md Claim 12's third instance; cross-referenced in tests/tools/ctags_reader.py's cobol notes. @@ -230,7 +230,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `cobol` class existence: ctags agree, GitGalaxy differ -*2-vs-1 -- 2 occurrences as of 2026-08-29T21:16:48Z* +*2-vs-1 -- 2 occurrences as of 2026-08-29T21:43:07Z* **Verdict** (by gemini-3.1-pro-high (agy), dispatched via tri-comparison-ledger-sweep, self-corrected and reviewed by claude-sonnet-5, 2026-08-19): > Confirmed real GitGalaxy pipeline defect, generalizes to all 19 occurrences. ctags correctly tags PROGRAM-ID as a program (kind P). Initial framing (that ctags is over-matching/GitGalaxy is semantically correct to return null, since PROGRAM-ID is not an OOP class) turned out to be wrong once the actual pipeline was traced -- GitGalaxy's own cobol class_start regex ALREADY matches PROGRAM-ID correctly and identically to ctags (verified directly: LANGUAGE_DEFINITIONS["cobol"]["rules"]["class_start"] matches "PROGRAM-ID. BANKDATA." at cics-banking-sample-application-cbsa/BANKDATA.cbl:35, capturing "BANKDATA", exactly matching ctags' own reading). The null is produced further downstream: gitgalaxy/core/detector.py's _CLASS_START_NAMED_EXTRACTION_LANGS allowlist (added by epic #1295, closed 2026-08-12, 11/13 languages) gates which languages' own class_start regex is used for named-entity extraction; cobol was never added (not decided out like css/html, simply never in scope since #1295's verification method requires a tree-sitter grammar cobol lacks). Every language missing from that allowlist falls through to a hardcoded generic fallback regex (class|struct|interface|trait|enum) that cannot structurally match COBOL syntax at all -- so the correct class_start match is computed internally (feeding the numeric risk-signal count) but discarded before named-entity output. Same bug class #1295 already fixed for 11 other languages. This corrects and supersedes issue #1858's original diagnosis (which claimed the regex itself never matches -- it does); posted a correcting comment on #1858 rather than filing a duplicate. Investigated via gemini-3.1-pro-high (agy); the dispatched agent itself caught and corrected Gemini's initial (semantically-plausible but pipeline-unverified) verdict by tracing the real extraction code, then Claude independently re-verified both the regex match and the allowlist gap directly against source before applying. credit_tools=[ctags] applied: ctags' claim (19 real PROGRAM-ID classes) is fully confirmed real, and GitGalaxy's non-corroboration is a confirmed pipeline bug in GitGalaxy, not an open question about ctags -- this is the textbook credit_tools case per tri_comparison_ledger.py's own VERIFIED ADJUSTMENTS docstring. @@ -244,10 +244,10 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `cpp` function existence: tree-sitter agree, GitGalaxy, ctags differ -*2-vs-1 -- 200 occurrences as of 2026-08-29T21:16:53Z* +*2-vs-1 -- 136 occurrences as of 2026-08-29T21:43:10Z* **Verdict** (by Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-21T00:00:00Z): -> UPDATED: count grew (98 -> 194) as a direct, correct consequence of the OPCODE-macro fix documented in the sibling agree[gitgalaxy,tree_sitter]_vs[ctags] entry -- tree-sitter's OWN OPCODE-family misparse (godot/gdscript_vm.cpp's bytecode-dispatch macro) was previously MASKED by GitGalaxy sharing the identical mistake (both wrong == they 'agreed', landing in the other shape instead of this one). Now that GitGalaxy correctly excludes known macro invocations, tree-sitter's own grammar limitation on this exact pattern is cleanly isolated here instead, confirmed via the shape's own examples (repeated `OPCODE` entries at godot/gdscript_vm.cpp, tree_sitter line set, ctags/gitgalaxy both None). The remaining, previously-documented causes (bare control-flow-keyword/`void` hallucination, conversion-operator naming-suffix convention) are unchanged and still contribute the bulk of the original 98. No credit/debit applies -- tree-sitter is the one that's wrong here, alone. +> UPDATED: count grew (98 -> 194) as a direct, correct consequence of the OPCODE-macro fix documented in the sibling agree[gitgalaxy,tree_sitter]_vs[ctags] entry -- tree-sitter's OWN OPCODE-family misparse (godot/gdscript_vm.cpp's bytecode-dispatch macro) was previously MASKED by GitGalaxy sharing the identical mistake (both wrong == they 'agreed', landing in the other shape instead of this one). Now that GitGalaxy correctly excludes known macro invocations, tree-sitter's own grammar limitation on this exact pattern is cleanly isolated here instead, confirmed via the shape's own examples (repeated `OPCODE` entries at godot/gdscript_vm.cpp, tree_sitter line set, ctags/gitgalaxy both None). The remaining, previously-documented causes (bare control-flow-keyword/`void` hallucination, conversion-operator naming-suffix convention) are unchanged and still contribute the bulk of the original 98. No credit/debit applies -- tree-sitter is the one that's wrong here, alone. 2026-08-29 follow-up (#2455): the tree_sitter_accuracy_audit / tri_comparison_gatherer shared tree-sitter reader now canonicalizes C++ conversion-operator and destructor names to GitGalaxy's own convention -- `Variant::operator String() const` -> `Variant::operator String`, `operator BitField() const` -> `operator BitField`, and the `_FORCE_INLINE_`-macro-mangled `~Variant` / `operator T` forms recovered from the sibling ERROR node -- and treats a small tree-sitter-cpp ERROR node (a function-like macro with no visible definition immediately before a special member) as a blind-spot region, promoting GitGalaxy's correct extraction there to ground truth. cpp function precision 95.0% -> 100.0%, extra_functions 68 -> 0. That resolved the 'conversion-operator naming-suffix convention' cause named above (~55 occurrences moved into GitGalaxy+tree-sitter consensus, count ~194 -> ~136). The residual is the genuine OPCODE-macro and bare-control-flow-keyword / bare-`void` hallucinations, still tree-sitter's alone. | file | name | GitGalaxy | tree-sitter | ctags | |---|---|---|---|---| @@ -262,29 +262,9 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest | godot/gdscript_vm.cpp | `OPCODE` | *(n/a)* | 1034 | *(n/a)* | | godot/gdscript_vm.cpp | `OPCODE` | *(n/a)* | 1068 | *(n/a)* | -### ✅ `cpp` function existence: GitGalaxy, ctags agree, tree-sitter differ - -*2-vs-1 -- 66 occurrences as of 2026-08-29T21:16:53Z* - -**Verdict** (by Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-21T00:00:00Z): -> Confirmed real tree-sitter-cpp grammar limitation, not a GitGalaxy or ctags defect. Every sampled case (GDScriptFunction::call, Main::setup, Main::setup2, Main::start, Object::Connection::operator Variant) is a large, complex function -- GDScriptFunction::call in particular (godot/gdscript_vm.cpp:499) is a bytecode interpreter's main dispatch loop using GNU 'labels as values' computed-goto syntax (`&&OPCODE_LABEL`) via the same OPCODES_TABLE/OPCODE macro family documented in the sibling agree[gitgalaxy,tree_sitter]_vs[ctags] entry -- a non-standard GNU extension tree-sitter-cpp's grammar does not support, which plausibly causes a parse error cascade that loses the enclosing function_definition node entirely rather than just misreading the body. ctags and GitGalaxy both correctly find and name these functions regardless of body content, since neither one needs to fully parse the function body to recognize its signature. tree-sitter's non-detection is a confirmed