Chunk Boundaries
Also called Chunk Boundaries
Where you cut a document into pieces — and how a bad cut can slice an idea in half.
Think of it like
Tearing a recipe across the middle of the ingredient list — each half is useless without the other.
Example
A naive splitter cuts a document every 500 characters and splits "do not exceed" onto one chunk and "the dosage" onto the next, so neither retrieves the warning intact.
How it actually works
Chunking has to happen for retrieval, but where you cut matters enormously. Fixed-size splits are simple but ignore meaning, stranding half a thought on either side of a boundary. Smarter approaches cut on natural seams — paragraphs, headings, sentences — so each chunk is a coherent unit. Get boundaries wrong and even perfect retrieval returns fragments the model can’t use.
For product teams
An unglamorous knob that quietly decides whether retrieved snippets are usable or garbled.
For engineers
Prefer semantic/structural split points (headings, sentences) over fixed-length cuts to keep chunks self-contained.
Related
- Chunking — The broader operation.
- Overlap Window — A soft fix that stitches boundaries back.
- Semantic Search — Coherent chunks retrieve better.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome