>this is about 0.4%. I find that a lot,

Why do I find that a lot? I did a very rough estimation, maybe too rough but may be it should be communicated as well:
The "bump" is around 0.003 at 650nm (for this frequency the 2 year slope is actually not pronounced because there irradiance had risen since 2003, but anyways) . On a rough browse the slope increases until around 950nm (see 947.35) at the line 981.15nm it decreases already rather sharply.
So let's assume the shape of the bump is approximately a step on the solar spectrum flank.
So roughly it's half of a quadrilateral with width 300nm and height 0.008 W/m^2 nm.
That is $$0.5*300*0.008 W/m^2= 1.2 W/m^2$$. I assume that this is the power per squaremeter as received on earth, i.e. the distance to the satellite had already been cared for. But thats a lot,
especially as apparently "Between 300 and 800 nm the stratosphere is only weakly absorbing and most of the solar radiation at these wavelengths is transmitted into the troposphere."
On a 1000 m^2 lawn patch you could easily power an electric lawn mower with that.
But as said I have no idea how much of that radiation stays, interacts and at other places of the spectrum you have "negative bumps" etc.