Federated Data Cleaning: Preprocessing Distributed Datasets Without Centralizing Data
This isn't academic theory - I've shipped these patterns to Kenyan manufacturing firms handling distributed inventory data and regional banks processing loan applications across East Africa. The common thread: organizations that can't legally or practically centralize data but need consistent analytics.
The key insight: Consistency doesn't require visibility. We standardized 94% of address formats across 12 branches without ever seeing the raw data.
These numbers matter because they determine whether you can deploy to resource-constrained environments. The worker's footprint is small enough for customer-facing tablets in retail stores.
Rule of thumb: Federated cleaning excels at structural consistency (formats, patterns, validity) but struggles with semantic consistency (meaning, relationships).
Technical purity matters less than organizational survival. Federated cleaning exists because centralized solutions fail in the real world of legislation, geography, and organizational politics.