claims about what it does or does not should not be expressed in plain and simple phrases

It appears that from this you draw the conclusion that any given plain and simple phrase can and should and will be clearly understood to refer to something easier to make such claims about with confidence. I draw a different conclusion: we shouldn't make claims with plain and simple phrases that are liable to be understood in terms of things we don't have a good handle on.

easier to trick oneself with

I am not at all convinced. It is very, very easy to trick oneself into seeing patterns that aren't there, and they will quite often appear to hit you between the eyes. Have a look at some random noise:

These are twelve randomly generated datasets with statistics crudely resembling those of the global warming data from 1960 to 2014. None of them has any sort of hiatus in the underlying process; they're all ramp + white noise. I'd say at least half have "hiatuses" inflicting at least as much interocular trauma as the actual global mean surface temperature graph's "hiatus" does.

If you have MATLAB you can generate similar graphs yourself:

n=55; f=7.5; x=1:n; for i=1:12; y=(1:n)+f*randn(1,n); subplot(3,4,i); plot(x, y, 'r-'); bestj=0; bestm=1; bestk=0; for j=1:(n-14); x1=j:(j+14); y1=y(x1); c=lscov([0*x1'+1 x1'], y1'); if c(2)<bestm; bestj=j; bestm=c(2); bestk=c(1); end; end; if bestm<=0.5; x1=bestj+(0:14); hold on; plot(x1,bestk+bestm*x1,'b-','LineWidth',3); hold off; end; end;

(This only plots the 15-year trend lines when the gradient over those 15 years is <= half the underlying gradient. You will notice that in my plots, every subplot has a trend line plotted. Yours probably will too.)