The current approach to per-section text editing is inherently fragile:

• Sections links are generated based on rendered HTML headings (before macro injections).
• Section extraction from the markup source is based on simplified regexp-based parsing of the markup, independent on the fully featured markup parser.

The issue is going to be more significant after relaxing restrictions on markup syntax (#32424, #35035).
But as #35036 shows, even the current restrictions do not guarantee correct section extraction. See also the exmple below.

I can imagine two approaches to the solution:

For CommonMark - using the @sourcepos@ feature of the renderer. Also applicable to other eventual formatters with similar feature.¶

For any markup - introducing validation of the section extraction results, which would detect that the extraction fails and would disable per-section edit links.¶

I can offer creating a PoC of the @sourcepos@ approach for CommonMark format after #32424 is incorporated.

A difficult-to-solve example of Markdown with broken section extraction follows (copied from the skipped unit test in #35036):

# Title

## Heading 2

- item
one
-
not a heading

## Heading 2
Nulla nunc nisi, egestas in ornare vel, posuere ac libero.