limitation in tree-sitter itself -- but no credit_tools adjustment applies: ctags and GitGalaxy are already a 2-of-3 AGREEING PAIR on this shape (agreeing_tools has 2 members), which already satisfies reconcile_symbols' own `len(present) >= 2` precision-credit condition naturally with no ledger adjustment needed. credit_tools exists for a LONE, single-tool claim (agreeing_tools with exactly 1 member) the base algorithm can't otherwise corroborate -- applying it to an already-mutually-corroborating pair would double-count (confirmed: this exact mistake briefly pushed ctags' precision past 100% before being caught and reverted in this same session). - -| file | name | GitGalaxy | tree-sitter | ctags | -|---|---|---|---|---| -| godot/gdscript_vm.cpp | `GDScriptFunction::call` | 499 | *(n/a)* | 499 | -| godot/main.cpp | `Main::setup` | 1027 | *(n/a)* | 1027 | -| godot/main.cpp | `Main::setup2` | 3007 | *(n/a)* | 3007 | -| godot/main.cpp | `Main::start` | 3987 | *(n/a)* | 3987 | -| godot/object.cpp | `Object::Connection::operator Variant` | 108 | *(n/a)* | 108 | -| godot/object.h | `operator=` | 970 | *(n/a)* | 971 | -| godot/object.h | `RequiredResult` | 976 | *(n/a)* | 977 | -| godot/object.h | `RequiredResult` | 985 | *(n/a)* | 986 | -| godot/object.h | `RequiredResult` | 994 | *(n/a)* | 995 | -| godot/object.h | `RequiredResult` | 1003 | *(n/a)* | 1004 | - ### ✅ `cpp` function existence: ctags agree, GitGalaxy, tree-sitter differ -*2-vs-1 -- 30 occurrences as of 2026-08-29T21:16:53Z* +*2-vs-1 -- 30 occurrences as of 2026-08-29T21:43:10Z* **Verdict** (by Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-21T00:00:00Z): > Confirmed ctags-only limitation: ctags parses INSIDE C++ macro DEFINITION bodies as if they were real, already-expanded code. godot/object.h's GDCLASS/_FORCE_INLINE_-based macros (`#define GDCLASS(m_class, m_inherits) ... _FORCE_INLINE_ bool (Object::*_get_get() const)(...) {...} ...`) never run as written -- they only produce real code once expanded at a `GDCLASS(SomeClass, Base)` call site elsewhere -- but ctags tags `_get_get`/`_get_set`/`_get_bind_methods`/`_get_bind_compatibility_methods`/`_get_notification`/`_get_property_can_revert`/`_get_property_get_revert`/`_get_validate_property`/`_get_get_property_list` (all 9 sampled cases, all from this same macro) as if they were ordinary member functions. Neither GitGalaxy nor tree-sitter are fooled by this. Documented in ctags_reader.py's KIND MAPS section (cpp bullet) rather than fixed -- this is ctags' own parser behavior, nothing in this repo's tooling can distinguish a macro-definition body from real code without reimplementing preprocessing. @@ -302,48 +282,49 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest | godot/object.h | `_get_validate_property` | *(n/a)* | *(n/a)* | 564 | | godot/object.h | `operator Ref` | *(n/a)* | *(n/a)* | 1036 | -### ✅ `cpp` function existence: GitGalaxy agree, tree-sitter, ctags differ +### ✅ `cpp` function existence: GitGalaxy, tree-sitter agree, ctags differ -*2-vs-1 -- 12 occurrences as of 2026-08-29T21:16:53Z* +*2-vs-1 -- 21 occurrences as of 2026-08-29T21:43:10Z* **Verdict** (by Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-21T00:00:00Z): -> Compound shape, three distinct causes confirmed via source, not one. (1) Most of the sample (OPCODE_WHILE x2, OPCODE_SWITCH, OPCODE) is the same GG+tree-sitter shared macro-misparse family documented in the sibling agree[gitgalaxy,tree_sitter]_vs[ctags] entry (godot/gdscript_vm.cpp's OPCODE/OPCODE_WHILE/OPCODE_SWITCH dispatch macros) -- here landing as 'GitGalaxy alone' because tree-sitter's own error recovery on this repeated macro pattern isn't fully deterministic across every occurrence, not because the underlying cause differs. GitGalaxy is WRONG for this portion (same debit as the sibling entry, not double-counted here since debit_tools is per-shape). (2) `attribute_buffer_applier_factories_`/`m_draggingState`/`std::thread` are a separate, real GitGalaxy FALSE POSITIVE: a lambda passed as a constructor argument or member-initializer-list entry (`m_draggingState([this]() {...}),`, `std::thread([...]() {...}).detach();`) is misread as a function definition. Filed as https://github.com/squid-protocol/gitgalaxy/issues/2013 . (3) `Variant::operator ::RID`/`Variant::operator ::AABB`/`Variant::operator Object *` are real functions GitGalaxy correctly finds (confirmed via godot/variant.cpp source) that didn't rank-match ctags/tree-sitter's own readings of the same functions by exact name string -- a residual, low-priority naming-comparison edge case (global-scope `::`-prefixed conversion-operator return types) not chased further in this pass. No credit/debit applied given the mixed, three-cause nature of this shape. +> UPDATED after a real production fix. The OPCODE-family shared mistake documented in this shape's prior verdict (~96 of the original 105 occurrences) is now FIXED: GitGalaxy's func_start (and by the same mechanism, C's) now scans each file for `#define NAME(...)` function-like macro definitions and excludes any match whose captured name is a known macro -- the exact same fact universal-ctags itself already used (confirmed via a direct `ctags -f -` run: ctags tags `OPCODE` only once, at its own #define line, kind `d`, and produces zero tags at any invocation site -- it isn't smarter about the invocation's shape, it just already knows the name is a macro and never re-tags it). This also fixed the already-documented C `RICHCMP_WRAPPER`/`SLOT0`/`SLOT1`/`SLOT1BINFULL`/`DICT___REVERSED___METHODDEF` false positives as a bonus (same mechanism, `lang_id in ("c", "cpp")`). Verified via 3 repeated full-corpus scans producing byte-identical function lists, the full extraction gauntlet, and both golden masters re-blessed. The remaining residual (9 occurrences, previously miscounted as more of the same shared mistake in the earlier verdict's blanket debit -- corrected here) is a DIFFERENT, unrelated finding: GitGalaxy and tree-sitter are CORRECT here, ctags is wrong. Two sub-patterns confirmed: (1) the already-documented macro-as-return-type-prefix pattern (`IFACEMETHODIMP_(void)` immediately before the real `FancyZones::Run() noexcept` -- ctags tags the macro invocation itself and loses the real name, powertoys/FancyZones.cpp, 4 occurrences); (2) a newly-confirmed, NOT yet root-caused ctags miss on an ordinary virtual method with no macro involvement at all (`virtual RID mesh_create_from_surfaces(const Vector &p_surfaces, int p_blend_shape_count = 0) override { ... }`, godot/rendering_server_default.h -- ctags produces zero tags anywhere near this real method, 5 occurrences). No credit_tools adjustment applies: GitGalaxy and tree-sitter are already a 2-of-3 agreeing pair here, which already satisfies reconcile_symbols' own natural precision-credit condition. 2026-08-29 follow-up (#2455): the tree_sitter_accuracy_audit / tri_comparison_gatherer shared tree-sitter reader now canonicalizes C++ conversion-operator and destructor names to GitGalaxy's own convention -- `Variant::operator String() const` -> `Variant::operator String`, `operator BitField() const` -> `operator BitField`, and the `_FORCE_INLINE_`-macro-mangled `~Variant` / `operator T` forms recovered from the sibling ERROR node -- and treats a small tree-sitter-cpp ERROR node (a function-like macro with no visible definition immediately before a special member) as a blind-spot region, promoting GitGalaxy's correct extraction there to ground truth. cpp function precision 95.0% -> 100.0%, extra_functions 68 -> 0. Several conversion operators GitGalaxy and tree-sitter now agree on (ctags names them without the `operator ` prefix or misses the `_FORCE_INLINE_`-macro forms) shifted into this shape, count ~9 -> ~21 -- still GitGalaxy+tree-sitter correct, ctags limited. | file | name | GitGalaxy | tree-sitter | ctags | |---|---|---|---|---| -| godot/object.h | `operator Ref` | 1035 | *(n/a)* | *(n/a)* | -| godot/variant.cpp | `Variant::operator ::RID` | 2002 | *(n/a)* | *(n/a)* | -| godot/variant.cpp | `Variant::operator Vector` | 2219 | *(n/a)* | *(n/a)* | -| godot/variant.cpp | `Variant::operator Vector` | 2229 | *(n/a)* | *(n/a)* | -| godot/variant.cpp | `Variant::operator Vector` | 2247 | *(n/a)* | *(n/a)* | -| godot/variant.cpp | `Variant::operator Vector` | 2266 | *(n/a)* | *(n/a)* | -| godot/variant.cpp | `Variant::operator Vector` | 2283 | *(n/a)* | *(n/a)* | -| godot/variant.cpp | `Variant::operator ::AABB` | 1882 | *(n/a)* | *(n/a)* | -| godot/variant.cpp | `Variant::operator Object *` | 2024 | *(n/a)* | *(n/a)* | -| godot/variant.h | `operator BitField` | 478 | *(n/a)* | *(n/a)* | +| godot/object.h | `operator Ref` | 1035 | 1036 | *(n/a)* | +| godot/rendering_server_default.h | `mesh_create_from_surfaces` | 363 | 363 | *(n/a)* | +| godot/rendering_server_default.h | `material_create_from_shader` | 324 | 324 | *(n/a)* | +| godot/rendering_server_default.h | `shader_create` | 278 | 278 | *(n/a)* | +| godot/rendering_server_default.h | `texture_create_from_native_handle` | 222 | 222 | *(n/a)* | +| godot/rendering_server_default.h | `redraw_request` | 111 | 111 | *(n/a)* | +| godot/variant.cpp | `Variant::operator ::RID` | 2002 | 2002 | *(n/a)* | +| godot/variant.cpp | `Variant::operator Vector` | 2219 | 2219 | *(n/a)* | +| godot/variant.cpp | `Variant::operator Vector` | 2229 | 2229 | *(n/a)* | +| godot/variant.cpp | `Variant::operator Vector` | 2247 | 2247 | *(n/a)* | -### ✅ `cpp` function existence: GitGalaxy, tree-sitter agree, ctags differ +### ✅ `cpp` function existence: GitGalaxy, ctags agree, tree-sitter differ -*2-vs-1 -- 9 occurrences as of 2026-08-29T21:16:53Z* +*2-vs-1 -- 14 occurrences as of 2026-08-29T21:43:10Z* **Verdict** (by Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-21T00:00:00Z): -> UPDATED after a real production fix. The OPCODE-family shared mistake documented in this shape's prior verdict (~96 of the original 105 occurrences) is now FIXED: GitGalaxy's func_start (and by the same mechanism, C's) now scans each file for `#define NAME(...)` function-like macro definitions and excludes any match whose captured name is a known macro -- the exact same fact universal-ctags itself already used (confirmed via a direct `ctags -f -` run: ctags tags `OPCODE` only once, at its own #define line, kind `d`, and produces zero tags at any invocation site -- it isn't smarter about the invocation's shape, it just already knows the name is a macro and never re-tags it). This also fixed the already-documented C `RICHCMP_WRAPPER`/`SLOT0`/`SLOT1`/`SLOT1BINFULL`/`DICT___REVERSED___METHODDEF` false positives as a bonus (same mechanism, `lang_id in ("c", "cpp")`). Verified via 3 repeated full-corpus scans producing byte-identical function lists, the full extraction gauntlet, and both golden masters re-blessed. The remaining residual (9 occurrences, previously miscounted as more of the same shared mistake in the earlier verdict's blanket debit -- corrected here) is a DIFFERENT, unrelated finding: GitGalaxy and tree-sitter are CORRECT here, ctags is wrong. Two sub-patterns confirmed: (1) the already-documented macro-as-return-type-prefix pattern (`IFACEMETHODIMP_(void)` immediately before the real `FancyZones::Run() noexcept` -- ctags tags the macro invocation itself and loses the real name, powertoys/FancyZones.cpp, 4 occurrences); (2) a newly-confirmed, NOT yet root-caused ctags miss on an ordinary virtual method with no macro involvement at all (`virtual RID mesh_create_from_surfaces(const Vector &p_surfaces, int p_blend_shape_count = 0) override { ... }`, godot/rendering_server_default.h -- ctags produces zero tags anywhere near this real method, 5 occurrences). No credit_tools adjustment applies: GitGalaxy and tree-sitter are already a 2-of-3 agreeing pair here, which already satisfies reconcile_symbols' own natural precision-credit condition. +> Confirmed real tree-sitter-cpp grammar limitation, not a GitGalaxy or ctags defect. Every sampled case (GDScriptFunction::call, Main::setup, Main::setup2, Main::start, Object::Connection::operator Variant) is a large, complex function -- GDScriptFunction::call in particular (godot/gdscript_vm.cpp:499) is a bytecode interpreter's main dispatch loop using GNU 'labels as values' computed-goto syntax (`&&OPCODE_LABEL`) via the same OPCODES_TABLE/OPCODE macro family documented in the sibling agree[gitgalaxy,tree_sitter]_vs[ctags] entry -- a non-standard GNU extension tree-sitter-cpp's grammar does not support, which plausibly causes a parse error cascade that loses the enclosing function_definition node entirely rather than just misreading the body. ctags and GitGalaxy both correctly find and name these functions regardless of body content, since neither one needs to fully parse the function body to recognize its signature. tree-sitter's non-detection is a confirmed limitation in tree-sitter itself -- but no credit_tools adjustment applies: ctags and GitGalaxy are already a 2-of-3 AGREEING PAIR on this shape (agreeing_tools has 2 members), which already satisfies reconcile_symbols' own `len(present) >= 2` precision-credit condition naturally with no ledger adjustment needed. credit_tools exists for a LONE, single-tool claim (agreeing_tools with exactly 1 member) the base algorithm can't otherwise corroborate -- applying it to an already-mutually-corroborating pair would double-count (confirmed: this exact mistake briefly pushed ctags' precision past 100% before being caught and reverted in this same session). 2026-08-29 follow-up (#2455): the tree_sitter_accuracy_audit / tri_comparison_gatherer shared tree-sitter reader now canonicalizes C++ conversion-operator and destructor names to GitGalaxy's own convention -- `Variant::operator String() const` -> `Variant::operator String`, `operator BitField() const` -> `operator BitField`, and the `_FORCE_INLINE_`-macro-mangled `~Variant` / `operator T` forms recovered from the sibling ERROR node -- and treats a small tree-sitter-cpp ERROR node (a function-like macro with no visible definition immediately before a special member) as a blind-spot region, promoting GitGalaxy's correct extraction there to ground truth. cpp function precision 95.0% -> 100.0%, extra_functions 68 -> 0. `Object::Connection::operator Variant` (doubly-qualified conversion operator) is now in consensus. The residual (`GDScriptFunction::call`, `Main::setup*`) is the computed-goto / OPCODES_TABLE parse cascade that loses the whole `function_definition` node -- GitGalaxy and ctags both still correct, count ~66 -> ~14. | file | name | GitGalaxy | tree-sitter | ctags | |---|---|---|---|---| -| godot/rendering_server_default.h | `mesh_create_from_surfaces` | 363 | 363 | *(n/a)* | -| godot/rendering_server_default.h | `material_create_from_shader` | 324 | 324 | *(n/a)* | -| godot/rendering_server_default.h | `shader_create` | 278 | 278 | *(n/a)* | -| godot/rendering_server_default.h | `texture_create_from_native_handle` | 222 | 222 | *(n/a)* | -| godot/rendering_server_default.h | `redraw_request` | 111 | 111 | *(n/a)* | -| powertoys/FancyZones.cpp | `FancyZones::OnKeyDown` | 473 | 473 | *(n/a)* | -| powertoys/FancyZones.cpp | `FancyZones::Run` | 214 | 214 | *(n/a)* | -| powertoys/FancyZones.cpp | `FancyZones::Destroy` | 286 | 286 | *(n/a)* | -| powertoys/FancyZones.cpp | `FancyZones::VirtualDesktopChanged` | 303 | 303 | *(n/a)* | +| godot/gdscript_vm.cpp | `GDScriptFunction::call` | 499 | *(n/a)* | 499 | +| godot/main.cpp | `Main::setup` | 1027 | *(n/a)* | 1027 | +| godot/main.cpp | `Main::setup2` | 3007 | *(n/a)* | 3007 | +| godot/main.cpp | `Main::start` | 3987 | *(n/a)* | 3987 | +| godot/object.h | `operator=` | 970 | *(n/a)* | 971 | +| godot/object.h | `RequiredResult` | 976 | *(n/a)* | 977 | +| godot/object.h | `RequiredResult` | 985 | *(n/a)* | 986 | +| godot/object.h | `RequiredResult` | 994 | *(n/a)* | 995 | +| godot/object.h | `RequiredResult` | 1003 | *(n/a)* | 1004 | +| godot/object.h | `RequiredResult` | 1012 | *(n/a)* | 1013 | ### ✅ `cpp` class existence: ctags agree, GitGalaxy, tree-sitter differ -*2-vs-1 -- 2 occurrences as of 2026-08-29T21:16:53Z* +*2-vs-1 -- 2 occurrences as of 2026-08-29T21:43:10Z* **Verdict** (by Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-21T00:00:00Z): > Not a real existence disagreement -- a template-argument name-formatting difference in the comparison tooling. `godot/variant.h`'s `HashMapComparatorDefault` and `is_zero_constructible` are template CLASS specializations; GitGalaxy and tree-sitter both read the name WITH its template argument baked in (matching the instantiation as written in source), while ctags strips the `<...>` template-argument suffix from its own class tag name. All three tools found the exact same class definition at the exact same line -- confirmed via the sibling agree[gitgalaxy,tree_sitter]_vs[ctags] entry, which is the same pair of classes from the opposite direction. Low magnitude (2 occurrences) -- not chased to a code fix in this pass, but a plausible future micro-fix would strip a trailing `<...>` from gg/tree-sitter's class name before matching, mirroring how ctags_reader.py's operator-name normalization already handles a similar formatting mismatch. @@ -355,7 +336,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `cpp` class existence: GitGalaxy, tree-sitter agree, ctags differ -*2-vs-1 -- 2 occurrences as of 2026-08-29T21:16:53Z* +*2-vs-1 -- 2 occurrences as of 2026-08-29T21:43:10Z* **Verdict** (by Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-21T00:00:00Z): > Same template-argument name-formatting difference as the sibling agree[ctags]_vs[gitgalaxy,tree_sitter] class entry, viewed from the opposite direction -- `HashMapComparatorDefault`/`is_zero_constructible` (ctags' bare names) vs. `HashMapComparatorDefault`/`is_zero_constructible` (GitGalaxy/tree-sitter's names, template argument included). Not a real disagreement about whether these classes exist -- see the sibling entry for the full explanation. @@ -369,7 +350,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `csharp` function existence: GitGalaxy, ctags agree, tree-sitter differ -*2-vs-1 -- 271 occurrences as of 2026-08-29T21:16:58Z* +*2-vs-1 -- 271 occurrences as of 2026-08-29T21:43:13Z* **Verdict** (by Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-21T18:13:27Z): > Confirmed: this shape is entirely tree-sitter-c-sharp's own known parse-error cascade in roslyn/LanguageParser.cs, already root-caused and evidence-backed as Claim 3 in docs/why_gitgalaxy_beats_ast_here.md (issues #1427/#1567). A syntax construct at line ~5198 triggers a parse error whose recovery fails to resynchronize for the rest of the (14,680-line) file -- tree.root_node.type itself becomes ERROR, so tree-sitter's own recall for this file collapses to near-zero past that point. GitGalaxy's regex and ctags' text-based parser are both unaffected and correctly find these functions (all sampled: ParseVariableDeclarator, GetPrecedence, ParsePrimaryExpression, ScanType x3 -- real, ordinary private methods, exact line-number agreement between ctags and gitgalaxy on every one). This is the tri-comparison ledger's own version of a phenomenon already fully diagnosed for the 2-way tree_sitter_accuracy_audit.py tool via its cascade_promotable logic -- that logic was never ported to tri_comparison_gatherer.py/tri_comparison_reconcile.py, which is why this shows as an unvalidated ledger shape despite being a known, closed question. Not a GitGalaxy or ctags defect. CORRECTION (2026-08-21): credit_tools was originally set here, but this is wrong -- these tools already mutually corroborate each other (2-of-3 agreement), so their occurrences were already counted in base precision before any credit was applied. Adding credit on top double-counted them, pushing gitgalaxy's func_precision matched_consensus past its own total_slots (1297/965, >100%, a real, visible chart-rendering bug). credit_tools is only valid for a shape where a tool is completely ALONE and otherwise uncorroborated (e.g. this language's agree[gitgalaxy]_vs[ctags,tree_sitter] shape), never for an already-mutually-agreeing pair. Reset to empty; the validation itself (which tool is factually correct here) is unaffected and remains correct. @@ -389,7 +370,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `csharp` function existence: GitGalaxy, tree-sitter agree, ctags differ -*2-vs-1 -- 108 occurrences as of 2026-08-29T21:16:58Z* +*2-vs-1 -- 108 occurrences as of 2026-08-29T21:43:13Z* **Verdict** (by Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-21T18:13:44Z): > Confirmed real ctags parser limitations, not a GitGalaxy or tree-sitter defect, via direct source reading and raw `ctags -x` runs against the flagged files. Two distinct mechanisms: (1) Local/nested functions -- roslyn/CSharpCompilation.cs's `validateSignature` (nested inside a larger method) and `isSupportedType` (a `static bool isSupportedType(...)` local function) are both structurally invisible to ctags, whose csharp kind map is only 'm' (top-level methods; "C# has no free functions" per ctags_reader.py's own comment) with no local-function concept at all -- this generalizes to every local function in the corpus, not just the sampled two. (2) Complex-signature misses and overload collisions on ordinary top-level private methods -- `FindEntryPoint` (nullable return/param types plus a generic `out` parameter) and `GetSourceDeclarationDiagnostics` (5 params, two with default values, one a `Func<...>` generic delegate) get no ctags tag at all, confirmed via `ctags -x` showing tags immediately before/after but not at their own lines -- isolated misses, not a wider blind region (unlike the tree-sitter cascade in the sibling ledger shape). `ReportUnusedImports` has two overloads at different lines; ctags tags only the first, silently dropping the second. Documented in tests/tools/ctags_reader.py's csharp KIND MAPS bullet. Credited on the strength of mechanism (1) generalizing structurally to the whole shape (ctags cannot see ANY local function, by construction) even though only ~5 of 107 occurrences were individually read -- mechanism (2)'s complex-signature misses are a smaller, less certain-to-generalize contributor to the same shape but point the same direction (ctags-side, not GitGalaxy/tree-sitter). CORRECTION (2026-08-21): credit_tools was originally set here, but this is wrong -- these tools already mutually corroborate each other (2-of-3 agreement), so their occurrences were already counted in base precision before any credit was applied. Adding credit on top double-counted them, pushing gitgalaxy's func_precision matched_consensus past its own total_slots (1297/965, >100%, a real, visible chart-rendering bug). credit_tools is only valid for a shape where a tool is completely ALONE and otherwise uncorroborated (e.g. this language's agree[gitgalaxy]_vs[ctags,tree_sitter] shape), never for an already-mutually-agreeing pair. Reset to empty; the validation itself (which tool is factually correct here) is unaffected and remains correct. @@ -409,7 +390,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `csharp` function existence: GitGalaxy agree, tree-sitter, ctags differ -*2-vs-1 -- 46 occurrences as of 2026-08-29T21:16:58Z* +*2-vs-1 -- 46 occurrences as of 2026-08-29T21:43:13Z* **Verdict** (by Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-21T21:45:53Z): > Mixed shape, confirmed via a full name-diff against gather_language('csharp') rather than the capped sample alone: 44 of 48 are genuine local (nested) functions inside roslyn/LanguageParser.cs's tree-sitter parse-error cascade region (line >= ~5198, same mechanism as the sibling agree[ctags,gitgalaxy]_vs[tree_sitter] shape / Claim 3) -- tree-sitter is fully blind there, and ctags additionally has no concept of a local/nested function at all (only top-level 'm' method tags), so GitGalaxy is the only tool that can see these. The remaining 2 are genuine GitGalaxy false positives with a confirmed, unrelated root cause: roslyn/CSharpCompilation.cs:2282's GetWellKnownType( (a call, no declaration exists anywhere in the file) and roslyn/LanguageParser.cs:2338's this.EatToken( (a call inside a switch-expression arm) are both mis-captured because func_start's return-type-loop character class allows unbalanced parens/commas in a single 'token', letting it swallow a real call-expression fragment as if it were part of a return type and land on the wrong identifier as the 'function name'. Filed as GitHub issue #2035 with root cause and fix direction; fix in progress in an isolated worktree as of this writing. Not crediting/debiting any tool on this shape since it's genuinely mixed (44 real local-function finds, 2 real false positives) -- re-visit once #2035 merges and re-run to confirm the shape drops to 44 (or fewer, if further false positives of the same class surface once #2035's fix generalizes across the full corpus). FOLLOW-UP (2026-08-21, post-#2035/#2036 re-verification): re-checked this shape with a full, uncapped corpus diff rather than the capped example sample. The 2 originally-confirmed false positives (GetWellKnownType, this.EatToken) are gone as expected. However 2 MORE confirmed false positives remain, distinct mechanisms from #2035's fix: CSharpCompilation.cs's `ref mdName, ..., CreateReflectionTypeNotFoundError(` (a real call site mistaken for a declaration because `ref` is a legitimate Branch A modifier keyword when used in a real declaration, but here it's a call-argument prefix) and LanguageParser.cs:2336's `_syntaxFactory.TypeConstraint(this.ParseType())` (a ternary `?` operator consumed by the return-type loop's nullable-type-marker allowance, same general shape as this.EatToken but not covered by #2035's specific fix). Filed as issue #2054. This shape remains genuinely mixed (the overwhelming majority are real cascade-region local functions, but a small residual of false positives keeps surfacing) -- still correctly left uncredited pending #2054. CLOSED (2026-08-21): this shape is now fully clean. Both remaining false-positive mechanisms (issue #2054: bare-comma call-argument absorption, ternary-? consumption, and the nested-generic-return-type regression its own fix introduced and #2061 corrected) are fixed and merged. A full uncapped re-diff confirms all 44 remaining occurrences are genuine local functions inside LanguageParser.cs's tree-sitter parse-error cascade region (Claim 3), zero non-cascade residue. credit_tools is now correctly applicable -- gitgalaxy is alone on this shape (no tool to share the credit-double-counting error this session already found and fixed on the sibling agree[ctags,gitgalaxy]/agree[gitgalaxy,tree_sitter] shapes with), and every one of its 44 claims here is confirmed real with the reason ctags (no local-function concept) and tree-sitter (parse cascade) don't corroborate it being a confirmed limitation in THEM, not an open question about GitGalaxy. @@ -429,7 +410,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `csharp` class existence: GitGalaxy, ctags agree, tree-sitter differ -*2-vs-1 -- 9 occurrences as of 2026-08-29T21:16:58Z* +*2-vs-1 -- 9 occurrences as of 2026-08-29T21:43:13Z* **Verdict** (by Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-21T18:13:27Z): > Same root cause and file as the sibling function-existence shape (roslyn/LanguageParser.cs's tree-sitter parse-error cascade from line ~5198, Claim 3 in why_gitgalaxy_beats_ast_here.md, #1427/#1567) -- with the parse tree corrupted into a single ERROR node, tree-sitter can't resolve any class_declaration structure in the corrupted region either, including the outer LanguageParser class itself (whose body spans the entire cascade) and every enum/nested class declared inside it (VariableFlags, NameOptions, ScanTypeArgumentListKind, ScanTypeFlags, ParseTypeMode, etc.). ctags and GitGalaxy both correctly find these via text-based parsing, unaffected by tree-sitter's own AST failure. Not a GitGalaxy or ctags defect. CORRECTION (2026-08-21): credit_tools was originally set here, but this is wrong -- these tools already mutually corroborate each other (2-of-3 agreement), so their occurrences were already counted in base precision before any credit was applied. Adding credit on top double-counted them, pushing gitgalaxy's func_precision matched_consensus past its own total_slots (1297/965, >100%, a real, visible chart-rendering bug). credit_tools is only valid for a shape where a tool is completely ALONE and otherwise uncorroborated (e.g. this language's agree[gitgalaxy]_vs[ctags,tree_sitter] shape), never for an already-mutually-agreeing pair. Reset to empty; the validation itself (which tool is factually correct here) is unaffected and remains correct. @@ -448,7 +429,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `csharp` function args: GitGalaxy, tree-sitter agree, ctags differ -*2-vs-1 -- 2 occurrences as of 2026-08-29T21:16:58Z* +*2-vs-1 -- 2 occurrences as of 2026-08-29T21:43:13Z* **Verdict** (by Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-21T20:06:45Z): > Mixed shape. 