check out the sensor here: 

https://hackaday.io/project/143014-compact-25-spectrometer

This is a spectrometer

The shown height / intensity is normalized in all three modes. The right bar shows calculated intensity (µw per cm^2 per IntTime) or % of T/R coefficient. In the top / middle the used dynamic range of the channel with the highest value in % is show underneath the sum of intensity all channels.

The optical cable is some old audio SPDIF cable and ends in front of a diffuser, covering the sensor (the diffuser must come out, when probing R & T). It used to have a good transmittance over the used range. The LEDs are between 405 nm and 933 nm. I made reference measurements of all of them with a 1000 line Spectrometer (the range is actually from 396.0 nm up to 933.9 nm.

Case V1 (not my design): https://www.thingiverse.com/thing:146...
Case V2 : https://www.thingiverse.com/thing:334...

I haven't checked the intensity, but i have noticed a drift in 2 or 3 channels over time while probing low to very low intensities (610, 680 and 705 nm it was).

I have checked other spectra with good results. To reference the reflected LED intensity is/was quite tricky; i could not repeat the result of a reflective surface. I think the board re-reflects - i thought of a non glossy black silkscreen or so.. maybe by painting some mate black on it. So to get a good reflective coefficient i think i should let the 'user' make his own reference (100%) measurement.

The transmittive coefficient works like a charm. While entering the T-Mode a automatic reference is taken from the background light. but: you shouldn't move the device after that.