While investigating T198911: ContentTranslation substitutes some templates in the references a regression on template adaptation was found. After the integration of source page inspect mode feature, the data sent to cxserver for adaptation has an empty data-cx. The template node has data-cx="{}" along with data-mw.
cxserver understand this type of data as nodes that are already adapted because of the presense of data-cx attribute and skips from adaptation. Resulting all inline templates not adapted.
To test: Translate "Tradícia_(zvyky)" from Slovak (sk) to Czech (cs). Translate the first paragraph starting with "Tradícia je antropologický" using an MT service like Google. The reference after "medzigeneračného prenosu." is a template. You can see that it is not adapted even though the target template exists. According to Wikidata ( https://www.wikidata.org/wiki/Q6242419#sitelinks-wikipedia ) the equivalent of sk:Harvbz is cs:Harvnb. But you can see that the target template is still sk:Harvbz