The issue you've encountered with MarkdownElementNodeParser not passing exclude_llm_metadata_keys and exclude_embed_metadata_keys from the document to its child nodes is indeed due to its implementation. Unlike parsers that inherit from MetadataAwareTextSplitter, MarkdownElementNodeParser does not handle these specific metadata exclusion keys. This behavior is because it focuses on parsing markdown content into nodes without considering the exclusion of specific metadata keys.

To address this, you can modify the _build_node_from_split method within MarkdownElementNodeParser to filter out the unwanted metadata keys before they are added to the node. Here's how you could implement such a filter:

def _build_node_from_split(
    self,
    text_split: str,
    node: BaseNode,
    metadata: dict,
) -> TextNode:
    """Build node from single text split."""
    node = build_nodes_from_splits([text_split], node, id_func=self.id_func)[0]

    if self.include_metadata:
        # Filter out specific metadata keys
        filtered_metadata = {key: value for key, value in metadata.items() if key not in ['source_id', 'source_date']}
        node.metadata = {**node.metadata, **filtered_metadata}

    return node

In this example, filtered_metadata is created by excluding 'source_id' and 'source_date' from the metadata dictionary before it's assigned to node.metadata. This modification directly addresses the issue by ensuring that these keys are not passed to child nodes.

Please note, this solution requires modifying the LlamaIndex codebase, which might not be ideal for all situations and would need to be reapplied after any updates to the library.

To continue the conversation, mention @dosu.

@logan-markewich can you implement a fix that does not require modifying source code? ideally all node parsers / transformations inherit these qualities from base document

Comes down to these two methods from what I can tell:

build_nodes_from_splits in node utils:

elif isinstance(document, Document):
            node = TextNode(
                id_=id_func(i, document),
                text=text_chunk,
                embedding=document.embedding,
                excluded_embed_metadata_keys=document.excluded_embed_metadata_keys,
                excluded_llm_metadata_keys=document.excluded_llm_metadata_keys,
                metadata_seperator=document.metadata_seperator,
                metadata_template=document.metadata_template,
                text_template=document.text_template,
                relationships=relationships,
            )
            nodes.append(node)

versus:
get_node_from_elements in base_node.py:

        for node in nodes:
            if metadata_inherited:
                node.metadata.update(metadata_inherited)
        return [node for node in nodes if len(node.text) > 0]