I built a schemaless intake for my CRM. It silently dropped files three times.

Short summary

A developer building a personal CRM discovered that schemaless file intake silently drops files without errors or warnings—hitting this failure mode three times during prototyping. The fix was a deterministic index layer between intake and interpretation that records path, date, and content hash for every file, making 'what got picked up' a diffable question rather than an assumption. The design borrows Palantir's Object/Link/Action separation but is a weekend prototype, not production-tested.

• •Schemaless intake pipelines can silently drop files with zero errors or warnings
• •A deterministic index layer recording path, date, and content hash makes drops detectable
• •Design borrows Palantir's Object/Link/Action ontology split for separating intake from interpretation

Generated with AI, which can make mistakes.