I've checked the https://en.wikipedia.org/api/rest_v1/page/mobile-sections-remaining/Chiang_Kai-shek and found it only returns one section, which is not a correct result.

Not sure how the iOS app gets the sections list, and maybe the bug is related to the content with some incomplete formats.

@Mholloway @bearND do you any idea?

The sectioning for this endpoint is base on the Parsoid output for the same page: https://en.wikipedia.org/api/rest_v1/page/html/Chiang_Kai-shek. Looks like all the sections in this document have a data-mw-section-id of -1. I'm not sure why that is for this article. The sections are not transcluded. Is that a Parsoid bug?

In T197479#4303753, @bearND wrote:

The sectioning for this endpoint is base on the Parsoid output for the same page: https://en.wikipedia.org/api/rest_v1/page/html/Chiang_Kai-shek. Looks like all the sections in this document have a data-mw-section-id of -1. I'm not sure why that is for this article. The sections are not transcluded. Is that a Parsoid bug?

The wikitext for this page has a {{stack begin}} at the top of the page which leaves an open <div> tag that isn't closed anywhere. So, it triggers the -2 pseudo-section there. But should check why the rest are -1ed. Later today / tomorrow.

In T197479#4303845, @ssastry wrote:

In T197479#4303753, @bearND wrote:

The sectioning for this endpoint is base on the Parsoid output for the same page: https://en.wikipedia.org/api/rest_v1/page/html/Chiang_Kai-shek. Looks like all the sections in this document have a data-mw-section-id of -1. I'm not sure why that is for this article. The sections are not transcluded. Is that a Parsoid bug?

The wikitext for this page has a {{stack begin}} at the top of the page which leaves an open <div> tag that isn't closed anywhere. So, it triggers the -2 pseudo-section there. But should check why the rest are -1ed. Later today / tomorrow.

Oh yes, of course ... this is because the entire page is marked as template generated because the unclosed <div> tag comes from a template. Right now, pages like these aren't easily editable in VE either.

I don't have any good solution for this here right now. Parsoid's template-wrapping and section-wrapping analysis would have to get smarter about exposing more top-level content that gets pulled into template-wrapped content because of unbalanced HTML tag scenarios like this, but it is tricky to get right in the general case.

See T187384: How should the apps handle Parsoid non-editable and pseudo-sections (section IDs -1 and -2)??

As described by @ssastry above, this is a template problem on this specific page. This comes up from time to time (see T182349: Section parsing bug on :en:Wikimedia Foundation for a previous example). It is not caused by Parsoid emitting -1 and -2 IDs for non-editable sections. Sections are being parsed incorrectly here because of the underlying content problem. It can probably be fixed just by editing a template invocation. We'll take a look at the page and get it fixed.