In one of the previous project with cross-site publishing we faced with strange behavior when Sharepoint showed not latest publishing version in Content by search web part from search index. Here is the scenario: we used classic architecture with single authoring web application and multiple publishing sites:
On publishing sites managed metadata was used in order to show different authoring content on different publishing sites (e.g. Country term set and different publishing sites for different countries). In order to simplify content creation on authoring site custom ribbon action was implemented for Pages doclib which copied publishing pages in the same doclib on the authoring site. I.e. page created for all languages was copied, then translators translated this copy to another languages and changed Country managed metadata to appropriate language. Also they removed this language from original page’s metadata:
It worked quite well, i.e. English content was shown on English publishing site, Russian – on Russian. But some time after site worked in production we found one problem: on publishing site if user changed browser language, then Content by search web part showed original English version of the page, which was targeted to both languages. But as I wrote above after translators translated the page they removed their language from the original page (Page 1 on the picture above) and published it with major version.
We used continuous crawl which worked quite well in other scenarios and the problem still reproduced after several hours after it was found, i.e. continuous crawl didn’t fix it. We tried to make incremental and full crawls after that also without result. This behavior means the following:
1. Browse language is used for filtering content in Content by search web parts. This is not new finding, I wrote about how to avoid it here: Configure content by search web parts in Sharepoint 2013 to not use client language.
2. Sharepoint may show content from the search index which is not last published version. I.e. even if greater major versions exist for the page, Content by search web part may still show previous version. This looks like a bug, because only latest content should be used in search index.
In order to fix the problem we changed our content by search web parts to not use client language like described in the article above. After that Content by search web parts started to use specified language, not Accept-Language http header of client’s browser, but the original problem was still not resolved.
Hope that this information will be useful if you will face with the same problem when will work with cross-site publishing in Sharepoint.