Command Palette

Search for a command to run...

All challenges
RAGAdvanced

Embedding Model Version Drift

You upgraded your embedding model from text-embedding-3-small to text-embedding-3-large. Search quality dropped 40% overnight despite the new model being 'better'.

Symptoms
  • Recall@5 dropped 0.88 → 0.53
  • New model is objectively better on benchmarks
  • Drop happened immediately after upgrade
  • Old queries return wrong results
Evidence

Upgrade process

- Updated embed() to call new model
- Did NOT re-embed existing 50k documents
- Vector index still contains old embeddings

Vector dimensions

Old: 1536-dim
New: 3072-dim
Index schema: 1536-dim
Tasks
  • 1Diagnose the failure
  • 2Design the correct upgrade path
  • 3Add guardrails to prevent recurrence