# diff_core
Semantic PDF diff engine for matching extracted document nodes.
`diff_core` compares two `pdf_semantic::SemanticDocument` values and produces a
stable `spdfdiff_types::DiffDocument`. It focuses on semantic block matching,
text hunks, move detection, layout-only changes, confidence scoring, and neutral
severity defaults for PDF comparison reports.
## What This Crate Provides
- Exact matching through deterministic semantic anchors.
- Ordered fuzzy matching inside unmatched exact-anchor windows.
- Resource-bounded matching with deterministic fallback when matrix limits are
exceeded.
- Inserted, deleted, modified, moved, and layout-changed semantic changes.
- Text hunks with token ranges for modified text.
- Character-level fallback hunks for small non-numeric word replacements.
- Structured layout evidence for page and bounding-box movement.
- Default severity classification that does not emit legal/business `Critical`
severity.
## Pipeline Context
`diff_core` is the comparison stage:
```text
old SemanticDocument + new SemanticDocument -> diff_core -> DiffDocument
```
Reports are generated by `diff_report`; this crate stays independent of JSON,
Markdown, HTML, and CLI rendering.
## Current Compatibility Boundary
Matching quality depends on the semantic nodes produced upstream. This crate can
separate moved content and layout-only changes when text anchors and geometry
support it, but it does not solve OCR, full visual diffing, arbitrary table-cell
semantics, or legal/business classification. Domain-specific severity can be
provided by a caller-supplied classifier.