A hands-on investigation into how OutSystems Developer Cloud (ODC) handles text chunking for its native Semantic Search feature. The author built a test app loading the same document into four entities — each using a different chunking method (Smart, Fixed-Size, Sentence-Based, Recursive) — and compared results across Markdown and plain text formats. Key findings: Fixed-Size cuts mid-word with no structural awareness; Sentence-Based explodes chunk count (65–66 chunks) and loses context for code/structured data; Recursive can produce garbage micro-chunks via its character-level fallback; Smart performs best overall but still splits code blocks. All four methods share a blind spot: no heading context injection and no concept of code block atomicity. The native options work well for plain prose (support tickets, knowledge base articles) but degrade silently when content is structured (code, tables, hierarchical documents). The author hints at a follow-up covering a custom C# External Logic chunking library that addresses these gaps.