1 of 4 is a confirmed genuine ctags defect: GetHashCode's tuple-parameter overload (`GetHashCode((ImmutableArray ContentHash, int Position) obj)`, 1 real param) -- ctags reports 2, mis-splitting the tuple-typed parameter's own internal comma as a second top-level parameter, the same family as the already-documented tuple-related ctags limitations in tests/tools/ctags_reader.py's csharp bullet. The other 3 (GetModifierExcludingScoped and two SetCurrentSolution rows) are NOT real tool disagreements at all -- SetCurrentSolution has 4 real overloads (1/6/4/5 real params respectively) in roslyn/Workspace.cs; every individual reading from every tool across both rows (ctags=6, gitgalaxy=1, tree_sitter=1, ctags=4, gitgalaxy=6, tree_sitter=6) independently matches SOME real overload's true count exactly when checked against the source -- the ledger shape only exists because the reconciler's rank-based pairing compared different tools' readings of DIFFERENT overloads against each other, not because any tool actually miscounted anything. Confirmed via direct line-by-line source reading of all 4 overloads. Credit gitgalaxy+tree_sitter for the 1 real ctags defect only; the shape as a whole is left without a clean credit/debit since 3 of 4 occurrences aren't a real disagreement to adjudicate. @@ -460,7 +441,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `csharp` function existence: ctags agree, GitGalaxy, tree-sitter differ -*2-vs-1 -- 1 occurrence as of 2026-08-29T21:16:58Z* +*2-vs-1 -- 1 occurrence as of 2026-08-29T21:43:13Z* **Verdict** (by Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-21T18:13:27Z): > Confirmed genuine ctags false positive, not a GitGalaxy or tree-sitter gap. roslyn/CSharpCompilation.cs:2539's real declaration is `public bool Equals((ImmutableArray ContentHash, int Position) x, (ImmutableArray ContentHash, int Position) y)` -- an Equals overload with tuple-typed parameters. ctags' lightweight C# parser misreads the return-type/tuple-parameter boundary and tags the match under the name `bool` instead of `Equals`. Documented in tests/tools/ctags_reader.py's csharp KIND MAPS bullet alongside this shape's sibling ctags limitations (local-function blindness, complex-signature misses, overload-name collision -- see the agree[gitgalaxy,tree_sitter]_vs[ctags] shape). Not crediting/debiting: a single-tool false positive from an otherwise-unaffected precision baseline, no shared-mistake mechanism applies. @@ -473,7 +454,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `css` function existence: GitGalaxy, tree-sitter agree, ctags differ -*2-vs-1 -- 3 occurrences as of 2026-08-29T21:16:59Z* +*2-vs-1 -- 3 occurrences as of 2026-08-29T21:43:15Z* **Verdict** (by Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep, 2026-08-21T00:00:00Z): > Confirmed ctags-side structural limitation, not a GitGalaxy defect. GitGalaxy's own func_start regex (language_standards.py) deliberately matches CSS at-rule keywords (@media/@supports/@container/@layer/@keyframes/@-webkit-keyframes) as the closest function-shaped construct CSS has, since CSS has no true function-definition/call syntax. tree-sitter's CSS grammar independently corroborates this: media_statement/supports_statement/keyframes_statement are real, distinct node types, already mapped 1:1 onto GitGalaxy's own at-rule set per issue #1313 (see tree_sitter_accuracy_audit.py's css NODE_MAPS entry and its media_statement/supports_statement name-extraction branches). Re-ran gather_language('css') directly against the full 4-file corpus (not just the ledger's capped 3-example sample): GitGalaxy and tree-sitter agree exactly on function count in every file (3/3 total, zero diff files) -- the agreement generalizes completely, it is not a partial/mixed sample. ctags' own FUNCTION_KIND_MAP already documents 'css': set() # no function-equivalent (ctags_reader.py) -- confirmed empty (ctags_funcs: []) across all 4 corpus files too. ctags structurally cannot tag CSS at-rules as anything function-like; it isn't wrong about a claim, it has no claim to make here at all. No new doc note needed -- both the GitGalaxy/tree-sitter convention (#1313) and the ctags limitation are already documented in code. GitGalaxy+tree-sitter's agreement is real corroboration, not a shared mistake, so no credit/debit adjustment applies; ctags' lack of a comparable slot here is already handled correctly by the reconciler's own total_slots mechanics. @@ -488,7 +469,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `dart` function existence: GitGalaxy agree, tree-sitter differ -*2-vs-1 -- 12 occurrences as of 2026-08-29T21:17:03Z* +*2-vs-1 -- 12 occurrences as of 2026-08-29T21:43:17Z* **Verdict** (by Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep, 2026-08-21T00:00:00Z): > Confirmed real GitGalaxy engine defect, not a tree-sitter limitation. Root-caused two distinct false-positive bugs in dart's func_start regex's zero-prefix branch, both in gitgalaxy/standards/language_standards.py: (1) Dart 3 switch-expression arms (`Pattern => result,` including bare `_ => result,`) matched as getter definitions, because the branch's bare-`=>` lookahead alternative was unconditional (no `get` keyword required) even though Dart's real grammar has no parameterless `=> expr` construct without `get`. (2) Call statements with a lambda argument (`obj.method((x) => ...)`, `setState(() {...})`, bare self-calls like `visitChildren((Element child) {...})`) matched as function definitions, because the paren-lookahead used a naive non-balanced-paren check (`\([^)]*\)`) that stopped at the lambda's own closing paren instead of the outer call's. A full corpus-wide before/after diff (language-crucible/data/dart, 7 Flutter framework files) confirmed 126 false-positive matches removed and only 1 new match gained (a genuine recall fix for EditableText's own multi-line constructor) -- every one of the 126 manually spot-checked as a real false positive (dotted call receiver, or a statement ending `);`, or a switch-expression arm), none a lost real definition. Fixed in gitgalaxy#2071 (merged via this ledger-sweep PR) by gating the bare-arrow alternative behind the existing `get`-group conditional and switching to the same balanced-paren pattern already used elsewhere in this regex. A smaller, related false positive (multi-line class headers with with/implements clauses absorbed as a phantom function's return-type prefix, e.g. `implements AutofillClient {` matching as a function named AutofillClient) was found in the same investigation but is NOT fixed -- the straightforward fix (excluding implements/with/extends from the return-type-prefix vocabulary) was tested and confirmed to break a much more common legitimate pattern, generic method type-parameter bounds (`static Future pushNamed(...)`), which also uses `extends`. Tracked in gitgalaxy#2072 along with several other confirmed-but-deferred recall gaps found in the same pass. @@ -508,7 +489,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `dart` function existence: tree-sitter agree, GitGalaxy differ -*2-vs-1 -- 9 occurrences as of 2026-08-29T21:17:03Z* +*2-vs-1 -- 9 occurrences as of 2026-08-29T21:43:17Z* **Verdict** (by Claude (Sonnet 5), direct investigation via tri-comparison-ledger-sweep, 2026-08-21T00:00:00Z): > Confirmed real GitGalaxy recall gaps (tree-sitter is correct), several distinct causes found via direct source investigation against language-crucible/data/dart. Some entries in this shape's original capped example list were incidentally resolved as a side effect of the func_start fix in gitgalaxy#2071 (e.g. EditableText's own multi-line constructor, generic-return-type getters like `Iterable get children =>`, and dotted-name getters like `T? get currentState => switch (...) {...}` all now regex-match correctly and reach the final pipeline output). A post-fix full-corpus name-diff still shows 12 residual missing_from_gg cases across 5 distinct confirmed root causes, none yet fixed: (1) EditableText's constructor and a private named constructor (_DiscreteKeyFrameSimulation._(...)) regex-match (confirmed via direct probe) but don't reach the final pipeline output -- a detector.py-level extraction bug downstream of the regex, not yet traced; (2) static getters with a dotted/prefixed-import return type (`static ui.BoxHeightStyle get defaultSelectionHeightStyle {`) never regex-match at all, because the return-type-prefix character class excludes `.