Shot noise in optics There are many explanations to be found about shot noise in optics, but the answers I find are incompatible. There are three ways shot noise in optics  is explained. 
(Note that according to Wikipedia, in general, shot noise is a type of noise which can be modeled by a Poisson process.)


*

*It is the noise purely arising form (vacuum) fluctuations of the EM-field. For example, the book of Gerry and Knight states that "In an actual experiment, the signal beam is first blocked in order to obtain the shot-noise level." I guess the number of photons you would detect in this way follows a Poissonian distribution, hence the name `shot noise'.  (For context, see screenshot of relevant section below - courtesy of Google Books)

*It is due to 'the particle nature of light'. Semi-classically, a low intensity laser beam will emit photons following a Poisson distribution. If the beam is incident on a photon detector, this detector will receive a fluctuating number of photons per time bin (according to the Poissonian). Thus the intensity (~number of photons per time bin) will fluctuate. These fluctuations are the `shot noise'.

*A laser beam emits a coherent state $|\alpha \rangle$. The probability to find $n$ photons upon measurement follows the poisson distribution, $P(n)=|\langle n | \alpha \rangle|^2= \frac{\bar n}{n!}e^{-\bar n}$ with $\bar n = |\alpha|^2= \langle \alpha | a^\dagger a | \alpha \rangle $ the average number of photons. Thus there is shot noise in the number of photons. (Here $| n \rangle $ is in the Fock basis but $|\alpha \rangle $ is in the coherent state basis.) 
So what is shot noise? Can you have multiple sources of shot noise, throw them all on one heap and call the combination 'the shot noise'. Then how can you 'measure the shot noise level' as in 1, or 'measure at the shot noise level'? 
Explanation 1 is incompatible with 2 and 3, for both 2 and 3 will cause no photons at all to be counted in the vacuum state. (The vacuum state is the coherent state with $\alpha=0$.)


 A: The explanations that are provided in points 2 and 3 give a good description of the shot noise in light. They are consistent with each other (right?).
The situation in point 1 is probably related to shot noise in detectors. This is electronic shot noise and not related to light. All detectors have a dark current, which exists even without the presence of light. Dark currents produce shot noise in the detector.
A: In general there is noise in light itself, called photon shot noise (different from electronic shot noise), i.e. if a bulb or laser is lit and has a flux of 1M photons per second it really is releasing 1M +- square root of 1M or 1000 photons per second, one std deviation. So that's typically 999,000 or 1001,000 photons actually measured. Light itself is statistical in its emission.  You can't take one atom, stimulate it and not know precisely when the photon will emit.  Yes 2 sources can mix, i.e. 2 bulbs gives 2,000,000 photons with a std error of 2000 but a single build emitting 2,000,000 gives a shot noise of 1414.
A: 
Then how can you 'measure the shot noise level' as in 1

For the balanced homodyne detector (BHD) described in the relevant section of Gerry and Knight, and with the signal port blocked, the sum of the two photodetector (PD) outputs will have noise contributions from the local oscillator (LO) field shot-noise as well as other laser related noise contributions, e.g., relative intensity noise (RIN).
However, the non-shot-noise contributions are common-mode and so the difference of the two PD outputs will (ideally) have only the LO shot-noise.

or 'measure at the shot noise level'?

At the PD difference output, amplitude variations of the signal as well as the shot-noise scale with the LO power.  Increasing the LO power generally improves the signal-to-noise ratio (SNR) when other sources of noise, e.g., electronic noise in the PDs, dominate.
However, if the shot-noise contribution of the LO dominates the other noise sources, increasing the LO power does not improve the SNR; the sensitivity of the apparatus is then shot-noise limited.  This is why, in figure 7.12 of Gerry and Knight, a strong coherent field is specified as being injected into the LO port.
