Is the HTML content necessary if using JSON LD for semantic data negotiation?

Hi to all!

Specially after looking this video about getting events into knowledge
graph by using Schema / JSON LD, http://goo.gl/zBFftH, i got these thoughts
and questions again:

Google repeatedly mentions its embracing of JSON LD as the
developer-friendly art of semantic data negotiation. I guess the cause
behind it to make the implementation of Schema's data easier / broader. The
questions i nervously ask are:

• whether the HTML content is in general still necessary, if using JSON LD?
• Whether / how algorithm checks the correlation of semantic data provided
in the same web document by JSON LD and the pure content? I assume, that
using in the same web document both of inline Schema's markup and JSON LD
simultanously isn't a good practice at all... right?
• If this correlation isn't checked, how algorithm decides about misuse /
spam usage of content provided by JSON LD snippet?
• Even if this correlation is checked, but meanings provided by content and
JSON LD are more or less different, which "meaning's part" will finally
rank: the one provided by JSON LD, or the one algorithm got from the
analyzing of pure content?

I remember tons of auto-populated websites years ago and see the similar
coming - tons of empty websites containing only JSON LD snippets. Providing
semantic data as inline markup gives added value to the content, but at the
same time it builds a barrier against spam misuse of inline markup, cause
the implementation of it takes place on quasi-manual way. Making
implementation easier (easier means in this context possibility to
*simple *script-based
data querying, negotiation and publication) could backfire with thousands
of auto-populated stub-sites inside of the SERP.

Any (not nervous) thoughts and answers?
Evgeniy