Okay, so then I just have to make a metadata plugin that collects the TOC from the input and stuffs it into some arbitrarily named metadata field, and then configure the template to include that metadata in the sidebar?

And then in the template logic I can handle the sidebar of the non-post pages by just checking whether that field exists?

Yeah, that should work. Although removing the auto-generated ToC from the post text may be difficult.

I think there might be a miscommunication here. I want to scan over the entire Markdown document to look for # lines, and generate a TOC from that. These could be spread out through all of the post Markdown, so I can't just set a split_metadata_re; the extraction needs to examine the whole document and also not remove the lines that it processes. But I don't see why there would be "auto-generated ToC in the post text" in the first place; the idea is that the auto-generated TOC appears only in a metadata field, not in the Markdown given to the renderer.

... Maybe I should do this at the PageCompiler level instead.

Oh, I thought you wanted to use the toc Markdown extension.

You can certainly parse the Markdown source to extract headers and put it in a meta field in a MetadataExtractor plugin. Note that just searching for # can cause issues, e.g. if you have code blocks in your posts that happen to contain Python comments.

Another idea might be to parse the generated HTML to extract h1-h6 tags. That would probably not be done at the metadata level, but you could write a parser function, pass it into GLOBAL_CONTEXT, and then call it from the template.

One caveat for either approach is that python-markdown does not generate ids for headers by default, so you would need to handle that yourself. The toc extension can insert IDs, so perhaps you could write your own Markdown extension based on it to add the ids.