I am creating my own Document nodes to put into a VectorStoreIndex, and I am setting page_label as metadata via doc.metadata["page_label"]=pgno.

But when I then try to query a QueryEngine derived from this vector index using metadatafilters fromDIcts {"key": "page_label", "value": p} I never get any results.

It's completely opaque to me how to tell how the filters are working, or what metadata they may need. How can I debug this? I'd like the issue fixed, but I'm also looking for general techniques to debug these classes.

Hi everyone. Very general question here. I'm having a hard time getting code generated with GPT or Claude to run. It seems to reference pre 0.10 packages and my suspicion is that a lot of packages moved recently. Is this true? Are there AI assistants that can generate llamaindex code reliably?

As a human I'm also finding that current examples are hard to find. Some things such as the PDFReader location are deeply nested but seem like they "should" be in a more accessible location (e.g.from llama_index.readers.file.docs.base import PDFReader). Am I approaching development, docs and code gen all wrong?