The cube-root units are totally arbitrary. You can think of them as “steps” of perceived brightness. So, 500 lm looks about 3 “steps” brighter than 100 lm, and 1000 lm looks about 2 “steps” brighter than 500 lm. Here’s an example with nice round numbers from my ramp calculator:
./level_calc.py 1 10 7135 1 1 1000
1: visually 1.00 (1.00 lm): 1.00/255
2: visually 2.00 (8.00 lm): 2.78/255
3: visually 3.00 (27.00 lm): 7.61/255
4: visually 4.00 (64.00 lm): 17.02/255
5: visually 5.00 (125.00 lm): 32.53/255
6: visually 6.00 (216.00 lm): 55.66/255
7: visually 7.00 (343.00 lm): 87.95/255
8: visually 8.00 (512.00 lm): 130.92/255
9: visually 9.00 (729.00 lm): 186.10/255
10: visually 10.00 (1000.00 lm): 255.00/255
PWM1 values: 1,3,8,17,33,56,88,131,186,255
I find that a useful mode spacing should have roughly the same number of steps between each level, and that an interval of ~2 steps looks pretty good. So, on a light which goes up to 1000 lumens, I’d normally do about 6 or 7 steps instead: (this example is something like a “moonlight special” driver)
./level_calc.py 2 6 7135 2 0.2 140 7135 2 2 860
1: visually 0.58 (0.20 lm): 2.00/255, 0.00/255
2: visually 2.47 (15.03 lm): 28.84/255, 0.00/255
3: visually 4.35 (82.36 lm): 150.69/255, 0.00/255
4: visually 6.23 (242.26 lm): 255.00/255, 31.56/255
5: visually 8.12 (534.79 lm): 255.00/255, 117.82/255
6: visually 10.00 (1000.00 lm): 255.00/255, 255.00/255
PWM1 values: 2,29,151,255,255,255
PWM2 values: 0,0,0,32,118,255
OTOH, for a light which does smooth ramping, like the BLF Q8, I find that it looks better to have only ~0.1 or ~0.2 steps between levels… and a lot of levels. IIRC, it has a ramp 128 levels long.
./level_calc.py 2 128 7135 2 0.2 140 FET 2 10 4000
1: visually 0.58 (0.20 lm): 2.00/255, 0.00/255
2: visually 0.71 (0.35 lm): 2.27/255, 0.00/255
3: visually 0.83 (0.56 lm): 2.66/255, 0.00/255
...
126: visually 15.63 (3820.73 lm): 255.00/255, 243.22/255
127: visually 15.75 (3909.68 lm): 255.00/255, 249.06/255
128: visually 15.87 (4000.00 lm): 0.00/255, 255.00/255
It may seem weird that low levels have only 0.15 lm between while high levels have 90 lm between, but it looks linear in practice.