`; (3) bodyless default constructors (`_EditableTextTapOutsideAction();`) don't regex-match; (4-5) several more names (recorder, ChildSemanticsConfigurationsResultBuilder, OrdinalSortKey, SemanticsData, SemanticsProperties, extension) not yet individually root-caused. All tracked as a consolidated follow-up in gitgalaxy#2072 -- deliberately not rushed, since the sibling existence shape's investigation already found one naive fix (broadening the return-type charclass or the keyword-exclusion list) can silently break a much more common legitimate pattern elsewhere, so each of these needs its own regression check before shipping. @@ -529,7 +510,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `fortran` function existence: GitGalaxy, ctags agree, tree-sitter differ -*2-vs-1 -- 15 occurrences as of 2026-08-29T21:17:08Z* +*2-vs-1 -- 15 occurrences as of 2026-08-29T21:43:22Z* **Verdict** (by Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-21T23:35:49Z): > Confirmed already-documented tree-sitter-fortran limitation (#1709/Claim 7 in docs/why_gitgalaxy_beats_ast_here.md), re-verified directly against this exact shape's 14 occurrences (11 in wrf/module_physics_init.F: bl_init, ra_init, landuse_init, mp_init, cu_init, shcu_init, CAM_INIT, z2sigma, fdob_init, fg_init, ALLOCATE_CAM_ARRAYS; 3 in wrf/wrf_timeseries.F: calc_p8w, calc_ts, write_ts). Directly ran tree_sitter_accuracy_audit.py's own _find_blind_spot_ranges() against both files' real parse trees: every one of the 14 occurrences' start lines falls inside a tree-sitter ERROR/preproc_* blind-spot range (e.g. bl_init/mp_init/cu_init/shcu_init all sit inside one continuous (2203,4393) span; all 3 wrf_timeseries.F occurrences fall inside a (1,1200) span). ctags and GitGalaxy are both correct; tree-sitter-fortran's own grammar cascades into ERROR nodes on the CPP #if/#endif guards surrounding these subroutines. Added as new evidence to the existing Claim 7 in docs/why_gitgalaxy_beats_ast_here.md (this ledger shape is independently sourced from the original accuracy-audit finding -- a raw per-file name diff via tri_comparison_gatherer.py, not the audit tool's promotion logic). @@ -549,7 +530,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `fortran` function args: GitGalaxy, ctags agree, tree-sitter differ -*2-vs-1 -- 1 occurrence as of 2026-08-29T21:17:08Z* +*2-vs-1 -- 1 occurrence as of 2026-08-29T21:43:22Z* **Verdict** (by Antigravity (direct sweep), 2026-08-24T04:14:26Z): > GitGalaxy/ctags correctly extract 677 args. Tree-sitter truncates at 39. @@ -560,7 +541,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `fortran` function existence: GitGalaxy agree, tree-sitter, ctags differ -*2-vs-1 -- 1 occurrence as of 2026-08-29T21:17:08Z* +*2-vs-1 -- 1 occurrence as of 2026-08-29T21:43:22Z* **Verdict** (by Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-21T23:35:49Z): > Confirmed GitGalaxy correct on both. 'vint' (module_initialize_real.F:5375, inside #ifdef VERT_UNIT) and 'foo' (module_initialize_real.F:7519, inside #if 0) are both real, syntactically valid `PROGRAM name` declarations. tree-sitter misses both via the already-documented #1709 ERROR/preproc_* blind-spot mechanism (both sit inside #if/#ifdef-guarded regions). ctags misses 'foo' only because CTAGS_FUNC_KINDS['fortran'] was {'f','s'} (functions/subroutines only) -- ctags itself DOES correctly tag `foo` as a 'p' (program) kind (confirmed via `ctags -x --kinds-Fortran=p`), just invisible to this comparison. Fixed in tests/tools/ctags_reader.py: CTAGS_FUNC_KINDS['fortran'] now {'f','s','p','e'} (added program, entry -- matching GitGalaxy's own func_start scope of FUNCTION|SUBROUTINE|PROGRAM|ENTRY). ctags separately, genuinely fails to tag 'vint' at all under any kind in this specific file (confirmed it tags an isolated test file with identical #ifdef-wrapped `program vint` syntax fine, so it's a real, unexplained ctags-fortran-parser corruption specific to this file's content, not a preprocessor-conditional-skip decision) -- a genuine, narrow ctags limitation, not chased further for a single occurrence. @@ -573,7 +554,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `go` function args: GitGalaxy, ctags agree, tree-sitter differ -*2-vs-1 -- 75 occurrences as of 2026-08-29T21:17:10Z* +*2-vs-1 -- 75 occurrences as of 2026-08-29T21:43:24Z* **Verdict** (by Antigravity (direct sweep), 2026-08-24T04:14:26Z): > GitGalaxy/ctags correctly parse multiple arguments sharing a type. Tree-sitter groups them. @@ -595,7 +576,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `haskell` function existence: ctags agree, GitGalaxy, tree-sitter differ -*2-vs-1 -- 102 occurrences as of 2026-08-29T21:17:12Z* +*2-vs-1 -- 102 occurrences as of 2026-08-29T21:43:27Z* **Verdict** (by Claude Sonnet 5 (dispatched agent investigation), 2026-08-19T00:00:00Z): > All 10 sampled cases are ctags-side artifacts, not real GitGalaxy/tree-sitter misses -- confirmed via direct `ctags -x` output against the corpus. Three distinct ctags Haskell-parser weaknesses cover the sample: (1) multi-clause double/triple-tagging -- ctags tags every pattern-matched equation line as its own occurrence of the name (writerFn/writeFnBinary/expandFilterPath: confirmed 2-3 raw ctags tags per function, one per clause; a file-wide count found 45 such extra same-name tags across the 7-file corpus, e.g. blockToInlines alone has 14). GitGalaxy/tree-sitter both correctly anchor to the FIRST clause only, leaving ctags' later-clause tags as the ones unpaired. (2) keyword-as-identifier misparsing -- `class`/`where`/`pattern` (from PatternSynonyms) get tagged as function names when ctags fails to parse past the keyword; confirmed at Options.hs:62 (`class HasSyntaxExtensions`), Parsing.hs:184 (module-header `where`), and 3 PatternSynonyms declarations in Options.hs. (3) CAF/value-vs-function kind collapse -- defaultAbbrevs/defaultKaTeXURL/defaultMathJaxURL/defaultWebTeXURL are zero-arg top-level VALUES (non-arrow type signatures), not functions; ctags has no value/variable kind at all (`ctags --list-kinds-full=Haskell` shows only constructor/function/module/type) so it lumps every `name = expr` binding into "function". GitGalaxy (language_standards.py haskell rules, #1312) and tree-sitter's own audit tooling (_find_haskell_signature_for_bind, #1566) both independently make the same value-vs-function distinction correctly -- real cross-tool corroboration, not coincidence. (4) TH-splice call sites misread as definitions -- deriveJSON at Options.hs:454/458 is a Template Haskell splice INVOKING an imported function, not defining one; ctags misparses the call as a definition. No GitGalaxy/tree-sitter defect anywhere in this shape; purely a ctags parser limitation, same category as the already-documented empty CTAGS_CLASS_KINDS['haskell']. @@ -615,7 +596,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `haskell` function existence: GitGalaxy, tree-sitter agree, ctags differ -*2-vs-1 -- 69 occurrences as of 2026-08-29T21:17:12Z* +*2-vs-1 -- 69 occurrences as of 2026-08-29T21:43:27Z* **Verdict** (by Claude Sonnet 5 (dispatched agent investigation), 2026-08-19T00:00:00Z): > Confirmed: all 10 sampled misses (and, by cross-check against additional non-sampled instances in Options.hs/Shared.hs, plausibly all 69) are locally-scoped function definitions -- `instance ... where` methods, `where`-clause helpers, or `let`-bound names inside `do` blocks -- never top-level module definitions. ctags' Haskell parser has no layout-rule/scope awareness and only tags equations anchored at column 1; it correctly handles multi-clause