Above the pixel grid’s Nyquist limit. At this viewing distance and pixel pitch the display cannot carry the frequency being asked for, so what reaches the eye is an alias of it — a different frequency, at a different contrast. Move closer, raise the pitch, or ask for less.
The contrast asked for does not fit about this background. A Michelson contrast is symmetric about the mean in luminance and the display cannot go below zero, so the excursion is capped at the largest one both phases survive and the delivered figure is reported. Clipping instead would flatten one phase and measure a threshold for something else.
Every contrast here is Michelson contrast in relative luminance, through the sRGB transfer curve, about the background stated in the rail. The page used to build the grating in 8-bit code values with no transfer curve anywhere: a requested 0.02 arrived as 0.051 and a requested 0.1 as 0.218, with the error largest at exactly the contrasts a detection threshold lives at. Sessions recorded on the old axis are not rescalable after the fact, because the error depends on the mean each excursion was taken about and the old export did not record it.