Repository Unicode audit

Versions marked “overview” do not contain this page. Selecting one opens that version’s documentation overview.

The audit covers the tracked repository text after integrating main at af82146cf9d23fa1fa5d7ffb587fb5053ca23a92. All 926 regular files decode as UTF-8 and were scanned, including source, documentation, workflows, scripts, manifests, lockfiles, and this report. The two tracked symbolic links have ASCII targets: .claude/CLAUDE.md and tests/e2e/topologies/grid-glb-demo/deploy. Their targets are already in the scanned set. No tracked binary, vendor directory, or text file was excluded.

Typographic dash punctuation, prose ellipses, and decorative box-drawing separators were replaced with ASCII punctuation or ordinary section comments. Generated CRD descriptions carry the same replacements as their Rust source. The documentation index uses colons consistently between links and descriptions. Links, command arguments, configuration values, and executable Rust tokens are preserved; changed runtime strings are diagnostic or assertion messages.

No non-breaking spaces, typographic quotes, zero-width or bidirectional control characters, private-use characters, or non-ASCII identifier lookalikes remain. The remaining non-ASCII characters occur in 73 files and are intentional:

The missing-data value is DOWN_VALUE in the dashboard’s SitePanel.tsx. Changing it would change the UI contract. crates/xtask/src/env.rs uses a Unicode ellipsis to mark truncated evidence. The diagram characters are confined to docs/adding-provider.md, docs/installation/existing-clusters.md, and crates/swim/src/crypto.rs; .github/workflows/tests.yaml also matches the actual box-drawing output of cargo tree. The accented character in the dashboard metrics client’s test and the euro sign in the MCP probe test exercise UTF-8 byte boundaries. These are functional fixtures, not prose punctuation.

Reproduce the inventory

Run this from the repository root. It reads every tracked regular file without following symbolic links and prints each non-ASCII code point’s count:

python3 - <<'PY'
import collections
import pathlib
import subprocess
import unicodedata

paths = subprocess.check_output(['git', 'ls-files', '-z']).decode().split('\0')
counts = collections.Counter()
regular = 0
for name in filter(None, paths):
    path = pathlib.Path(name)
    if path.is_symlink():
        continue
    text = path.read_text(encoding='utf-8')
    counts.update(char for char in text if ord(char) > 127)
    regular += 1
print(f'{regular} regular UTF-8 files scanned')
for char, count in sorted(counts.items()):
    print(f'U+{ord(char):04X} {unicodedata.name(char)}: {count}')
PY

The existing Unicode Safety Check workflow continues to detect hidden Unicode in PR diffs. It does not ban the intentional visible characters above; an additional blanket ASCII gate would reject valid fixtures and diagrams.