Build the Index

Basic Usage

iw index build

This runs the full pipeline and writes .iw/index.db.

Options

Option	Default	Description
--depth <mode>	structured	structured or full (see below)
--include <glob>	—	Only index files matching glob
--exclude <glob>	—	Skip files matching glob
-v, --verbose	off	Show per-stage progress

Depth Modes

structured (default)

Scans only headings, bold text, code spans, and identifiers in documents. Fast and precise — best for everyday use.

iw index build
# or explicitly:
iw index build --depth structured

full

Adds body text scanning with dictionary matching and IDF-based noise filtering.

iw index build --depth full

Full mode produces significantly more annotations:

Metric	Structured	Full
Annotations	6,721	11,533 (+72%)
Grounded links	2,548	7,360 (+189%)
Co-occurrences	1,099	2,631 (+139%)
Build time	1.1 s	2.8 s

The IDF filter removes noise by penalizing terms that appear in nearly every document (common words like “data”, “type”, “config”). A baseline of 50 stopwords with a ceiling of 0.15 keeps precision high while allowing the richer recall.

Pipeline Stages

The build runs these stages in order:

1. AX — AST extraction (tree-sitter) → builds a symbol registry with all classes, functions, interfaces, types, exports
2. KWX — Keyword extraction → scans documents for entity mentions. In full mode, receives the symbol dictionary for body text matching
3. COX — Co-occurrence scoring → finds entity pairs mentioned together
4. TCG — Git analysis → co-change Jaccard scores, hotspot detection, ownership, staleness flags
5. Annotate — Matches document mentions to code symbols, applies IDF penalties
6. Write — Persists everything to SQLite

Filtering Files

# Only index src/ and docs/
iw index build --include "src/**" --include "docs/**"

# Exclude test files
iw index build --exclude "**/*.test.ts" --exclude "**/__tests__/**"

Output

The build creates .iw/index.db with these tables:

Table	Contents
symbols	Code symbols from AST (name, kind, file, line, exported)
annotations	Doc spans → code symbols (confidence, source, IDF score)
co_occurrences	Entity pairs co-mentioned in docs or co-imported in code
co_changes	File pairs that change together in git (Jaccard + recency)
files	Per-file metadata (last modified, churn, hotspot, owner, content hash)

Next Steps

• Retrieve — query the index for relevant files
• Incremental Update — re-index only changed files