Supporting Content:Cognitive Search:Gives chunks that sometimes starts abruptly and in few occasions doesn't complete the paragraph or sentence.

[vikassalaria2412]

Please provide us with the following information:

This issue is for a: (mark with an x)

- [X ] bug report -> please search issues before submitting
- [X ] feature request
- [ ] documentation issue or request
- [ ] regression (a behavior that used to work and stopped in a new release)

Minimal steps to reproduce

Ask a question.LLM will respond.Click on citation.Go to supporting content tab.

Any log messages given by the failure

Expected/desired behavior

Expected: The chunk should start forming a proper sentence and should stop ending correctly.Currently for few references the chunk supporting doc starting from middle of the reference document or sometimes stops without completing complete sentence.
For example it starts from: and development ....... and end without proper closing of sentence.

OS and Version?

Windows 7, 8 or 10. Linux (which distribution). macOS (Yosemite? El Capitan? Sierra?)
Win 10

azd version?

run azd version and copy paste here.

Versions

azd version 1.3.0 (commit 6605c5768cb3ab6d9e0e6c2d892c9e8b49e1f720)

Mention any other details that might be useful

May be this is happening because of token limitation the chunks getting created beforte creation of embedddings.

---

Thanks! We'll be in touch soon.

[github-actions]

This issue is stale because it has been open 60 days with no activity. Remove stale label or comment or this issue will be closed.

[pamelafox]

Thanks for reporting this and for the details you provided. The behavior you described — supporting content starting mid‑sentence or ending abruptly — can happen when content is chunked to respect model/token limits. The sample uses an overlapping/sliding‑window chunking strategy to minimize context loss, but the snippets shown in Supporting content may still be fragments in some cases.

Since there hasn’t been any activity since 2023 and we don’t have a reproducible example (sample document/index, chunking/overlap settings, ingestion path, versions), we’re going to close this out as stale. The sample has also evolved since then.

If you can still reproduce this on the latest main branch, please open a new issue with:

• A minimal sample document (or redacted example) and how it was ingested
• Your chunk size/overlap or ingestion settings and language/runtime used
• Azure Cognitive Search/index settings (if applicable)
• Exact steps and screenshots of the Supporting content panel

We’re happy to investigate with concrete details. Thanks again!