A "reference for 24bit ADC" means the Vref fluctuation (thermal, 1/f, other noise, long term stability, hysteresis, etc) is (well) below 1 LSB (0.15uV in your case, imho).
Not quite, noise 0.15uVp-p in sub 10 Hz range, this is why I'm saying "NO single Voltage Reference on a market for 24-bits ADC".
I'm considering one parameter only, it's written in the topic's name, so /aging /temp. stability /long-term / humidity - not to discuss, please.
I start this research during evaluation SD ADC from AD, and explain in another thread that there is No solution to get noise below 0.15 nVp-p
except using simple RC network filtering . You may continue to critisize this , but w/o offering alternative to RC filter, words would not have any sense to me.
TL431B is evaluated for one parameter only, and it's works exceptionally/ perfectly well - my SD ADC running almost for a week now
Comments to image: 3 TL431B in the test, one is connected to adc V.Ref input, two others - to two diff adc inputs.
Filtering is swithed off for all 3 at the moment of capture image, but I also tested with 3 identical (1500 uF + 10k) filters .
Wiggle lines (about 4 uV) similarity on both channels proves that I have a problem not with the filters but with variation on the input current at Ref. Input of the ADC. Strange things, DS says +-100 nA but measurements imply +- 3-5 nA absolute values, with 4 uV / 10k = 0.4 nA bumps