Technical notes on web development, DevOps, and AI integration.
1 article
The corrective import of PPKD cultural heritage data failed because the SK number wasn't unique. Candidate pools and overlap coefficient solve 1:1 matching.
TL;DR: Importing PPKD heritage data broke when SK numbers turned out to be shared across multiple objects, so direct key matching corrupted good records. The fix pairs each row with the best name match from an SK's candidate pool using the overlap coefficient, plus a used guard preventing double assignment. Column ownership rules keep verified coordinates and media untouched.