Thanks to Dan Burzo <https://observablehq.com/@danburzo> for a fix here.

I've just realized while plotting a JzCzhz chart (notebook here: https://observablehq.com/d/17dbaff3f61474dc) that `PQ_inv` will result in NaNs when its argument is negative. I wonder if this function can yield more useful values in this situation...

Update: I ended up handling NaN as 0 with `PQ_inv(val) || 0`.

The ZCAM paper refers to the X,Y,Z inputs for Jzazbz as "absolute tristimulus values (in units of 𝑐𝑑/𝑚²)" here: https://opg.optica.org/oe/fulltext.cfm?uri=oe-29-4-6036&id=447640 In order for the Jzazbz luminance curve to line up with the L* curve as in the Jzazbz paper, a display_white_luminance of 203 cd/m² seems best. That also corresponds to the ITU-R BT.2408 "Nominal recommendations for test charts and graphics for PQ and HLG production" of 203 cd/m², 58% PQ, 75% HLG. Notable values: 50 - yields #808080 at 0.5 B/W 80 - sRGB white luminance: https://drafts.csswg.org/css-color-4/#predefined-sRGB 140 - close to XYB luminance curve 160 - Adobe White Point Luminance: https://en.wikipedia.org/wiki/Adobe_RGB_color_space 203 - BT.2408 reference, brighter 70% very close to Lab; https://drafts.csswg.org/css-color-hdr/#valdef-color-rec2100-pq 220 - yields #777777 at 0.5 B/W