# Context-preserving multi-granularity embeddings > Source-anchored research for Context-preserving multi-granularity embeddings. Run status: BUDGET_EXHAUSTED. Parent: [Concept knowledge architecture strategies](https://llms-explorer.com/tree/concept-knowledge-architecture-strategies/) · 3 facets · 15 facts · page: https://llms-explorer.com/tree/context-preserving-multi-granularity-embeddings/ ## How it works - The late-chunking paper finds no benefit when surrounding document text is irrelevant to the target information. (low confidence) — [source](https://arxiv.org/html/2409.04701v3) - Long late chunking uses overlapping long-context windows when the document exceeds the embedding model's context capacity. (low confidence) — [source](https://arxiv.org/html/2409.04701v3) ## Comparisons and alternatives - Anthropic reports limited gains from generic document summaries added to chunks, unlike its chunk-specific context. (low confidence) — [source](https://www.anthropic.com/research/contextual-retrieval) ## Facts and statements - Voyage contextualized embedding accepts grouped chunks so each chunk is encoded with its document peers. (low confidence) (verified as of 2026-10-22) — [source](https://www.mongodb.com/docs/voyageai/models/contextualized-chunk-embeddings.md) - Voyage contextualized embedding preserves chunk order in the returned document embeddings. (low confidence) (verified as of 2026-10-22) — [source](https://www.mongodb.com/docs/voyageai/models/contextualized-chunk-embeddings.md) - Voyage supports query and document input types that add task-specific prompts before embedding. (low confidence) (verified as of 2026-10-22) — [source](https://www.mongodb.com/docs/voyageai/models/contextualized-chunk-embeddings.md) - Anthropic contextual retrieval prepends chunk-specific explanatory context before embedding and BM25 indexing. (low confidence) — [source](https://www.anthropic.com/research/contextual-retrieval) - Late chunking contextualizes document tokens before pooling token spans into individual chunk vectors. (low confidence) — [source](https://arxiv.org/html/2409.04701v3) - Dense X defines propositions as minimal facts that include the context required to interpret them independently. (low confidence) — [source](https://arxiv.org/html/2312.06648v3) - Late chunking can be applied to suitable long-context embedding models without additional training. (low confidence) — [source](https://arxiv.org/html/2409.04701v3) - Dense X reports retrieval and QA improvements from proposition units under its evaluated computation budgets. (low confidence) — [source](https://arxiv.org/html/2312.06648v3) - Chunk boundaries, size and overlap remain empirical design choices in Anthropic's contextual retrieval recipe. (low confidence) — [source](https://www.anthropic.com/research/contextual-retrieval) - Dense X's evaluated retrieval corpus is Wikipedia, limiting direct evidence about other corpora. (low confidence) — [source](https://arxiv.org/html/2312.06648v3) - The late-chunking paper's overlap ablation does not show a clear retrieval advantage from overlapping chunks. (low confidence) — [source](https://arxiv.org/html/2409.04701v3) - Dense X's manual proposition audit found both unsupported propositions and propositions lacking standalone context. (low confidence) — [source](https://arxiv.org/html/2312.06648v3)