Indexing gap: reference #2570

codebase-memory-mcp reports a best-effort indexing gap for the project Users-gautamparab-Documents-antigravity-decisionnew. index_status lists 32 files with parse_partial ranges. 27 are HTML files under one reference folder, 2 are markdown files, and 1 is a CSS file. No code is included; the failing constructs are described below.

Three kinds of construct fail to parse:

1. Bare & and > in HTML text and titles. Examples: a <title> containing Benchmarks & ..., <div class="eyebrow"> text containing &, and <text> or <p> content containing >3 times?, S×O > 40. The flagged lines are title, eyebrow, heading and paragraph text, not attributes.
2. A CSS selector list with a trailing comma before the closing brace, e.g. .x span:not(:first-child), }. This is invalid CSS, so the parser recovers at that point.
3. Markdown table rows with unbalanced pipes, including two rows joined on one line and cells containing |. The parser recovers mid-table.

Impact: the flagged lines may be missing from the graph; the rest of each file is indexed. The affected files are mostly reference examples, so the impact on code search is small. The coverage note says the signal is best-effort, which matches this output.

Environment: codebase-memory-mcp version as installed on macOS (Darwin 25.6.0). Index status shows status: ready with 453,028 nodes and 470,027 edges.