The current logic for the generation is the following, page per page:

1. page_read_source
2. page_markdown
3. page_content

This approach leads to inconsistencies in the generated content, when a page uses the metadata from one or multiple other page(s). Use-cases are for example a page listing other pages with a certain metadata content (like the pages of some author, the pages that have a certain status, etc.).

A specific related issue when using the built-in server and monitoring file changes: when you change file A that uses metadata from file B, the newly generated content now have access the B metadata that was missing during the initial generation. As a consequence the generated content of A changes without any link to the effective changes in A.

A correction would be to load all files, including metadata, and only then do the content generation accordingly. This would have the following positive impacts:

1. solving situations where generated content is inconsistent,
2. improve features (like the one found in jekyll or hugo generators

However, this would also negatively impact the memory usage, like the proposal in #2669

All: it would be interesting to have your end-user feedback here (please vote) to evaluate the willingness to have such feature rather than performance optimization.