Instead, it looks like a deeper inconsistency between: - token positions in `abstract_inverted_index` - and how phrase queries are actually evaluated
Even when: - tokens are adjacent - order is correct - and no unusual whitespace is present
…the full phrase still fails to match.
---
### Conclusion
This suggests a broader issue with phrase search reliability in OpenAlex, where: - exact phrase queries do not consistently match reconstructed abstract text - behavior is inconsistent even for simple word sequences
This is especially problematic for: - scientific terms (like "B 12") - and general phrase-based search workflows
---
It would be helpful to understand: - how phrase queries are evaluated internally - and whether this behavior is expected or a bug