UV index monitoring (VEML6075, ultra-violet)

I live in Australia so UV is a big deal. I have VEML6075 connected to Tasmota, running fine (well, i2c works and sensor provides data), the sensor looks at the clear sky through a thin glass (and I checked it reduces readings for less than 10%), typical winter readings are:

|VEML6075 UVA intensity|3973 W/m²|
|VEML6075 UVB intensity|5491 W/m²|
|VEML6075 UV Index|10.0|

The actual sensor provides UVA/UVB readings as “counts/μW/cm2” so these thousands of watts look odd but it is likely an interface problem (should be 39.73W/m2, and so on?).

The main problem is the UV index which is calculated by the formula on page6 of [3]. It is winter here now and the actual index is less than 3 but the formula returns 10. So something is really wrong bit I cannot pinpoint what?

Are these indexes the same as the local meteorology service (bom.gov.au) uses? A chart from [3] suggests the max index of 12 but the calculations return 20+ during the summer.
Or actual readings are too high for some reason, then what are the expected numbers? [1] suggests it should be mW/m2, not W.

Has anyone figures this out? Or may be esphome gets it right (does not like it is any different, the same math).

Useful links:
[1] Ultraviolet index - Wikipedia
[2] https://cdn.sparkfun.com/assets/3/c/3/2/f/veml6075.pdf
[3] https://cdn.sparkfun.com/assets/3/9/d/4/1/designingveml6075.pdf
[4] Tasmota/xsns_70_veml6075.ino at adff41f962f7157fe0678f1a5ab9683d0e7535aa · arendst/Tasmota · GitHub

1 x/cm2 ➜ 10,000 x/m2
10,000 μx ➜ 0.01 x
1 μW/cm2 ➜ 0.01 W/m2 or 10 mW/m2

Just for comparison, this is what my Ecowitt weather station reported today in Sydney:

And if you want to try that VEML6075 sensor with ESPHome you can use this custom component code: esphome-customisations/veml6075 at master · exxamalte/esphome-customisations · GitHub