Function code_nodes_schema 

pub fn code_nodes_schema() -> Schema

Schema for the CodeNodes table.

Columns 0–12 (original):

• id: fully-qualified identifier (e.g. func:brain/perception/signal_fusion.py::fuse)
• kind: object type (dictionary-encoded string, Dictionary<Int8, Utf8>)
• parent_id: containment pointer (function→class→module→file)
• name: object name
• signature: function/method signature
• docstring: documentation text
• body_hash: SHA-256 of the object body for equality checks
• body: full source text of the node
• embedding: semantic vector (FixedSizeList<f32, 768>)
• loc: lines of code
• cyclomatic_complexity: complexity metric
• coverage_pct: test coverage percentage
• last_modified: timestamp from git history

Columns 13–18 (EX-3168 / V12a-1 — position metadata, all nullable):

• start_line: 1-indexed source line where the node begins
• end_line: 1-indexed source line where the node ends
• start_col: 0-indexed column within start_line
• end_col: 0-indexed column within end_line
• file_path: relative path from crate root (e.g. "src/lib.rs")
• byte_offset: byte position from file start (for fast editor seek)

Position columns are nullable for backward compatibility with existing Parquet files. The tree-sitter Rust parser (EX-3120) populates them for new nodes.