css tri-comparison: func precision + recall to 100% (audit-reader fixes, no engine change) by squid-protocol · Pull Request #2500 · squid-protocol/gitgalaxy

squid-protocol · 2026-08-30T19:07:46Z

What

CSS func precision and recall to 100%, by fixing two css-only defects in the
tri-comparison audit reader (tests/tools/tree_sitter_accuracy_audit.py::_get_node_name).
Zero GitGalaxy engine change — detector.py / language_standards.py / prism.py untouched,
golden masters untouched. Both defects made the audit disagree with GitGalaxy where GitGalaxy's
own css func_start is correct. Same bug class as the closed #1313.

Surfaced by a tri-comparison-ledger-sweep pass on css (fixes #2499).

Bug 1 — `keyframes_statement` name branch was dead code (precision)

The branch looped for a child of node type at_keyword, but the pinned tree-sitter-css grammar
names that child literally @keyframes / @-webkit-keyframes (only @media/@supports use the
at_keyword shape). It always fell through to return None, so tree-sitter reported 0
@keyframes corpus-wide despite the nodes being in its parse tree. GitGalaxy correctly matches
@keyframes. Ledger shape css/function/existence/agree[gitgalaxy]_vs[ctags,tree_sitter], 5 occ
(tailwindcss_atrules/theme.css:443/449/457/463 nested in a Tailwind-v4 @theme block;
gutenberg_css_modules/button.module.css:237).

Fix: accept the @keyframes / @-webkit-keyframes child types, fall back to "keyframes".

Bug 2 — bodyless `@layer a, b;` counted as a function (recall)

The at_rule branch returned "layer" for both the block form @layer name { … } (which
GitGalaxy matches) and the bodyless cascade-layer ordering statement @layer a, b, c; (which
GitGalaxy deliberately does not — its func_start counts only block-bearing at-rules, the same
body-bearing-definition-only convention that drops C/C++ forward declarations). Ledger shape
css/function/existence/agree[tree_sitter]_vs[ctags,gitgalaxy], 4 occ.

Fix: at_rule branch returns None when the node has no block child. @container is never
bodyless, so only @layer a, b; is affected.

Scope

at_rule / keyframes_statement appear in no other language's NODE_MAPS entry. Verified:
the full --all chart regen moves only css cells plus the summary tallies that roll up from them
(diff below). --all --ci (javascript/typescript/zig baselines) passes unchanged.

Verification

recall_audit.py css: 4 counted misses → 0
Chart: css Func Found 25*/24*/0* → 25/25/0; css Func Precision 20/25* & 20/24* →
25/25 & 25/25, asterisk-free
python tests/tools/tri_comparison_chart.py --all --ci → all OK
test_css.py + test_css_strict.py → 92 passed; test_html_embedded_ts_funcs.py,
test_tri_comparison_{ledger,reconcile}.py → pass
ruff_audit.py --ci / mypy_audit.py --ci → no new findings
No golden-master re-bless — tests/tools/tree_sitter_accuracy_audit.py is never imported by
anything under gitgalaxy/; GitGalaxy scan output is byte-identical.

Also in this PR

Both css ledger shapes → status: validated with full verdicts (no credit/debit — real
corroboration once the reader is fixed, and a single-tool over-count respectively)
Regenerated tri_comparison_chart.svg / tri_comparison_ledger.json /
tri_comparison_points_of_interest.md
New docs/language_status/css.md (§1–8 via the language-status skill, §9 tri-comparison
capstone by hand) + README index row

🤖 Generated with Claude Code

@container

…s; css func precision+recall to 100% tree_sitter_accuracy_audit.py _get_node_name had two css-only defects (same class as #1313), both making the tri-comparison audit disagree with GitGalaxy where GitGalaxy's own css func_start is correct: - keyframes_statement name branch was dead code: it looked for a child of node type `at_keyword`, but this grammar version names it `@keyframes` / `@-webkit-keyframes`. It always fell through to `return None`, so tree-sitter reported 0 @Keyframes corpus-wide. Fixed: accept the `@keyframes` / `@-webkit-keyframes` child types, fall back to "keyframes". - the `at_rule` branch returned "layer" for the bodyless `@layer a, b, c;` ordering statement, which GitGalaxy deliberately does not match (its func_start counts only block-bearing at-rules, the same body-bearing-definition-only convention that drops C/C++ forward declarations). Fixed: return None when the at_rule node has no `block` child. @container is never bodyless, so only `@layer a, b;` is affected. at_rule / keyframes_statement appear in no other language's NODE_MAPS, so the change is css-only -- verified: the full --all chart regen moves only css cells plus the summary tallies that roll up from them. Ledger shapes validated (verdicts in tri_comparison_ledger.json): - css/function/existence/agree[gitgalaxy]_vs[ctags,tree_sitter] (5 occ, keyframes) -> audit-reader defect; tree-sitter now corroborates all 5. No credit/debit. - css/function/existence/agree[tree_sitter]_vs[ctags,gitgalaxy] (4 occ, @layer) -> GitGalaxy correct; audit no longer counts the bodyless form. No credit/debit. recall_audit.py css: 4 counted misses -> 0. Chart: css Func Found 25*/24*/0* -> 25/25/0; css Func Precision 20/25* & 20/24* -> 25/25 & 25/25, no asterisks. Zero GitGalaxy engine change -- golden masters untouched. Refs #2499. Same bug class as #1313. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Lsfg2qxhocsSsiejiP1XqC

