test(tree-sitter-audit): re-bless stale css baseline after #2500's audit-reader fix

[squid-protocol]

Closes #2507.

What was failing

tree-sitter-accuracy-audit red on main for css, exit 1:

real_functions: baseline=24 current=25

Only the ground-truth-drift guard fails CI. Everything else already moved the right way:
found_functions 20→25, extra_functions 5→0, args_exact_match 14→19.

Root cause (issue item 2 — "investigate before regenerating")

Not corpus drift. files_scanned is 21 in both the committed baseline and now — #2478 already
regenerated this baseline 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 tests/tree_sitter_accuracy_baseline_css.json:

• the fixed dead-code keyframes_statement name branch (previously always return None → 0
  @keyframes corpus-wide) now recognises keyframes — +funcs
• the at_rule branch now returns None for the bodyless @layer a, b; ordering form, which
  GitGalaxy deliberately doesn't match — −funcs

Net +1 (24→25). #2500 updated the tri-comparison chart/ledger and docs/language_status/css.md
but missed this JSON — its own commit message records the target: "css Func Found 25*/24*/0* ->
25/25/0". Zero GitGalaxy engine change, zero corpus change.

Issue item 1 — tighten css func_start to drop @media/@supports/@container/@layer

Not done, deliberately. It contradicts decisions already validated:

• all 3 css function entries in tri_comparison_ledger.json are status: validated;
  css/function/existence/agree[gitgalaxy,tree_sitter]_vs[ctags] (25 occ) verdict: "not a
  GitGalaxy defect … func_start deliberately matches CSS at-rule keywords … the closest
  function-shaped construct CSS has."
• tree_sitter_accuracy_audit.py NODE_MAPS deliberately counts
  media_statement/supports_statement/keyframes_statement/at_rule as ground-truth functions
  (tree_sitter_accuracy_audit.py reports real_functions=0 for css/kotlin/powershell/zig despite confirmed function-shaped nodes in real corpus files #1313). Dropping them from func_start would crater measured recall and force a NODE_MAPS revert.
• after this re-bless, css sits at 100% func recall / 100% func precision in both the 2-tool
  audit and the tri-comparison — the "precision 100%→83.3%" note from the v1.2.0 cut was already
  resolved by css tri-comparison: func precision + recall to 100% (audit-reader fixes, no engine change) #2500 as an audit-reader bug, not an engine defect.

The only residue is the informational args_comparable mismatch (@media all and (max-width:600px)
→ GitGalaxy got=3, tree-sitter real=0, because at-rule nodes have no parameters field). This
PR documents it in the SCOPE docstring — same "no formal signature at the declaration site" shape as
the shell/perl note (#1518/#1519). The gated args_exact_match is unaffected and improved.

Changes

File	Change
tests/tree_sitter_accuracy_baseline_css.json	regenerated (--regenerate)
gitgalaxy/standards/language_standards.py	summary-table row: Css 83.3%/80.0% → 100.0%/100.0% (auto-refreshed by --regenerate)
tests/tools/tree_sitter_accuracy_audit.py	SCOPE-docstring note on the at-rule args_comparable gap

No golden-master re-bless — no engine/regex change.

Verification

• tree_sitter_accuracy_audit.py --lang css --ci → exit 0
• tree_sitter_accuracy_audit.py --all --ci → 31 languages, all OK, no other language moved
• tree_sitter_accuracy_audit.py --summary-table → already matches
• pytest -k "tree_sitter_accuracy or summary_table or language_standards_strict" → 21 passed

🤖 Generated with Claude Code

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab