Caveats with the existing TextExtracts service are documented here:
We should also surface these in the API response as warnings where possible.
 When `exsentences` is used in html mode a warning should be sent to the consumer "exsentences is not guaranteed to work when used in HTML mode."
 In HTML mode for all output show a warning "HTML may be malformed and/or unbalanced and may omit inline images. Use at your own risk. Known problems are listed at https://www.mediawiki.org/wiki/Extension:TextExtracts#Caveats"
 The message should be configurable via messages. WikimediaMessages for instance may want to point users to the alternative RESTBase service.
[x] Should we also point them to the RESTbase service in these warning messages?
This is up to the wiki which can customise the message.
[x] Should we add `math` to the elements that should be excluded from text extracts and check that no math tags appear inside https://en.wikipedia.org/w/api.php?format=jsonfm&action=query&prop=extracts&titles=Planck%20constant ? Alternatively we keep them - as consumers can parse them and display them if they are capable of doing so.
No: Let's keep the status quo. Changing how the extract behaves may interfere with how people are currently using the API and it provides little gain for ourselves.
= Developer notes
You can add warnings like so when certain combos are used:
$main->addWarning( new RawMessage( 'existing warning' ), 'existing-warning' );