TOP-LEVEL definitions (verified via expandFilterPath, writeFnBinary, writerFn -- all tag fine, clauses and all), so this is a pure scope blind spot, not a clause-counting bug (distinct from the tree-sitter clause-splitting bug fixed earlier in this same effort, which shares 2 of the 10 sample names by coincidence of subject matter, not root cause). GitGalaxy and tree-sitter are both correct; ctags is not wrong so much as structurally incapable of seeing these. Known, expected limitation of ctags' Haskell parser, now documented alongside its existing Haskell notes in tests/tools/ctags_reader.py -- not a GitHub issue, nothing in this repo to fix. @@ -635,7 +616,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `haskell` class existence: GitGalaxy, tree-sitter agree, ctags differ -*2-vs-1 -- 16 occurrences as of 2026-08-29T21:17:12Z* +*2-vs-1 -- 16 occurrences as of 2026-08-29T21:43:27Z* **Verdict** (by Claude Sonnet 5 (session investigation), 2026-08-19T00:00:00Z): > Not a real discrepancy -- structural tooling gap, already documented in the codebase. tests/tools/ctags_reader.py:39-40,228 sets CTAGS_CLASS_KINDS['haskell'] = set() on purpose: "ctags' Haskell parser has no class-shaped kind at all (constructor/function/module/type only)". Every example in this shape (data/newtype/class declarations -- ReaderOptions, CiteMethod, HasSyntaxExtensions, etc.) is real; ctags structurally cannot report any of them for this language, not a sample-specific miss. No action needed beyond this note. @@ -655,7 +636,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `haskell` function existence: GitGalaxy, ctags agree, tree-sitter differ -*2-vs-1 -- 1 occurrence as of 2026-08-29T21:17:12Z* +*2-vs-1 -- 1 occurrence as of 2026-08-29T21:43:27Z* **Verdict** (by Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-29): > Reconciler rank-pairing artifact from ctags-haskell's per-line tagging. All three tools FIND the function -- pandoc/Shared.hs's `addPandocAttributes`: GitGalaxy at line 299 (the bare name line of a multi-line definition), tree-sitter-haskell at line 301 (the first equation), each exactly once. ctags-haskell emits THREE `f` tags for it, at lines 300, 301 and 302 (the `::` type-signature line plus both `addPandocAttributes ... = ...` equation lines). tri_comparison_reconcile pairs same-named occurrences by rank: GitGalaxy's single occurrence rank-pairs with ctags' rank-1 tag (@300) and tree-sitter's single occurrence rank-pairs with ctags' rank-2 tag (@301), so the GitGalaxy+ctags slot is left with no tree-sitter partner and surfaces as `agree[ctags,gitgalaxy]_vs[tree_sitter]` for 1 occurrence. Not a tree-sitter recall gap and not a GitGalaxy defect -- purely ctags' surplus tags shifting the rank alignment. Documented in tests/tools/ctags_reader.py's haskell CTAGS_FUNC_KINDS comment. No credit/debit: ctags' extra tags are its own per-line behaviour, not a shared mistake with another tool, and neither GitGalaxy nor tree-sitter has a confirmed limitation here. @@ -666,7 +647,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `haskell` function existence: tree-sitter, ctags agree, GitGalaxy differ -*2-vs-1 -- 1 occurrence as of 2026-08-29T21:17:12Z* +*2-vs-1 -- 1 occurrence as of 2026-08-29T21:43:27Z* **Verdict** (by Claude (Sonnet 5), dispatched via tri-comparison-ledger-sweep, 2026-08-26T00:00:00Z): > Stale, superseded by extensive GitGalaxy Haskell hardening between this shape's original capture and today (closed issues #1312, #1435, #1442, #1532, #1564, #1565, #1614, #1615, #1616 all touch this exact case). A fresh gather of every sampled example (writerFn, writeFnBinary in App.hs; expandFilterPath in Filter.hs; tabFilter, compactify, camelCaseStrToHyphenated, blockToInlines in Shared.hs) shows GitGalaxy now correctly anchors to the function's TYPE SIGNATURE line, tree-sitter anchors to the first pattern-match equation, and ctags tags every equation clause separately (up to 14 tags for blockToInlines' many clauses). Rank-paired, slot 0 is now full 3-way agreement; the extra ctags clause-tags fall into the already-validated shape `haskell/function/existence/agree[ctags]_vs[gitgalaxy,tree_sitter]`, which already covers this exact multi-clause-tagging phenomenon. @@ -677,7 +658,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `haskell` function existence: GitGalaxy agree, tree-sitter, ctags differ -*2-vs-1 -- 1 occurrence as of 2026-08-29T21:17:12Z* +*2-vs-1 -- 1 occurrence as of 2026-08-29T21:43:27Z* **Verdict** (by Claude Sonnet 5 (session investigation), 2026-08-19T00:00:00Z): > Mixed shape as originally investigated (2 occurrences): (1) Options.hs:438 getExtensions -- real function, `instance HasSyntaxExtensions WriterOptions where getExtensions opts = writerExtensions opts`. A sibling instance for ReaderOptions at Options.hs:80 has the identical shape. Tree-sitter's Haskell grammar doesn't expose typeclass-instance-method clause bodies the way it does top-level bindings, and ctags' Haskell parser has no instance-method kind either -- GitGalaxy is correct, both other tools have a real recall gap on typeclass instance methods. (2) Shared.hs:475 extensionEnabled -- was NOT a real function (imported from Text.Pandoc.Extensions, only ever appears as a guard-clause call inside a multi-line `||` condition) -- a genuine GitGalaxy false positive, filed as #2082 and fixed in PR #2083 (2026-08-22): `_slice_by_indentation` now skips an equation-form func_start match whose immediately preceding line ends in `||`/`&&`. Reconciled post-fix: only the getExtensions occurrence remains, so this shape is now a clean, unambiguous GitGalaxy win -- credited accordingly. @@ -688,7 +669,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `haskell` function args: none agree, GitGalaxy, tree-sitter differ -*3-way split -- 9 occurrences as of 2026-08-29T21:17:12Z* +*3-way split -- 9 occurrences as of 2026-08-29T21:43:27Z* **Verdict** (by Claude Sonnet 5 (session investigation), 2026-08-19T00:00:00Z): > One systematic cause, confirmed by reading source for 3 of the 9 (getMetadataFromFiles App.hs:395-397, splitTextByIndices Shared.hs:142-143, tabFilter Shared.hs:256-259) and consistent with the shape of the remaining 6. Every case is a point-free/eta-reduced Haskell equation: the type signature declares N params, but the specific clause GitGalaxy and tree-sitter both align to only explicitly binds N-1 of them, handling the trailing argument via composition (`.`) or `\case`. GitGalaxy counts arity from the full type signature (the true logical arity); tree-sitter's declaration-only reading counts only the clause's explicitly-bound patterns (correct for that one equation, but undercounts true arity). Neither reader is wrong about what it's measuring -- same shape as Claim 1 in docs/why_gitgalaxy_beats_ast_here.md. GitGalaxy's answer is arguably the more useful coupling signal; recommend documenting as a candidate Claim rather than treating as an engine defect to fix toward matching tree-sitter. @@ -709,7 +690,7 @@ Sorted 2-vs-1 splits before 3-way splits, unvalidated before validated, biggest ### ✅ `html` function existence: GitGalaxy, tree-sitter agree, ctags differ -*2-vs-1 -- 1 occurrence as of 2026-08-29T21:17:13Z* +*2-vs-1 -- 1 occurrence as of 2026-08-29T21:43:28Z* **Verdict** (by Claude Sonnet 5, dispatched via tri-comparison-ledger-sweep, 2026-08-29): > GitGalaxy correct, corroborated by tree-sitter, ctags structurally cannot see it. cpython_jinja/layout.html:40 `@media only screen { ... }` inside a `