Sections 1-8 via the language-status skill (30/48 rule keys wired, 92 dedicated test cases, epic history #577/#737/#841/#955 + cross-language #645). Section 9 is the tri-comparison capstone from this sweep: 3/3 css discrepancy shapes resolved, zero GitGalaxy engine defects, recall 4-miss -> 0, both precision and recall at 100%. Documents the two audit-reader bugs fixed in #2499 and the bodyless-@layer recall-audit bucketing. Refs #2499. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Lsfg2qxhocsSsiejiP1XqC

github-actions · 2026-08-30T19:10:22Z

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab

…dit-reader fix (#2507) (#2509) `tree-sitter-accuracy-audit` was red on main for css with a ground-truth drift (`real_functions: baseline=24 current=25`). Not corpus drift: files_scanned is 21 in both the committed baseline and now -- #2478 already regenerated it against the v1.2.0 pin. The 24->25 is #2500 ("css tri-comparison ... audit-reader fixes, no engine change") landing its own `_get_node_name` changes without re-blessing this JSON: the fixed dead-code `keyframes_statement` branch (was always `return None`, so 0 `@keyframes` corpus-wide) adds keyframes; the `at_rule` branch now returning None for the bodyless `@layer a, b;` form removes those. Net +1. #2500 updated the tri-comparison chart/ledger and docs/language_status/css.md but missed this file -- its commit message records the target state ("css Func Found 25*/24*/0* -> 25/25/0"). Zero GitGalaxy engine change, zero corpus change. Regenerated via `--regenerate` (also refreshes the summary table in language_standards.py: css 83.3%/80.0% -> 100%/100%, matching the tri-comparison). All movement is an improvement or the explained audit-reader correction: found_functions 20->25, extra_functions 5->0, args_exact_match 14->19. Also documents the permanent css at-rule `args_comparable` discrepancy in the SCOPE docstring (GitGalaxy reads a param count off the at-rule prelude; `_get_param_count` has no parameters field for at-rule nodes -> real=0). Not an engine defect -- a CSS at-rule has no formal parameter list, and counting at-rules as functions is a ledger-validated design choice (`css/function/existence/agree[gitgalaxy,tree_sitter]_vs[ctags]`, status validated). The gated `args_exact_match` is unaffected. No golden-master re-bless: no engine/regex change. Claude-Session: https://claude.ai/code/session_01Tpu4YmAESTUeSoPQ5LqLWy Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>

squid-protocol and others added 2 commits August 30, 2026 15:01

squid-protocol merged commit 4cc5001 into main Aug 30, 2026
26 of 27 checks passed

squid-protocol deleted the feature/tri-comparison-css-precision-recall branch August 30, 2026 19:07

squid-protocol mentioned this pull request Aug 30, 2026

tree_sitter_accuracy_audit _get_node_name: css keyframes_statement branch is dead code; bodyless @layer counted as a function #2499

Closed

Navigation Menu

css tri-comparison: func precision + recall to 100% (audit-reader fixes, no engine change) - #2500

css tri-comparison: func precision + recall to 100% (audit-reader fixes, no engine change)#2500
squid-protocol merged 2 commits into
mainfrom
feature/tri-comparison-css-precision-recall

squid-protocol commented Aug 30, 2026

Uh oh!

Uh oh!

github-actions Bot commented Aug 30, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

squid-protocol commented Aug 30, 2026

What

Bug 1 — keyframes_statement name branch was dead code (precision)

Bug 2 — bodyless @layer a, b; counted as a function (recall)

Scope

Verification

Also in this PR

Uh oh!

Uh oh!

github-actions Bot commented Aug 30, 2026

🐦‍⬛ Muninn Security Scan

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Bug 1 — `keyframes_statement` name branch was dead code (precision)

Bug 2 — bodyless `@layer a, b;` counted as a function (recall)